Emerging Trends in Computer Science and Its Application
Uploaded by
CRC PRESSEmerging Trends in Computer Science and Its Application
Uploaded by
CRC PRESSInnovations in Computational Intelligence:
Trends and Technologies
Dr. Anurag Tiwari
Anurag Tiwari is an Associate Professor and Head of the Department of Computer Science
and Engineering at BBD National Institute of Technology and Management, Lucknow. A Sen-
ior IEEE Member, he has over 11 years of academic experience and has led the department to
NBA accreditation in 2023. He holds a Ph.D. in Computer Science, multiple M.Tech degrees,
and has authored several research publications. With multiple patents and consultancy pro-
jects to his credit, his research interests include machine learning and optimization techniques.
Dr. Manuj Darbari
Manuj Darbari, Senior Member IEEE (USA) and Chartered Engineer (India), is a visionary
leader in Computer Science and Engineering, specializing in AI, Cloud Computing, and IoT.
With a track record of pioneering research, patents, and numerous PhD supervisions, he is
recognized for driving innovation and excellence. Currently a Professor at BBDITM Lucknow,
Dr. Darbari has made significant contributions to cutting-edge technologies and has served as
a convener and reviewer for leading global conferences. His editorial expertise ensures that
the conference proceedings reflect the highest standards and capture the forefront of emerging
research trends.
Innovations in Computational Intelligence:
Trends and Technologies
Proceedings of the International Conference on Advances in
Emerging Trends in Computer Applications (ICAETC-2023)
December 21–22, 2023, Lucknow, India
Edited by
Anurag Tiwari
Manuj Darbari
Boca Raton London New York
CRC Press is an imprint of the
Taylor & Francis Group, an informa business
First edition published 2024
by CRC Press
4 Park Square, Milton Park, Abingdon, Oxon, OX14 4RN
and by CRC Press
2385 NW Executive Center Drive, Suite 320, Boca Raton FL 33431
© 2024 selection and editorial matter, Anurag Tiwari and Manuj Darbari; individual chapters, the
contributors
CRC Press is an imprint of Informa UK Limited
The right of Anurag Tiwari and Manuj Darbari to be identified as the authors of the editorial
material, and of the authors for their individual chapters, has been asserted in accordance with
sections 77 and 78 of the Copyright, Designs and Patents Act 1988.
All rights reserved. No part of this book may be reprinted or reproduced or utilised in any form
or by any electronic, mechanical, or other means, now known or hereafter invented, including
photocopying and recording, or in any information storage or retrieval system, without permission in
writing from the publishers.
For permission to photocopy or use material electronically from this work, access
www.copyright.com or contact the Copyright Clearance Center, Inc. (CCC), 222 Rosewood Drive,
Danvers, MA 01923, 978-750-8400. For works that are not available on CCC please contact
[email protected]
Trademark notice: Product or corporate names may be trademarks or registered trademarks, and are
used only for identification and explanation without intent to infringe.
British Library Cataloguing-in-Publication Data
A catalogue record for this book is available from the British Library
ISBN: 9781032999012 (pbk)
ISBN: 9781003606635 (ebk)
DOI: 10.1201/9781003606635
Typeset in Sabon LT Std
by HBK Digital
Contents
List of figures xiv
List of tables xxii
Chapter 1 A systematic review on recent controllers for an industrial wastewater
treatment plant 1
K. V. Reshma, A. Vimala Juliet, and G. Glan Devadhas
Chapter 2 Green SaaS adoption in cloud computing: An investigation into environment
sustainability13
Bhanu Priya, Bhupal Arya, K. M. Pandey, and Prashant Verma
Chapter 3 Anomaly detection in wireless sensor network 24
Mayank Shukla, Sneha Yadav, Abhay Pratap Singh, Fizza Rizvi, and Surya Vikram Singh
Chapter 4 Revolutionizing object recognition beyond CNN and YOLO with deep learning
breakthrough31
G. S. S. S. S. V. Krishna Mohan, Mahammad Firose Shaik, G. Usandra Babu,
R. G. V Prasanna, Pinagadi Venkateswara Rao, and Inakoti Ramesh Raja
Chapter 5 Investigating efficacy of emerging ICTs for the Indian construction sector 40
Sanjay Pandey and G. T. Thampi
Chapter 6 AgriTech: Empowering agriculture through integrated technology 50
Pankaj Kunekar, Anushka Popalghat, Onkar Borude, Vishal Gavali, Dnyanesh Gholap,
and Sarvesh Hadole
Chapter 7 Online and offline learning in universities: A student’s dilemma 59
Ruchi Agarwaland Rahul Pradhan
Chapter 8 Strengthening UAV network security: Advanced authentication-based intrusion
detection systems 65
Altaf C, Shanila Mahreen,C Atheeq, and Shaista Khanam
Chapter 9 Comparative analysis of wireless routing protocols for security threat—A review 74
Bhupal Arya, Amrita Kumari, and Jogender Kumar
Chapter 10 Cascadability analysis of free space optical wireless data centers 81
Anand Kumar Dixit, Meenakshi Srivastava, and Rajiv Srivastava
vi Contents
Chapter 11 Wireless network with machine learning and deep learning techniques 88
Bhupal Arya, Amrita Kumari, and Jogendra Kumar
Chapter 12 Assessing technology innovation of cellular health 100
M. Vasumathi Devi, Y. Muvvana, Y. Sai Likhitha, M. Gowri Bhavani, and M. Vishnu Priya
Chapter 13 Reform dentistry: A web-based platform for revolutionizing dentistry 107
Vineet Agrawal, Nikita Verma, Aman Yadav, and Khyati Maurya
Chapter 14 Approach to recognize fake news 114
Arya Srivastava, Palak Chaubey, Manasvi Sonkar, Rudrendra Bahadur Sing, and Vinayak
Chapter 15 Cybersecurity measures for safeguarding medical image data 121
Zeenath, K. Durga Devi, andJohn W. Carey M.
Chapter 16 Detection of phishing attacks using machine learning 129
Siddharth, Rajeev Srivastava, Harsh Raj, Shourya Dwivedi, Shourya Dwivedi,
Rohit Singh, and Nitish Chaurasiya
Chapter 17 A comprehensive exploration of aircraft detection in synthetic and satellite images 135
M. Madhu Bala, G. Ramesh Chandra, G. Pavan Teja, M. Sai Sudha, Sk. Sabeeha Kouser,
and M. Ajay Kumar
Chapter 18 Mitigating man-in-the-middle attack in UAV network using authentication
mechanism based on chaotic maps 144
C. Atheeq, Layak Ali,C. Altaf , and Aleem Mohammed
Chapter 19 A comprehensive multi-modal sentiment analysis approach for social media content
integration153
Sheela S., Suresh Balakrishnan T., Uttham Sing K., Sujitha E., Soundarrajan R., and Veena M.
Chapter 20 A proposed deep learning framework for internet-of-medical things 159
Faiyaz ahamad and Syed Hauider Abbas
Chapter 21 Analysis of novel routing protocol to minimize delay in VANET: A comprehensive
simulation for VANET 165
Kamlesh Kumar and Bobby Sharma
Chapter 22 Securing the secrets of 5G: Mitigating eavesdropping threats and enhancing network
integrity175
Mamidisetti Sai
Chapter 23 A comprehensive crop prediction model leveraging optimization algorithms and
machine learning classifiers 183
Neha Chauhan, Dinesh Sharma, and Deepika Dhaneja
Chapter 24 Malaria detection using convolutional neural networks: A deep learning approach 192
Navin Kumar Agrawal
Contents vii
Chapter 25 An optimal emergency vehicle path finding mechanism using routing algorithms 197
Hina Naaz, Roshan Jahan, and Faiyaz Ahamad
Chapter 26 Smart city solutions: Enhancing infrastructure with LoRa multi-hop networks 203
Faraz Ahmad, Saleha Mariyam, and Faiyaz Ahamad
Chapter 27 Enhancing online customer engagement through strategic optimization:
Customized page ranking and analytical insights via web mining techniques 210
Ambareen Jameel, Mohd Usman Khan, and Faiyaz Ahamad
Chapter 28 Towards unbreakable cloud security: Symmetric key cryptography and
blockchain synergy218
Khushabu Agrawal, Gunjan Verma, Priya, Divya Sharma, and Nidhi Pruthi
Chapter 29 Neuralink’s approach to spinal cord injuries and vision impairment
through pioneering brain–machine interfaces 226
Manoj M., Banupriya N., Gayathri V., Girinath D., and Shibi Kannan A.
Chapter 30 Honeypots and honeynets: investigating attack vectors 233
Pawandeep Kaur and Harpreet Kaur
Chapter 31 A review on deep learning frameworks for pose estimation during exercise for
rehabilitation240
Banupriya N., Haris Dominic Savier, Jeeva V., and Uma R.
Chapter 32 SwiftScan SpeedPro: OpenCV-powered vehicle speed monitoring 246
Vishal Kumar Sinha, Ayush Bharti, and Selvin Paul Peter J.
Chapter 33 Hawkeye-intelligent surveillance system 255
Raj Kalantri, Umang Thakur, and Selvin Paul Peter J.
Chapter 34 Sentimental analysis for amazon product review 263
Sumathi S., Mohanapriya A., Mohammed Aashif M J., Vidhur S., and Vijayalakshmi K.
Chapter 35 Crop recommender system using machine learning approach 271
Sujatha V., Lavanya N., Karunasri V., SaiSindhu G., and Madhavi R.
Chapter 36 Cloud malware detection using heuristic techniques 277
Lakshmi Triveni D., Alapati Naresh, Sree Harshitha A., Koteswari. K., and
Bhavana Lakshmi P.
Chapter 37 Guarding your home: A seamless Iot-driven security system with high security 285
Murala Divyasree, Alapati Naresh, Tammisetty Jyothi, Perumalla Sahithi, and Vinukonda
Ankitha
Chapter 38 The challenges of enforcing cybercrime laws in the age of E-governance:
A literature review 292
Manuj Darbari, Naseem Ahmed, and Abhishek Kumar Singh
viii Contents
Chapter 39 A exhaustive review on advancements and challenges in low power wireless sensor
networks297
Manuj Darbari, Naresh Chandrab, Diwakar Yagyasenc,
Anurag Tiwarid, and Sandeep Kumar Mishrae
Chapter 40 Automated personality evaluation of children based on emotional speech 302
Alapati Naresh, Basireddy Yasaswi, Golla Sri Lakshmi, Konakandla Meghana,
and Allamudi Yamini
Chapter 41 Redefining human activity recognition with LSTM and GRU networks 308
Vasumathi Devi M., Sai Sreeja S., Anusha P., Mounika Bindu B., and Sirisha M.
Chapter 42 Cybersaferoutes: Protecting IoT networks from hybrid attack with security
measures315
Ganga Bhavani P, Lakshmi Tirapathamma S, Reshma SK, Swetha B, and Himaja N
Chapter 43 Harmful content classification in social media using gated recurrent units and
bidirectional encoder representations from transformer 321
Sujatha V., Tejaswi Y., Pravalika V., Pavani P., and Ch Sravani
Chapter 44 Detection of missprounciation using deep learning 328
Naga Vardhani B, Naga Nandini K, Nandhakeerthi M, and Rupanjali M
Chapter 45 IOT-enabled cloud solutions for reliable health monitoring 335
Lakshmi Sravani G., Lalitha Chandrika N., Lavanya S., Vijaya Lakshmi S., and
Sri Lakshmi Prasanna S.
Chapter 46 Detection of melanoma using DenseNet-based adaptive weighted loss function 341
Rishi Agrawal
Chapter 47 Advanced reconfigured solar arrays to attenuate shading loss under partial shading
conditions: experimental validation 346
Vijay Laxmi Mishra, Yogesh K. Chauhan, and Kripa S. Verma
Chapter 48 Understanding the relationship of quantum mechanics and machine learning 352
Sushamana Sharma
Chapter 49 Employability prediction: A machine learning model for computer
science students 359
Abhishek Bhattacherjee and Vinay Kukreja
Chapter 50 A comprehensive review of innovations in stray cattle detection 365
Abdullah Nizami, Rudrendra Bahadur Singh, Haris Javed, Mohd Anas Nadeem, and
Alok Kumar
Chapter 51 Enhancing emotional insight: NLP powered sentiment analysis 373
Jyoti Gaur, Stuti Singh, and Shelja Sharma
Contents ix
Chapter 52 Decision-making using argumentation mining framework on English text 379
Rudrendra Bahadur Singh, Shobhit Sinha, Ankita Singh, Alok Kumar Thakur, Akhil Chaurasiya,
Nishtha Maurya, and Arpita Yadav
Chapter 53 Potato leaf disease classification and prediction using CNN 386
UdayPratap Singh, Sandeep Kumar Mishra, ShubhamVishwakarma, Suraj Kumar,
and Vinayak Pandey
Chapter 54 Exploring the impact of integrated technological solutions on student engagement
administrative efficiency 393
Anurag Tiwari, Ayush Maurya, Arushi Baranwal, Amritesh Gupta, and Anurag Pathak
Chapter 55 Early stage lung cancer detection using RESNET 399
Anjali Srivastava, Vinayak, Aarohi Rai, and Ainy Khan
Chapter 56 Cryptographic image concealment with neural networks 406
Chinnala Balakrishna,Shaik Saidhbi, Balajee Maram, Mallikharjuna Rao K., B. Santhosh
Kumar, and Sasibhushana Rao Pappu
Chapter 57 An efficient CNN-based approach for automated animal intrusion detection 412
Ramamani Tripathy, S V Achuta Rao, Maheswari. P, Mallikharjuna Rao K,
B. Santhosh Kumar, and Balajee Maram
Chapter 58 Next-generation smart cities: Integrating blockchain, IoT, deep learning, and
Dingo optimization 419
K. Sankar, Ramamani Tripathy, S. Nandhini, Aleem Ali, B. Santhosh Kumar,
and Balajee Maram
Chapter 59 Intelligent diagnosis of diabetic retinopathy: Leveraging machine
and deep learning 425
Balajee Maram, S. Arun Joe Babulo, B. Manivannan, B. Santhosh Kumar, T. Daniya, and
Sasibhushana Rao Pappu
Chapter 60 Digital dwelling hub: Revolutionizing real estate with AR and smart home
integration431
Tanya Omar, Rajeev Srivastava, Tanya Singh, Vaibhav Srivastava, Suraj Chhetri, and
Shaurya Kumar
Chapter 61 Mobilenet-v3: A comprehensive survey of object detection algorithms using CNN 438
Rasheeq Zehra, Omkar Sharma, Vinayak, and Rudrendra Bahadur Singh
Chapter 62 Rainstorm prediction system 445
Muskan Gupta, Ankit Khare, Kirti Pandey, Pritismita Thakur, and Muskan Gupta
Chapter 63 Advancements in assistive technologies for web accessibility: A systematic review 451
Ayushi, Padmini Mishra, Anusha Ansari, and Drishti Maurya
Chapter 64 Predicting bitcoin prices: A machine learning approach for accurate forecasting 457
Rishabh Jain, Shekhar Srivastava, and Prakhar Shukla
x Contents
Chapter 65 A comprehensive review on multi disease prediction web app: QuadraDiag 464
Khushi Saxenaa, Keerti Saxenab, Vishesh Guptac, Swatid, and Kamlesh Kumare
Chapter 66 A comprehensive review on tomato plant disease detection using
convolutional neural network 473
Rudra Pratap Singh, Sandeep Kumar Mishra, Samriddhi Chaurasia, Sameer Thakur,
and Sachin Yadav
Chapter 67 Building a decentralized non-fungible token marketplace: Leveraging blockchain
technology for secure digital asset exchange 480
Vaibhav Bansal, Shivam Arora, Sonia Deshmukh, and Ankit Saini
Chapter 68 CollegeGo: leveraging alumni data for improved college placements 487
Avneesh Agrahari, Subha Mishra, Rudrendra Bahadur Singh, Kumar Ronit,
and Pahun Ratna
Chapter 69 An in-depth review of AI/ML techniques in emotion-preserving multilingual video
translation for practical cross-cultural communication 492
Diwakar Yagyasen, Akshit Kumar Tiwari, Abhishek Kevin Gomes, Aditya Yadav, and Harsh
Anand Gupta
Chapter 70 Data analysis for revenue optimization in the hospitality sector 499
Pankaj Kunekar, Parul Patle, and Anagha Gajaralwar
Chapter 71 Technological solutions for E-waste management and disposal 506
Aditya Pratap Singh and Akshat Chaturvedi
Chapter 72 Comparative study of various transformation techniques in image fusion 513
Kunal Kishor Jha, Sumit Kumar Jha, Hari Om Shanker Mishra, Amit Dhawan,
Manish Tiwari
Chapter 73 Blockchain transaction processing: challenges and resolutions 521
Kamal Kant, Sarvesh Pandey, and Udai Shanker
Chapter 74 Fake social media profile detection using machine learning 527
Anurag Shukla, Shreya Chaurasia, Tanushri Asthana, Tej Narayan Prajapati, and
Vivek Kushwaha
Chapter 75 Prompted LLMs as chatbot modules for long open-domain conversation 534
Yash Kumar Singh, Deepak Pant, Syed Mohd Askari, and Shivam Kumar Pandey
Chapter 76 Smart locking system using machine learning technique 542
Sadhana Singh, Jhanak Verma, Aditya Gupta, and Priyanka Sharma
Chapter 77 Mind Matters: Psychological analysis using CBT Chatbot 547
Preeti Maurya, Gyanendra Kumar, Aditya Singh Yadav, Kumari Anushka Gupta,
and Adarsh Tiwari
Contents xi
Chapter 78 Optimizing patient outcomes through integrated health analysis and management
strategies 554
Anurag Shukla, Sambhav Pathak, Rahul, Rishikant Yadav, and Raj Gupta
Chapter 79 Data leakage detection 562
Neha Singh, Sandeep Kumar Mishra, Rahul Kumar, and Sandeep Vishwakarma
Chapter 80. Vision-based lip-reading system using deep learning 569
Abhishek Pratap SinghAmit Kumar Sachan, Akram Ansari, and Abhishek Dubey
Chapter 81 Reform dentistry: A web-based platform for revolutionizing dentistry 575
Vineet Agrawal, Nikita Verma, Aman Yadav, and Khyati Maurya
Chapter 82 AI genius: Integrated intelligence platform 582
Azhar Ahmad, Aneesh Sahu, Amir Ali, Anjali Maurya, and Saurabh kumar Jain
Chapter 83 Agriculture: Soil analysis for suitable crop and fertilizer prediction 589
Shreya Singh, Saurabh Kumar Jain, and Stuti Rastogi
Chapter 84 Comprehensive examination of system surveillance method employing keystrokes
logging technology 597
Bajrangi Vishkarma, Kamlesh Kumar, Aman Verma, Atul Singh, and Abhishek Singh
Chapter 85 Health prediction application: A comprehensive literature review 604
Tushar Agrawal, Prince Kumar Singh, Aditya Kumar, and Mohit Chowdhary
Chapter 86 E-portal for case management and hearing 610
Anushka Shukla, Anurag Shukla, Ashish Kumar Yadav, and Anushka Srivastava
Chapter 87 Enhancing electoral processes: The role of “Smart Voting Systems” 616
Madhulika Sharma, Anuj Singh, and Gulshan Varshney
Chapter 88 Advancements in rainfall prediction systems: A comprehensive review of
techniques and models 622
Adarsh Vishwakarma, Ankit Khare, Amit Verma, and Aditya Pandey
Chapter 89 FindServ 628
Mayank Srivastava, Pratibha Dixit, Adarsh Verma, Sadaf Ali
Chapter 90 IoT devices security against common threats 635
Mayank Srivastava and Prabudh
Chapter 91 Predict-a-disease: Survey 642
Mayank Srivastava, Aditya Singh Chauhan, and Abhinav Bhardwaj
Chapter 92 Securing MANETs against black hole attacks: A novel approach using secure
knowledge algorithm 648
Mohammed Abdul Bari, Arshad Ahmad Khan Mohammad, Arif Mohammad, and Abdul
xii Contents
Chapter 93 AI-powered navigation of the digital frontier: Analyzing modern trends in
digital marketing and their influence on consumer purchasing patterns with
special reference to pharma industry 655
Abhay Singh Rajawat, Ved Prakash, and Shailendra Singh Chauhan
Chapter 94 Fake video detection 665
Mahendra Singh, Saurabh Jain, Laraib Khan, and Iaswarchandra
Chapter 95 Olympic odyssey: Tracing 120 years of athletes sports, and global influence 673
Shreya Verma, Shreya Mishra, Prakriti Singh, Sneha Seth, and Sanjay Kumar Gupta
Chapter 96 Crop prediction system using machine learning 680
Neha Chauhan, Devansh Tomar, Gaurav Singh, Gaurav Mishra, and Anand Mishra
Chapter 97 A mobile application to assist the women in gestation period 686
V. lavanya, Madhu Hasitha Manne, Gnana Sri Manasa Veeramachaneni, Aditya Ram Manam,
and Sajid khan Mohammed
Chapter 98 I-tourism planner 692
Aayush Kumar, Himanshu Ojha, Pratham Singla, and Amit Kumar Jaiswal
Chapter 99 Blockchain based E-vault system for legal records 697
Shekhar Srivastava, Amritanshu, Maitri Jain, and Rahul Maurya
Chapter 100 Anomaly detection in wireless sensor network 704
Mayank Shukla, Sneha Yadav, Abhay Pratap Singh, Fizza Rizvi, and Surya Vikram Singh
Chapter 101 Enhancing fault tolerance in distributed systems through machine learning
techniques711
Mohd Haroon, Manish Madhav Tripathi, Jameel Ahmad, Shish Ahmad, and Mohd Husain
Chapter 102 Predicting breast cancer risk: A machine learning approach 717
Navin kumar Agrawal
Chapter 103 A multi-objective optimization algorithm for fog computing with trust and energy
awareness 723
Ruchi Agarwal, Saurabh Singhal, and Ashish Sharma
Chapter 104 Pediatric brain tumor detection in MRI: A machine learning perspective 729
Sandeep Kumar Mishra and Sheeba Praveen
Chapter 105 Enhancing trust in the social internet of things (SIoT): A comprehensive review 736
Gyanendra Kumar and Anuradha Misra
Contents xiii
Chapter 106 Fostering plant health: Utilizing YOLOv7 framework to identify leaf diseases 743
Rajya Lakshmi R, Sasikala V, Divya Likhitha D, Rishita Reddy Ch, and VenkataSai Vidyasri
Abhishainy A
Chapter 107 A review on-malignancy detection in lung nodules using CNN 751
Krish Yadav, Vinayak, Pratibha Yadav, Lakshmi Pandey, Himanshu Sharma, and Smita Yadav
Chapter 108 Next-generation parking system using IoT 756
K. Geetha, Gude Ganesh Satya Kumar, Raghu Ram C., and A. G. Sreedevi
Chapter 109 A review on smart health diagnosis 760
Kishan Kumar Rauniyar, Vijay Yadav, Padmini Mishra, Sanskar Singh, and
Ashish Kumar Yadav
Chapter 110 Android app threat analysis techniques: a review 765
Ujjwal Shukla, Gyanendra Kumar, Syed Haidar Abbas, and Vaibhav Paliwal
Chapter 111 AI Powered Legal Documentation Assistant 770
Yogesh Shekhawat, Utkarsh Tiwari, Syed Hasan Mehdi, and Himanshu Vaishy
Chapter 112 Ideas to boost fitness activities and assist in keeping fit 777
Neha Chauhan, Vasu Gupta, Harshita Bajpai, and Kim Gupta
Chapter 113 Using existing CCTV network for crowd management, crime prevention and work
monitoring: A novel approach for object detection 783
Manvi Agrawal
List of figures
Figure 1.1 Automatic control flow of aeration tank DO content by the improved IPI control
algorithm6
Figure 1.2 shows the process scheme of the WWTP to which MPC is applied 7
Figure 1.3 Workflow of the proposed deep learning-driven forecasting procedure 9
Figure 1.4 Benchmark Simulation Model 1(BSM 1) 10
Figure 2.1 Green IT components 15
Figure 2.2 Greener cloud placement in cloud computing technology 15
Figure 2.3 Green cloud service model of software as a service 19
Figure 4.1 YOLO architecture 34
Figure 4.2 Feature extracting process by RCNN methodology 35
Figure 4.3 Block diagram for text to speech conversion 36
Figure 4.4 Detection of object by camera module 37
Figure 4.5 List of detected objects 38
Figure 5.1 Types of construction companies contributed to the study 43
Figure 5.2 Cadre-based classification of respondents 43
Figure 5.3 Harnessing ICTs for construction industry tasks 45
Figure 5.4 Perceived benefits of using ICTs for construction companies 46
Figure 5.5 Hindering factors for ICT adaptation in the Indian construction sector 47
Figure 5.6 Overall performance of construction companies after ICT adaptation 48
Figure 6.1 Soil testing lab locations 52
Figure 6.2 Heatmap for crop recommendation 52
Figure 6.3 Crop recommendation 52
Figure 6.4 Plant disease prediction 52
Figure 6.5 Blockchain-based marketplace 52
Figure 6.6 Crop price prediction 53
Figure 6.7 Real-time weather monitoring 53
Figure 6.8 Gamified plant growth 54
Figure 6.9 Confusion matrix 54
Figure 6.10 Accuracy chart 54
Figure 6.11 Crop recommendation flow diagram 54
Figure 6.12 Visualizing plant growth using image processing 55
Figure 7.1 Subject distribution according to the ages 60
Figure 7.2 Subject distribution according to the gender 60
Figure 7.3 Subject distribution according to education qualification 60
Figure 7.4 Feedback– Distribution of Agreement and disagreement on above issues 62
Figure 8.1 Comparison of proposed method with existing with respect to PDR 71
Figure 8.2 Comparison of proposed method with existing with respect to delay 72
Figure 8.3 Comparison of proposed method with existing with respect to overhead 72
Figure 9.1 76
Figure 10.1 Communication between the ToRs using LoS and NLoS links 82
Figure 10.2 LoS communication between the cascaded racks (BUS topology) 82
Figure 10.3 LoS communication between the racks (16 Racks topology) 83
Figure 10.4 LoS communication between the racks is 3 meters (a) FOV=70° (b) FOV=10° 84
List of figures xv
Figure 10.5 LoS communication between the racks is 3√2 meters (a) FOV=70° (b) FOV=10° 84
Figure 10.6 LoS communication between the racks is 6 meters (a) FOV=70° (b) FOV=10° 85
Figure 10.7 LoS communication between the racks is 6√2 meters (a) FOV=70° (b) FOV=10° 85
Figure 10.8 LoS communication between the racks is 9 meters (a) FOV=70° (b) FOV=10° 85
Figure 10.9 LoS communication between the racks is 9√2 meters (a) FOV=70° (b) FOV=10° 86
Figure 10.10 Probability of Error for various distances between the racks 86
Figure 11.1 Deep learning in wireless network 90
Figure 11.2 ML using ANN and DNN 91
Figure 11.3 91
Figure 11.4 93
Figure 11.5 94
Figure 11.6 96
Figure 11.7 97
Figure 12.1 Conceptual framework. 102
Figure 12.2 Methodology schema 104
Figure 12.3 Types of admission. 105
Figure 12.4 City Code Patient Count 105
Figure 13.1 Block diagram 110
Figure 15.1 Comparison of proposed method with existing with respect to PDR 126
Figure 15.2 Comparison of proposed method with existing with respect to Delay 126
Figure 15.3 Comparison of proposed method with existing with respect to Overhead 126
Figure 15.4 Comparison of proposed method with existing with respect to Throughput 127
Figure 17.1 A Sample DOTA image of airplane category with OBB representation
(a) v1.0 (b) v1.5 and (c) v2.0waveform of BLDC motor 136
Figure 17.2 A Sample synthetic Image 137
Figure 17.3 Aircraft detection framework using YOLO 137
Figure 17.4 Architecture of Detectron2 137
Figure 17.5 FR-CNN architecture in aircraft detection 138
Figure 18.1 Comparison of proposed method with existing with respect to PDR 150
Figure 18.2 Comparison of proposed method with existing with respect to Delay 150
Figure 18.3 Comparison of proposed method with existing with respect to Overhead 151
Figure 19.1 Text data based sentiment analysis 155
Figure 19.2 Image-based sentiment analysis 155
Figure 19.3 Live video-based sentiment analysis 156
Figure 19.4 Audio-based sentiment analysis 156
Figure 19.5 Emotion detected using text 157
Figure 19.6 Emotion detected using Image 157
Figure 19.7 Emotion detected using Image 157
Figure 19.8 Emotion detected using Audio 157
Figure 20.1 The proposed model diagram 161
Figure 20.2 Comparison with existing frameworks 163
Figure 21.1 A Differential Value of VANET nodes 171
Figure 21.2 Decision of node movements 171
Figure 21.3 Simple AODV setup 172
Figure 21.4 Basic AODV 25 nodes 172
Figure 21.5 Starting simulation of basic AODV 25 nodes 172
Figure 21.6 AODV 20 nodes 172
Figure 21.7 Starting simulation of AODV 20 nodes 173
Figure 21.8 AODV 25 nodes 173
Figure 21.9 Starting simulation of AODV 25 nodes 173
Figure 21.10 AODV 30 nodes 173
xvi List of figures
Figure 21.11 Starting simulation of AODV 30 nodes 173
Figure 21.12 Bar chart during delay occurrence on different nodes 173
Figure 22.1 Eavesdropping attack 176
Figure 22.2 MOSPR working process 178
Figure 24.1 Classification of labelled image 194
Figure 24.2 Loss function for validation data 195
Figure 24.3 Accuracy on validation data 195
Figure 24.4 AUC-ROC curve 195
Figure 25.1 Grid map graph 199
Figure 25.2 Performance comparison graph 201
Figure 26.1 Layers of proposed framework for smart city 204
Figure 26.2 Proposed framework for integration of LoRa technology for cross platform
interoperability207
Figure 26.3 Significant improvement in data transmission efficiency, after integration of LoRa
technology in urban infrastructure 208
Figure 26.4 Gateway statistics facilitating inter sensor communication 208
Figure 26.5 Packet transmission statistics 208
Figure 27.1 Web data extraction techniques 211
Figure 27.2 https://www.shopify.com/in/aboutpage 213
Figure 27.3 Combined data 214
Figure 27.4 List of top countries for marketing Web Log Hit 214
Figure 27.5 List of top countries for Web Log Hit 216
Figure 27.6 Number of HITS 216
Figure 28.1 Classification of cryptography algorithm in cloud computing 220
Figure 28.2 The process of symmetric-based key algorithm 220
Figure 28.3 Represents the overall process of Proposed Cryptography algorithm using
blockchain methodology in cloud computing 221
Figure 29.1 Representation of BMI 227
Figure 29.2 Elon Musk’s Neuralink 228
Figure 29.3 BCI chip installation 229
Figure 29.4 Visual prosthesis representation 229
Figure 29.5 Prefrontal Cortex 230
Figure 29.6 Representation of Mental Synthesis Theory 231
Figure 30.1 Honeypot architecture 235
Figure 30.2 Classification of honeypot 235
Figure 31.1 The general GCN diagram from the input data to the desired output 241
Figure 31.2 Branches of classification in body 244
Figure 31.3 The dataset images for the pose estimation 244
Figure 31.4 The classification and the correction branch diagram for the image model 244
Figure 32.1 System flow 248
Figure 32.2 Framework architecture 250
Figure 32.3 Project model-object tracking 251
Figure 32.4 Speed calculation 251
Figure 32.5 Saves vehicle data 252
Figure 32.6 Use case diagram 252
Figure 32.7 Timing table 252
Figure 32.8 Stage by stage measurements 253
Figure 32.9 Stage-by-stage ROC curve 253
Figure 33.1 System flow 256
Figure 33.2 System architecture 258
Figure 33.3 Framework architecture 258
List of figures xvii
Figure 33.4 ResNet architecture 259
Figure 33.5 LSTM working 260
Figure 33.6 Beem search algorithm 260
Figure 33.7 Architecture 260
Figure 33.8 Sequence diagram 261
Figure 33.9 Class diagram 261
Figure 33.10 Upload video 261
Figure 33.11 Live captions generated 261
Figure 33.12 Comparison with other captioning models 261
Figure 34.1 General architecture 264
Figure 34.2 Implementation of dataset 266
Figure 34.3 Visualization of dataset 266
Figure 34.4 Visualization using word cloud 267
Figure 34.5 Tabulation of model 269
Figure 34.6 Visualization for F1-score 269
Figure 35.1 Architecture 272
Figure 35.2 Methodology 274
Figure 35.3 Accuracy comparison 275
Figure 35.4 Dataset distribution (area) 275
Figure 35.5 Dataset distribution (Production) 276
Figure 35.6 Heat map of the dataset 276
Figure 36.1 Generation of opcode sequences 280
Figure 36.2 Opcode sequence instructions 280
Figure 36.4 Bar graph representing opcode versus normalized frequency in benign and
malware281
Figure 36.3 Benign and malware images 281
Figure 36.5 Accuracy of CNN vs KNN, NB 283
Figure 37.1 Flowchart 289
Figure 37.2 Motion detection 290
Figure 37.3 Authentication 290
Figure 37.4 Authentication request message 290
Figure 37.5 Captcha generation 290
Figure 39.1 WSN deployed in forest area 299
Figure 40.1 Workflow of the model 304
Figure 40.2 CNN architecture 305
Figure 40.3 Output 306
Figure 41.1 LSTM and GRU model architecture 311
Figure 41.2 Applications of 52 312
Figure 41.3 Accuracy 312
Figure 41.4 Precision 312
Figure 41.5 Recall 313
Figure 41.6 F1-Score 313
Figure 42.1 Architecture diagram 318
Figure 42.2 Throughput 319
Figure 42.3 Packet delivery ratio 319
Figure 42.4 Energy delay 319
Figure 42.5 Control overhead 319
Figure 42.6 Loss 319
Figure 43.1 Hybrid GRU-BERT sentinel 324
Figure 43.2 Accuracy 325
Figure 43.3 Precision 325
xviii List of figures
Figure 43.4 Recall 325
Figure 43.5 F1-Score 326
Figure 43.6 Loss 326
Figure 44.1 CNN and BiLSTM architecture model 331
Figure 44.2 Accuracy 332
Figure 44.3 Loss 332
Figure 44.4 Precision 333
Figure 44.5 Recall 333
Figure 44.6 F1-Score 333
Figure 21.1 Architecture of reliable health monitoring 337
Figure 21.2 AES structure 338
Figure 21.3 Accuracy data 338
Figure 21.4 Precision data 338
Figure 21.5 Recall data 339
Figure 21.6 Loss data 339
Figure 21.7 F-score data 339
Figure 46.1 Samples and statistical distribution of ISIC 2019 Dataset 342
Figure 46.2 Melanoma detecting Uging DenseNet-201 344
Figure 46.3 Feature concatenation process of DenseNet 344
Figure 46.4 Proposed model performance graph for training and validation 344
Figure 47.1 Electrical circuit of a solar cell 347
Figure 47.2 Solar module connections 347
Figure 47.3 Experimental setup 348
Figure 47.4 Shading conditions 348
Figure 47.5 Output curves (a) TCT (b) PCR 348
Figure 47.6 Global power plot 349
Figure 47.7 Shading loss plot 349
Figure 47.8 Execution ratio plot 349
Figure 47.9 Fill factor plot 350
Figure 47.10 Experimental curves 350
Figure 48.1 Showing labeled data for supervised ML 354
Figure 49.1 Proposed methodology 361
Figure 51.1 Role of NLP 374
Figure 51.2 Domains of sentimental analysis 375
Figure 52.1 Data flow architecture 380
Figure 53.1 Healthy leaf/Early blight/Late blight 386
Figure 53.2 Process flow diagram 387
Figure 54.1 Process flow chart 394
Figure 55.1 AlexNet architecture 400
Figure 55.2 Problem identification using AlexNet 401
Figure 55.3 VGG16 model architecture 401
Figure 55.4 RESNET50 model architecture 402
Figure 56.1 Alice’s neural network setup 408
Figure 57.1 Animal Intrusion Detection Approaches 414
Figure 58.1 Blockchain’s Role in Smart Cities 420
Figure 61.1 Mobilenetv3 architecture 439
Figure 62.1 Diagram for rainstorm prediction system 446
Figure 64.1 459
Figure 65.1 465
Figure 67.1 System architecture 482
Figure 67.2 Working diagram 483
List of figures xix
Figure 67.3 Image categories minted 484
Figure 67.4 Trends in transaction 484
Figure 68.1 Proposed approach for application 488
Figure 69.1 Methodological Process Flow Chart 494
Figure 69.2 Metric analysis 494
Figure 70.1 Data Model (Star Schema) 502
Figure 70.2 Final dashboard created 502
Figure 70.3 Revenue tooltip for trend by week 502
Figure 72.1 DWT based image fusion 515
Figure 72.2 DCT based image fusion 515
Figure 72.3 NSCT based image fusion 516
Figure 72.4 Curvelet transform flow graph 516
Figure 72.5 (a) Picture 1, (b) Picture 2, (c) DCT based Fusion, (d) DWT based
Fusion, (e) NSCT based fusion, (f) Curvelet based fusion 517
Figure 72.6 (a) CT Picture, (b) MRI Picture, (c) DCT based fusion, (d) DWT based fusion,
(e) NSCT based fusion, (f) Curvelet based fusion 517
Figure 76.1 Different face templates 544
Figure 76.2 Different types of matching 544
Figure 76.3 Facial Recognition system working 544
Figure 76.4 Fingerprint Recognition 544
Figure 76.5 Processing of data 544
Figure 76.6 Different types of fingerprint postures 545
Figure 76.7 Fingerprint matching 545
Figure 77.1 Architecture of the application 550
Figure 78.1 MERN stack development 557
Figure 79.1 System architecture block diagram 564
Figure 80.1 Block layout for the lip-reader system 570
Figure 80.2 Architecture of ResNet50 572
Figure 81.1 Block diagram 578
Figure 82.1 583
Figure 82.2 583
Figure 84.1 System flow diagram 602
Figure 85.1 Data driven diabetes prediction 607
Figure 85.1 612
Figure 85.2 612
Figure 88.1 Correlation matrix of major Parameters 623
Figure 88.2 Accuracy and time limitations of ML models in cloudburst prediction 624
Figure 88.3 Efficiency of ML algorithms for Cloudburst Predicitons 626
Figure 89.1 Basic working flowchart of proposed model 629
Figure 89.2 Training phase of face Identification 630
Figure 89.3 Testing phase of face Identification 631
Figure 89.4 KNN sample case image 631
Figure 89.5 PCA sample case image 632
Figure 89.6 Sample image 633
Figure 90.1 Techniques for remote security management 637
Figure 90.2 Security Server Diagram 639
Figure 90.3 Changing Default Settings 640
Figure 90.4 Downloading Firewall 640
Figure 90.5 Activating Firewall 640
Figure 91.1 Accuracy result of decision tree classifier 644
Figure 91.2 Accuracy results of Random Forest Classifier 645
xx List of figures
Figure 91.3 Numerical Distribution of Dataset 646
Figure 91.4 Part of the Training Dataset with CSV Format 646
Figure 91.5 Models Comparison 646
Figure 92.1 Demonstrates the correlation between the packet delivery ratio and the number of
malicious nodes throughout a simulation duration of 100 seconds 652
Figure 92.2 Illustrates the correlation between the packet delivery ratio and the number of
malicious nodes, employing a simulation duration of 100 seconds 652
Figure 94.1 Working procedure of fake video detection 666
Figure 94.2 Data flow diagram of CNN Based Model 667
Figure 94.3 667
Figure 95.1 Most popular sports in the Olympics 675
Figure 95.2 Gender distribution 675
Figure 95.3 Number of participants in each season 675
Figure 95.4 Distribution of medals by season 676
Figure 95.5 Age distribution of Olympic athletes 676
Figure 95.6 Medal victories by age for gold, silver, and bronze medals 676
Figure 95.7 Age distribution of Olympic athletes 676
Figure 95.8 Most popular sports for Indian athletes 677
Figure 95.9 Gender distribution of Indian athletes 677
Figure 95.10 Age distribution of Indian athletes 677
Figure 95.11 Medal distribution for India 678
Figure 96.1 Methodology of crop prediction 682
Figure 96.2 Process flow model 682
Figure 96.3 Accuracy comparison of 3.2 models implications 684
Figure 97.1 Working model 687
Figure 97.2 No of calories for each trimester 688
Figure 97.3 Customized diet chart using random method 688
Figure 97.4 Architecture of the application 689
Figure 97.5 Differences between nearest-neighbor 689
Figure 97.6 Represents the nearby hospitals 690
Figure 97.7 Nearby hospitals 691
Figure 98.1 Popular tourism platforms 693
Figure 98.2 Architecture of the platform 693
Figure 99.1 698
Figure 101.1 Distributed system architecture 712
Figure 101.2 Reliability of fault in the distributed system 713
Figure 102.1 Performance evaluation using ROC curve 720
Figure 102.2 Performance evaluation using confusion matrix 720
Figure 102.3 Learning curve using logistic model 720
Figure 103.1 Flow diagram of trust-based system 725
Figure 103.2 Correlation between Fog Node Trust and Credibility 726
Figure 103.3 Correlation Between F-Rate and Number of Incoming Tasks 726
Figure 103.4 Energy Consumption Comparison Chart 726
Figure 103.5 Resource Optimization Graph Theta 727
Figure 104.1 Pediatric brain tumor detection and classification studies different model and their
accuracy733
Figure 104.2 Flow diagram of F1 score (adapted from toward data science) 734
Figure 105.1 Proposed flow chart 740
Figure 106.1 Workflow of object detection 745
Figure 106.2 Leaf image examples for object detection 747
Figure 106.3 Accuracy 748
List of figures xxi
Figure. 106.4 Loss 748
Figure 106.5 Precision 748
Figure 106.6 Recall 749
Figure 106.7 F1 Score 749
Figure 108.1 757
Figure 108.2 757
Figure 109.1 Data flow diagram 761
Figure 111.1 Project architecture 771
Figure 111.2 Workflow flowchart 773
Figure 112.1 Methodological process flow chart 779
Figure 113.1 784
List of tables
Table 1.1 Comparison of different optimal control strategies of the mean EQ and EC in storm
weather condition 4
Table 1.2 Statistical results of the proposed model’s performance evaluation (R2 and MSE) for
different stages of modeling (training, validation, testing and all data sets) 5
Table 1.3 The results of three different models 7
Table 1.4 Model Performance on the training, validation and test sets (R2 values) 7
Table 2.1 State-of-the-art of green cloud 18
Table 2.2 Green cloud SAAS performance metric 21
Table 5.1 Respondent’s opinion ranking of construction tasks performed by ICTs 45
Table 5.2 Ranking of the benefits as per respondent’s opinion 46
Table 5.3 Ranking of hindering factors for ICT adaptation 47
Table 6.1 Algorithm and accuracy 56
Table 6.2 Comparison with previous work 56
Table 7.1 Participants distribution on the basis of Age 61
Table 7.2 Participants distribution on the basis of Gender 61
Table 7.3 Participant’s distribution based on their educational background 62
Table 7.4 Responses in favor of online mode of learning 62
Table 8.1 Simulation Parameters 71
Table 10.1 Simulation parameters 83
Table 12.1 Mechanisms of NPT of mHealth in medicine 103
Table 12.2 Research questions and methodology 103
Table 15.1 Simulation requirements 125
Table 17.1 Results of algorithms used 140
Table 20.1 Execution evaluation of the proposed plan of action 162
Table 20.2 Execution comparison 162
Table 21.1 VANET Comparative Tables 172
Table 26.1 Technical layers in an urban infrastructure 204
Table 27.1 Experts’ contributions in web mining 212
Table 28.1 Comparative analysis of cryptographic approaches in cloud security 223
Table 30.1 Various solutions of honeypot 237
Table 31.1 Evaluation of results according to the datasets. 244
Table 41.1 Comparison with other methods 312
Table 46.1 Comparison with other state-of-the-art methods 344
Table 47.1 Quantitative Simulation Results 349
Table 49.1 Students’ employability features 361
Table 49.2 Performance Metrics for ML Algorithms 363
Table 51.1 Dataset popularly used 375
Table 53.1 Number of images with diseases 387
Table 53.2 Model performance comparison 389
Table 56.1 Comparing cryptographic image concealment with neural networks to traditional
steganographic techniques 410
Table 56.2 Comparison of PSNR, SSI, and BER 410
Table 57.1 Results comparison 416
List of tables xxiii
Table 59.1 Comparative Analysis 429
Table 61.1 Comparative study of different papers 440
Table 62.1 Critical analysis comparison 447
Table 66.1 Accuracy table 477
Table 69.1 Critical analysis comparison 495
Table 69.2 Parametric analysis comparison 496
Table 70.1 Key metrics list top of form 501
Table 70.2 Insights from dashboard 503
Table 72.1 Performance metrics for transform-based algorithms’ multi-focus pictures 517
Table 72.2 Performance metrics for transform-based approaches’ multi-model images 518
Table 73.1 Comparison of traditional and blockchain 523
Table 73.2 Comparative analysis of consensus algorithms based on characteristics 524
Table 89.1 Result of model based on features 633
Table 92.1 Parameters of the simulation environment 652
Table 93.1 Research methodology table 659
Table 93.2 Hypothesis test summary 660
Table 93.3 Model summary 660
Table 93.4 Statistical significance of the model 660
Table 93.5 Estimated model coefficient 661
Table 96.1 Comparative Study 683
Table 104.1 732
Table 105.1 Overview of related works 739
Table 107.1 754
Table 112.1 Critical analysis comparison 779
1 A systematic review on recent
controllers for an industrial wastewater
treatment plant
K. V. Reshma1,a, A. Vimala Juliet1,b, and G. Glan Devadhas2,c
1
Department of Electronics and Instrumentation Engineering, SRM Institute of Science and
Technology, Kattankulathur Campus, Chennai, India
2
Department Electronics and Instrumentation Engineering, Vimal Jyothi Engineering College,
Kannur, Kerala, India
Abstract: A wastewater treatment plant (WWTP), also referred to as a sewage treatment plant or a water
reclamation facility, is a facility designed to treat wastewater and sewage from residential, commercial, and
industrial sources. Wastewater treatment (WWT) facilities are essential for maintaining the well-being of our
communities by safeguarding water supplies, maintaining public health, and minimizing environmental pol-
lution. The specific design and processes employed by a WWTP can change based on variables like the size of
the facility, the characteristics of the wastewater, and the required effluent quality standards. This paper pre-
sents the comprehensive technological review of PID, FLC, PLC, and MPC controllers for an industrial WWT
plant with advantages and disadvantages of the same and results are shown. In addition, this overview includes
Artificial intelligence is revolutionizing the field of water treatment by enhancing model prediction and control
strategies.
Keywords: Deep learning model (DNN), fuzzy logic controller (FLC), machine learning (ML), model predic-
tive controller (MPC), programmable logic controller (PLC), proportional integral derivative controllers (PID),
wastewater treatment plants (WWTPs)
1. Introduction (Li et al., 2021; Fu et al., 2023). Because of
the importance of addressing sustainability
WWTPs are amenities dedicated to treating issues in WWTPs, new technologies and pro-
incoming waters and returning them to their cedures have been created often.
natural cycle, where the pollution reduction In this light, control mechanisms have
process of urban residual waters is carried been developed and implemented at WWTPs
out. The primary objective is to minimize the with the aim of keeping pollution levels
concentration of total suspended particles, within the limits set by the rules. However,
dissolved organic matter, phosphorus, and WWTPs come in a wide variety of designs, AQ1: As
per style,
nitrogen-derived components in the treated each optimized for a different set of needs. the refer-
waters such that they are safe for the aquatic As a result, a significant amount of time and ence style
should be
life and surrounding ecosystems. The cost effort has been spent over many years (Han et num-
of running a wastewater treatment (WWT) al., 2019; Iratni et al., 2019; Newhart et al., bered.
Please
plant continues to be a major concern, hence 2019) on the development of sophisticated check and
efforts are being made to improve WWT sys- controllers for WWT facilities. The conven- review for
tems while keeping operational expenses low tional PID controllers, however, impossible to clarity.
[email protected]; [email protected]; [email protected]
a
CHECKED AND CONFIRMED
DOI: 10.1201/9781003606635-1
2 Innovations in Computational Intelligence: Trends and Technologies
extensively employed for complicated process lot of data and some serious pre-processing
for treating wastewater (Subari et al., 2022) to work well. It’s crucial to remember that
because to the occurrence of nonlinearities AI can’t replace mechanical models. There-
and turbulences, as well the contemplation fore, the goal of Using Machine Learning for
of particular optimization targets for unfa- Wastewater Treatment is to make better deci-
miliar systems. The reality is that there is a sions depending on projected performance
dearth of AI-based optimization approaches for existing facilities (Sundui et al., 2021;
that work well in a complex, uncertain, and Nourani et al., 2021).
nonlinear setting. Therefore, figuring out how Because of their superior feature extrac-
to acquire intelligent optimum controllers for tion, self-learning, and generalization abilities,
complicated nonlinear systems have been deep learning algorithms have found various
seen as a crucial step in the development of fantastic Implementations in recent years
the area of advanced control. across a wide range of areas. In DL, RNNs
The utilization of single-objective opti- with recurrent links and hidden layers are
mization approaches has generally applied recommended for analyzing difficulties with
in the field of wastewater management of a a historical context, while LSTM is intended
considerable duration, with the aim of opti- to improve the system’s ability to deal with
mizing the effectiveness of the WWT process. dependencies over the long run. However, the
It’s tough to pin down the precise weighting duration of time for training is a major factor
elements, however. As a result, several dif- in DL. Recent advances in genetic based con-
ferent kinds of multi-objective optimization trol algorithms for biological WWT facilities
algorithms, such MOPSO and MODE (Han have allowed for significant improvements
et al., 2022) have been created. Traditional in effluent quality and reductions in operat-
multi-objective optimization algorithms ing costs. This type of controller is primarily
offer numerous benefits in WWT methods, intended to enhance the WWTP’s efficiency
but they also have drawbacks, such as weak by adjusting the set points of the lower-level
global search ability and high levels of unpre- Supervisors, such as PID controllers, in addi-
dictability in the ideal position section. tion to offers insights into the use of genetic
Intelligent technologies based on machine algorithms in the development of advanced
learning (ML) are now being used in the water control systems for sewage treatment facili-
industry to solve pressing problems worldwide. ties. It’s also possible to cut down on process-
Model capabilities for the dynamic aspects of a ing time by employing PSO, DE, or ACO in
WWTP operation have also been improved via the right situation.
the use of a hybrid method that blends several
ML techniques. It has been shown that artifi-
cial neural networks (ANNs) can learn and 2. Literature Review
construct mathematical models of extremely To conserve energy and cut down on emis-
non-linear interactions, making them useful as sions, researchers (Li et al., 2021) proposed
predicting tools in a variety of contexts. the development of an innovative CLSTMA
Artificial Neural Network Ant Lion for observing effluent nature within wide-
Optimization (ANN-ALO), (PSO-ANN) ranging paper industry treatment facility
(Mahadeva et al., 2022) is proposed to using continuous convolution neural network
enhance the effectiveness of conventional (CNN), long short-term memory (LSTM),
multi-objective optimization techniques on a and attention mechanism (AM). For each
global scale and prevent the algorithm from of the study’s output variables, the authors
settling to the point of reaching a regional employed standard performance measure-
maximum. However, AI models do need a ments such as Pearson correlation coefficient
A systematic review on recent controllers for an industrial wastewater treatment plant 3
(R), the Root Mean Squared Error (RMSE), effectiveness and durability of WWT pro-
and the Mean Absolute Error (MAE). The cesses. They also emphasize the need for fur-
findings demonstrated that the hybrid deep ther research and development in this field to
learning model achieved the highest levels of overcome the existing challenges and realize
accuracy, precision, and recall compared to the full potential of control Techniques for the
the other models. The authors argue that their treatment of sewage.
model may be used to lessen the ecological The degradation of proteins and amino
toll of paper manufacturing by increasing the acids in organic waste is the principal source
efficacy of WWT systems in the paper sector. of ammonia in wastewater, Subari et al. (2022)
The RME of the CLSTMA model for the set out to compile a comprehensive overview
effluent Chemical Oxygen Demand (CODeff) of the utilization of biological treatment
was lowered by 23.3–31.55%, the MAE was techniques has emerged as a viable alterna-
condensed by 38.89–74.50%, and the R of tive strategy for the removal of ammonia in
the CLSTMA model was enhanced by 8.29– water and sewage purification facilities. This
11.86% when compared to different mod- article discusses the present worldwide pollu-
els (CNN, LSTM, and CLSTM models). The tion condition of ammonia, the sources about
RMSE of the CLSTMA model decreased by ammonia impurity, and the usual restrictions
10.26% and 9.92% for the effluent suspended on regulation of ammonia concentration pos-
solids (SSeff), the MAE decreased by 5.37% sible dangers, recorded instances, and more.
and 3.44%, and the R of the CLSTMA model In addition, AI is being used into water
rose by 15.13% and 37.21%. Results from purification model prediction and control
computer simulations show that the suggested systems. The ANN is a talented computer
CLSTMA model might be very useful for keep- mechanism that can improve both the WTP
ing an eye on the WWT system in the paper sec- and WWTP in general. Due to the nonlinear-
tor so that cleaner products can be made. ity of their dynamic behavior, variability in
PID control, as well as more modern meth- parameter values over time, constraints on
ods like MPS, FLC and neural network con- manipulated variables, interdependencies
trol, are all discussed in detail by (Iratni et between manipulated and controlled vari-
al., 2019). The authors also explore how to ables, presence of unaccounted Interruptions,
effectively regulate the WWT process by inte- delays in input and measurement processes,
grating control technology with optimization and other factors, a standalone conventional
methods and real-time monitoring systems. PID controller is inadequate for effectively
The complexity and variety of wastewater tuning parameters or achieving optimal
composition, the absence of dependable sen- control outcomes in a WTP or WWTP. It is
sors, and the high cost of installing advanced necessary to know the mathematical connec-
control technologies are all highlighted in this tion between the system’s input and output
article as problems that control systems must in order to design an appropriate controller
overcome in WWT operations. Non-linear and also compares the outcomes of several
controllers like the NMPC controller and the specific analysis involving the optimization
neural-fuzzy controller can be viable alterna- of PID controllers. Hybridization with other
tives, but they demand a high level of knowl- tuning methods, like Ziegler–Nichols (ZN) or
edge due to the complexity of their algorithms, Cohen-Coon (CC), a Fuzzy control system, or
which can lead to increased computing needs. a metaheuristic approach, like Particle Swarm
In conclusion, the authors suggest that the inte- Optimization (PSO), Gravitational Search
gration of control technologies with advanced Algorithm (GSA), or Grey Wolf Optimization
monitoring and optimization techniques can (GWO), can greatly increase the effectiveness
also lead to significant improvements in the of a PID controller.
4 Innovations in Computational Intelligence: Trends and Technologies
Using a WWT process as an example, Han compares appropriate control options for use
et al. (2022) describe the results of a research during storms. MIOC’s optimization perfor-
into the development of a multi-objective mance may be confirmed by its best-in-class
integrated optimum control method. The mean EQ (7167) and its lowest-in-class mean
authors offer a control system that simultane- EC(3889). With a mean IAE of just (0.041),
ously maximizes the plant’s efficiency and the MIOC clearly outperforms conventional,
efficacy of the WWT process, therefore reduc- incrementally optimized control approaches.
ing both operational costs and environmen- To predict the operational effectiveness of
tal effect. Existing multi-objective optimum a desalination facility, Mahadeva et al. (2022)
control approaches, however, tend to analyze offer a model of PSO-ANN. The purpose
the procedure of maximizing efficiency and of this research is in order to improve the
the control method separately, which may dependability of projections and enhance the
lead to the set-point fluctuating wildly and benefits of the Reverse Osmosis (RO)-based
being difficult to follow, and therefore dete- desalination technology plant. Modelling and
riorate the control and practical performance simulation were accomplished with the help
of WWTPs. In order to address this issue, of MATLAB 2019a’s neural network tool-
this research suggests a Multi-Objective Inte- box. The literature implies that changes to
grated Optimal Control (MIOC) approach the modelling parameters of soft computing
for the Sludge Processing Facility. To accom- models may lead to significant gains. In order
plish synchronized optimization and control to determine what those ideal values are, this
action, a Multi-Objective model predictive study provides a comprehensive list of mod-
control framework is designed. elling parameters and an in-depth systematic
Both the model of controlled variables and examination of them. Table 1.2 displays the
the model of operational indices utilized in summary statistics for the suggested mod-
MIOC were constructed using AFNNs. In eling efficiency assessments (R2 and MSE)
BSM1, EQ and EC are used to assess WWTP’s throughout training, validation, testing, and
efficiency. Control performance and optimiza- all datasets. Table 1.2 shows details.
tion performance studies are examined using Model 2 had the greatest training results
simulation in all climates. Example: Table 1.1 (R2 = 99.8%, MSE = 0.000) while Model 1 had
Table 1.1. Comparison of different optimal control strategies of the mean EQ and EC in storm
weather condition
Optimal NH4 Ntot SS COD BOD EQ kg EC Percentages IAE
control mg L–1 mg L–1 mg L–1 mg L–1 mg L–1 poll units kW h % mg
methods L-1
MIOC 2.87 17.27 13.55 46.89 2,74 7167 3889 12.00 0.041
DMOPSO 3.07 17.48 13,02 47.79 2,78 7512 4089 7,42 0.101
(Han, Liu,
2021)
MOO 2.93 17.91 13.56 48.13 2.83 7583 4169 5.61 0.120
(Sweelapple
et al., 2014)
MOOC 2.98 17.39 13.86 48.64 2.71 7236 4410 0.15 0.054
AQ2: (Han et al.,
Please 2018)
provide
source for P1D 3.06 17.52 13.79 49.52 3.18 7970 4417 0 0.108
all tables. Source:
AUTHOR'S SELF GENERATED
A systematic review on recent controllers for an industrial wastewater treatment plant 5
Table 1.2. Statistical results of the proposed model’s performance evaluation (R2 and MSE) for
different stages of modeling (training, validation, testing and all data sets)
Model Performance evaluations of the PSO-
ANN models by different stages
Training (75%) validation (20%) Testing (05%) All (100%)
R2(%) MSE R2(%) MSE R2(%) MSE R2(%) MSE
Model - 1 99.0 0.005 98.7 0.014 99.6 0.003 98.9 0.007
Model - 2 99.8 ~0.000 98.5 0.016 94.7 0.057 99.0 0.006
Model - 3 99.3 0.004 98.9 0.011 98.6 0.015 99.1 0.006
Model - 4 99.1 0.005 98.6 0.014 96.9 0.024 98.8 0.008
Source:
Author's self generated
the ultimate testing outcomes (R2 = 99.6%, quality is enhanced by 5.6% in Scheme 2 and
MSE = 0.006); Model 3 had the ultimate by 20.3% in Scheme 3 as compared to Scheme
outcomes across the board. Overall, this arti- 1. As a consequence, scheme 3 has shown to
cle gives important insights into the use of be an effective method for treating wastewa-
advanced machine learning techniques, such ter. A neural network model-based regulat-
as PSO-ANN, for enhancing the efficiency of ing strategy might be used in the future to
water treatment desalination facilities. The enhance effluent quality while simultaneously
optimized model presented in this study can decreasing electrical energy usage.
help plant operators make informed decisions Ning et al. (2022) propose the PLC with
and optimize the plants performance, leading incremental proportion and incremental
to improved efficiency and cost savings. (IPI) control algorithm to preserve the con-
Saravana Kumar et al. (2021), begin by servation of the aquatic ecosystem and pro-
the difficulties connected with managing the mote supportable growth by automating the
quality of effluent from WWT plants, particu- adjustment of the dissolved oxygen level in
larly in light of increasingly stringent regula- the aeration tank during the WWT process.
tory standards. They introduce the concept of A BP-NN was utilized to improve method of
fuzzy logic control, which is a type of control regulation. Automatic DO content manage-
system that uses linguistic variables to model ment within the aeration tank presented in
complex and uncertain systems. The sug- Figure 1.1 through the PLC’s enhanced IPI
gested study provides two control techniques control algorithm. A PLC-based automated
(ammonia control and total nitrogen man- control system was implemented to miti-
agement) to enhance their performance of gate the astronomical increases in operating
treating wastewater, and the authors propose expenses and energy consumption caused
using a fuzzy logic controller to achieve these by the lack of automation. Under the guid-
goals. The simulation work was done Within ance of the enhanced IPI procedure, the total
the framework of the Benchmark Simulation nitrogen content of wastewater treated in
Model No.1 (BSM1), and this study’s offers the aeration tank was minimized, and its
a supervisory FLC method enhance the efflu- variability was reduced to a minimum. The
ent quality of a WWT facility and provides authors also highlight the challenges faced
numerical evidence of the effectiveness of this by WWT plant, including fluctuating influ-
control strategy. It’s noteworthy that the efflu- ent characteristics and variations in flow rate,
ent quality is maintained despite the higher which can lead to inefficiencies in treatment
AE and PE use in this system. The effluent and increased energy consumption and cost
6 Innovations in Computational Intelligence: Trends and Technologies
savings. Overall, the article provides a useful recycle loop. Even still, MPC may struggle to
example of how PLC-based control Systems deal with input disturbances and frequently
may be utilized to maximize productivity of needs many model coefficients to describe a
municipal WWT plant. response. The investigation found the MPC
Development and evaluation of a neuro- system was able to effectively control the
fuzzy-based Model Predictive Controller treatment process in real-time, reducing
(MPC) approaches that can estimate the key energy consumption and maintaining high
process factors and provide the appropriate levels of treatment efficiency. This paper con-
level of aeration to overcome the factors such cludes that the use of machine learning-based
as process uncertainty, seasonal variation, MPC systems able to offer significant benefits
and nonlinearity was discussed by Bernardelli to WWT plants, improving both their envi-
et al. (2020) in order to attain a streamlined ronmental and economic performance.
and cost-effective functioning. This article In order to forecast the dose of coagu-
describes the process of data collection and lant, Fang et al. (2022) employed the features
pre-processing, which involved the use of parameters such as pH, turbidity, electrical
sensors to monitor the different factors of conductivity, and flow velocity that character-
the treatment plant, such as flow rate, dis- ize influent water quality. Authors note that
solved oxygen levels. The data was then used traditional dosing algorithms for WWT plants
to train the machine learning models, which are often based on empirical formulas or trial
were integrated into the MPC system. and error, which can be inefficient and may
More testing using a weekly switch the result in suboptimal treatment outcomes.
between the two controllers (EW and CC) has In contrast, the proposed intelligent dosing
proven that the EW controller is more reli- algorithm model uses methods in the arena of
able at improving sewage quality while using artificial intelligence, such as machine learning
less energy. Especially by reducing air flow at neural networks to optimize dosing decisions
periods of low load times, but also by pre- based on real-time data from the treatment
venting oxidation during peak loads, as long plant and to suggest a sophisticated Light-
as the nutritional ceiling is never reached. GBM-based dosage prediction method, which
Figure 1.2 shows the process plan of a uses these data to estimate the dosage of coag-
WWTP that uses MPC. The dotted lines rep- ulant. The LightGBM algorithm’s findings for
resent the two control loops: the DO con- predictive power demonstrate its dominance.
trol loop controls the airflow (Ua), and the Overfitting and dataset compatibility are two
internal recycle (Qr) is controlled by the drawbacks of the LightGBM algorithm.
AQ3:
Please
provide
source Figure 1.1. Automatic control flow of aeration tank DO content by the improved IPI control
for all algorithm.
figures.
Source: Author's own resource
A systematic review on recent controllers for an industrial wastewater treatment plant 7
Figure 1.2. shows the process scheme of the WWTP to which MPC is applied.
Source:
AUTHOR'S SELF GENERATED
Table 1.3 provides an evaluation rubric capability is shown by these results. Overall,
for predicting outcomes using various this paper highlights the potential benefits of
approaches. Evaluating the efficacy of three using artificial intelligence methods that may
models using experimental data. The experi- be used to boost the performance of waste-
mental findings demonstrate that the R2 water remediation systems.
value reaches a maximum of 0.9957 for the Matheri et al. (2021) describe a study that
dataset model created by LightGBM, while was done by a group of researchers from dif-
the RMSE and MAE values are reduced to a ferent institutions in South Africa. They used
minimum of 0.3049 and 0.1474 correspond- data from a WWT plant to train and test
ingly. LightGBM’s advantage in predictive ANN. The models were built using a Multi-
layer Perception model and were utilized for
Table 1.3. The results of three different models forecasting potential issues with WWT sys-
tems, such as chemical composition, inherent
Model R2 RMSE MAE MAPE flow rate, and better safety factor. The result
MLR 0.9952 0.7504 0.4458 6.29% demonstrates that ANNs are an effective tool
DT 0.9748 0.4949 0.2978 4.78% for developing integrated wastewater systems
LightGBM 0.9957 0.3049 0.1474 1.34% and simulating the designs of WWTPs. For
the purpose of the real-time prediction of
Source:
nutrient removal efficiency, a neural network
based on the LSTM network (a descendant
Table 1.4. Model Performance on the training, of the RNN) is also suggested. However,
validation and test sets (R2 values) ANN is computationally intensive, difficult
Training Validation Test set to explain, and data-intensive. WWT process
set set issues were predicted using the artificial intel-
TSSe_RF 0.934 0.862 0.920 ligence (AI)-based ANN prediction model,
developed on the popular MATLAB platform
TSSe_ 0.935 0.892 0.920
DNN
and enhanced with machine learning (deep
learning). The research also stresses the need
PO4_RF 0.905 0.870 0.886
of taking into account numerous removal
PO4_DNN 0.904 0.908 0.872 input factors while creating such models to
Source: guarantee precise forecasts.
8 Innovations in Computational Intelligence: Trends and Technologies
Wang et al. (2021) present a unique ML- and require more training data to learn effec-
based method that aims to enhance effluent tively. In addition, GMM is highly sensitive
quality control in WWTPs by showing how to outliers. The best models for each algo-
practical factors and discharge parameters are rithm under test were determined by running
related. The author collected a large dataset multiple tests with different combinations of
of process data, including influent and efflu- their main parameters and selecting the ones
ent water quality parameters, operational lowest errors according to metrics taken into
conditions and environmental factors. This account, including MSE, RMSE, MAE, and
dataset was used to train several ML methods MAPE. With least RMSE, MSE, MAE, and
such as ANNs, decision trees, random forests, MAPE, the Transformer approach obtained
and linear regression. The framework is com- the highest Performance for predicting all of
prised of Random Forest (RF) models, Deep the critical variables.
neural network (DNN) models, and Variable When it comes to forecasting TSS and
Importance Measure (VIM) models and the NO3, LSTM methods fared the best, while
partial Dependence Plot (PDP) analyzers, and ANN and GMM algorithms produced the
it uses a unique method to explain temporal best results for NH4 and BOD5, respectively.
delays between processes. Thirty-two opera- Concerning the predictability of outcomes, it
tional variables and two effluent parameters— is lastly worth noting that the Transformer
Total Suspended Solids in Effluent (TSSe) and algorithm necessitates more time and effort
Phosphate in Effluent (PO4e)—are analyzed to train and make predictions. In accord-
in order to create RF models, which are then ance with the selected metrics, both ANN
verified using DNN models as references and and LSTM provided accurate predictions
demonstrated to be appropriate for VIM and while simultaneously using less computing
PDP in terms of clarifying the specific effects resources and a shorter amount of time to
of these two operational variables on TSSe train and provide predictions.
and PO4e. The model performance of DNN Classification models were shown by
and RF on training, validation and test data Wodecka et al. (2022) to be a useful tool
set as shown in Table 1.4. In conclusion, both for predicting changes in effluent character-
the RF and DNN models accurately forecast istics at the intake of WWT facilities, when
all data sets and demonstrated outstanding these values are directly proportional to the
Performance in generalization for TSSe and flow rate of incoming wastewater. In the
PO4e. considered proposal, the levels of quantified
Salles et al. (2022) suggested using machine wastewater characteristics parameters were
learning algorithms to forecast important divided into lower (lower-than-average indi-
variables (monitoring, control, and optimi- cator values), average (normal and maximum
zation of industrial processes) in the WWT values), and higher (increasing values). The
method. The value of the TSS, BOD5, NH4, results of the computations showed that the
and NO3 variables were approximations values of the chosen wastewater quality indi-
using the Gaussian Mixture Models (GMM), cators (BOD5, COD, TN, TP) can be detected
LSTM, ANFIS, Transformer and ANN algo- with adequate accuracy using the calculated
rithm. Greater computational resources and ensemble models blending support vector
longer preparation time are needed for the machines and boosted trees. The statistical
transformer algorithm. ANFIS has a limi- parameters (minimum, average, maximum,
tation of the type, quantity and position of and standard deviations) of seasonal fluc-
membership functions, the curse of dimen- tuation in the amount and quality of BOD5,
sionality, and the fact that LSTM models COD, TN, and TP were identified, with a
are more complex than conventional RNNs split between winter and autumn/spring. For
A systematic review on recent controllers for an industrial wastewater treatment plant 9
the purpose of maximize the effectiveness of included. The results of this study will help
WWT, this research finds that machine learn- professionals manage WWTPs based on data.
ing approaches may be utilized to forecast the The authors concluded that deep WWTP
quality of incoming wastewater. The findings learning-based models have the potential to
of this investigation could be applied to cre- be useful tools for forecasting key features
ate a real-time monitoring system for WWT of WWT plant and advise future research to
plants, which allow operators to improve take into account sensor networks that can
the treatment process based on the predicted provide data sets with greater dimensions
quality of incoming wastewater. and frequency.
Cheng et al. (2020) utilized training Figure 1.4 depicts the core of the work
and testing data collected over time from a of Do et al. (2021), which is the incorpora-
WWT facility in Norway. Standard LSTM tion of genetic algorithms (GAs) into the
and GRU, exponentially smoothed LSTM, Benchmark Simulation Model1 (BSM1) for
adaptive LSTM, and smoothed LSTM are the purpose of designing a higher-level con-
only some of the six deep learning models trol system for WWTPs. This research article
built using RNN-based soft sensors. Models proposes a unique Genetic Optimization for
built on GRUs are more efficient than LSTM Biological Wastewater Treatment Plants with
models because they converge more quickly. the objectives of increasing discharge qual-
For all important aspects, the LSTM soft- ity while decreasing operational expenses.
sensor performs best overall, succeeded by The suggested controller is meant to optimize
the exponentially-smoothed GRU and the the WWTP by adjusting the set points of the
LSTM. The dynamic models underperformed lower-level controllers, such as the PID con-
the other models in terms of their ability to trollers. This research provides the outcomes
predict the future. All that is needed are the of a simulation research was carried out to
data itself, as the offered models produce test the performance of proposed control-
good forecasting results. Figure 1.3 is a sche- ler. In this work, a GA-based hierarchical
matic illustration of the forecasting system controller was presented, with the bottom
that we propose. Using RNN-based soft sen- tier controller regulating S0,5 by manipula-
sors, this research constructed RNN-based tion of Kla,5, and the upper-level controller
models to predict WWTP essential param- adjusting S0,5 per SNH,4. In the higher-level
eters such influent flow, temperature, BOD, control, GA is utilized to figure out KSO,5
chloride, and energy consumption. To fur- and BSO.5 in order to achieve three prede-
ther reprocess unprocessed Data before deep termined goals in a variety of climates and
learning, the exponential smoothing filter is atmospheric circumstances.
Figure 1.3. Workflow of the proposed deep learning-driven forecasting procedure.
Source:
AUTHOR'S SELF GENERATED
10 Innovations in Computational Intelligence: Trends and Technologies
A tank 5 has recently implemented the for designing a suitable treatment system
desired control. For each of the three weather to ensure the safety, efficacy, and quality of
scenarios investigated, the simulated out- the treated wastewater. Here, the limitations
comes will be compared to those obtained faced by the existing research are described as
by using simply the lowest controller, which follows. PID, FLC, PLC, and MPC controllers
is an ILC paired with a PI regulator for the are all commonly used in WWT processes;
wastewater influent data. As a result, the However, the PID controller’s inefficient
suggested strategy might be used to various parameter coordination and substantial time
contexts to further decrease the OCI and delay make it difficult to provide precise con-
EQI. Overall, the research sheds light on trol, especially on complex systems or sys-
how genetic algorithms might be included tems that demand high performance.
into the planning of advanced controls for Moreover, FLC controller has disadvan-
WWT facilities. There may be far-reaching tages of lower speed and longer run time and
consequences for the environment and pub- lack of real time responses. This PLC-based
lic health if the suggested controller is not control system is difficult to adapt or replace,
implemented to enhance the performance of requires extensive training and is always diffi-
WWTPs. Furthermore, time spent might be cult to identify errors. In addition, MPC may
cut down by using PSO, DE, or ACO in effi- struggle to deal with input disturbances and
cient ways. frequently needs many model coefficients to
describe a response.
Since traditional mechanistic models suffer
3. Discussion from a number of drawbacks, scientists have
This paper comprises the discussion about turned to Machine Learning (ML) techniques
the waste water and their various waste to simulate WWTP processes. Therefore, the
water treatment methods which provides existing research utilized LightGBM, RF,
different types of techniques from different ANN, ANFIS, LSTM, and GMM. However,
methods. WWT processes have advanced fast ANN requires lots of computational power
with scientific breakthroughs in recent dec- hard to explain and it requires lots of data,
ades. The treatment effectiveness with respect over fitting and data set compatibility are two
to their benefits, uses, and limitations have drawbacks of the LightGBM algorithm. RF
been thoroughly addressed. Understanding model is not able to deal with unbalanced
the characteristics of wastewater are essential and missing data, also its runtimes are fast,
Figure 1.4. Benchmark Simulation Model 1(BSM 1).
Source:
AUTHOR'S SELF GENERATED
A systematic review on recent controllers for an industrial wastewater treatment plant 11
greater computational resources and longer Acknowledgement
preparation time are needed for the trans-
former algorithm. ANFIS has a limitation of The authors gratefully acknowledge the stu-
the type, quantity and position of member- dents, staff, and authority of Electronics and
ship functions, the curse of dimensionality Instrumentation department for their coop-
and the fact that LSTM models are complex eration in the research.
than conventional RNNs and require more
training data to learn effectively. In addition, References
GMM is highly sensitive to outliers. Moreo- [1] Bernardelli, A., Marsili-Libelli, S., Manzini,
ver, DNN is utilized to investigate the Indus- A., Stancari, S., Tardini, G., Montanari, D., ...
trial Wastewater and Venier, S. (2020). Real-time model pre-
Treatment however, DNN algorithm are dictive control of a wastewater treatment
not able to perform with smaller data and plant based on machine learning. Water Sci-
critical interpretability. ence and Technology, 81(11), 2391–2400.
[2] Cheng, T., Harrou, F., Kadri, F., Sun, Y., and
Leiknes, T. (2020). Forecasting of wastewa-
4. Conclusion ter treatment plant key features using deep
learning-based models: A case study. Ieee
The findings and analyses presented in this Access, 8, 184475–184485.
study underscore the paramount importance [3] Do, H. T., Van Bach, N., Van Nguyen, L.,
of WWT. Failing to address wastewater can Tran, H. T., and Nguyen, M. T. (2021). A
have dire consequences for both human well- design of higher-level control based genetic
being and environmental factors. algorithms for wastewater treatment plants.
Looking ahead, the significance of water Engineering Science and Technology, an
treatment is set to escalate both globally and International Journal, 24(4), 872–878.
particularly in India. The gradual deteriora- [4] Fang, X., Zhai, Z., Zang, J., and Zhu, Y.
tion of groundwater quality, exacerbated by (2022, April). An intelligent dosing algo-
rithm model for wastewater treatment plant.
industrial waste and pollution, has led to the
In Journal of Physics: Conference Series (Vol.
emergence of hazardous diseases. To combat 2224, No. 1, p. 012027). IOP Publishing.
these challenges, WWT methods will play an [5] Fu, X., Zheng, Q., Jiang, G., Roy, K., Huang,
increasingly pivotal role. These techniques L., Liu, C., ... and Wang, Z. (2023). Water
aim not only to minimize water wastage but quality prediction of copper-molybdenum
also to ensure that an ample supply is con- mining-beneficiation wastewater based on
served for future generations. the PSO-SVR model. Frontiers of Environ-
In our perspective, researchers should mental Science and Engineering, 17(8), 98.
give top priority to the creation of WWT [6] Han, H., Liu, Z., Hou, Y., and Qiao, J.
systems that excel in efficiency, energy con- (2019). Data-driven multiobjective pre-
servation, fouling prevention, cost-effective- dictive control for wastewater treatment
ness, and compactness. The survey results process. IEEE Transactions on Industrial
strongly indicate the potential for employ- Informatics, 16(4), 2767–2775.
[7] Han, H. G., Chen, C., Sun, H. Y., and Qiao, J.
ing hybrid methodologies, which could
F. (2022). Multi-objective integrated optimal
yield enhanced precision and superior out- control for a wastewater treatment process.
comes for Industrial Wastewater Treatment Control Engineering Practice, 128, 105296.
facilities. By focusing on these aspects, the [8] Iratni, A., and Chang, N. B. (2019). Advances
field of WWT can usher in transformative in control technologies for wastewater treat-
advancements that address environmen- ment processes: status, challenges, and per-
tal concerns and industrial demands more spectives. IEEE/CAA journal of automatica
effectively. sinica, 6(2), 337–363.
12 Innovations in Computational Intelligence: Trends and Technologies
[9] Li, X., Yi, X., Liu, Z., Liu, H., Chen, T., Niu, using machine learning models. In 2022
G., ... and Ying, G. (2021). Application of International Joint Conference on Neural
novel hybrid deep leaning model for cleaner Networks (IJCNN) (pp. 1–9). IEEE.
production in a paper industrial wastewater [16] Saravana Kumar, S., and Latha, K. (2021).
treatment system. Journal of Cleaner Pro- A supervisory fuzzy logic control scheme
duction, 294, 126343. to improve effluent quality of a wastewater
[10] Mahadeva, R., Kumar, M., Patole, S. P., and treatment plant. Water Science and Technol-
Manik, G. (2022). An optimized PSO-ANN ogy, 84(10-11), 3415–3424.
model for improved prediction of water [17] Subari, F., Harisson, H. F., Kasmuri, N. H.,
treatment desalination plant performance. Abdullah, Z., and Hanipah, S. H. (2022).
Water Supply, 22(3), 2874–2882. An overview of the biological ammonia
[11] Matheri, A. N., Ntuli, F., Ngila, J. C., Seodi- treatment, model prediction, and control
geng, T., and Zvinowanda, C. (2021). Per- strategies in water and wastewater treat-
formance prediction of trace metals and cod ment plant. Malaysian Journal of Chemical
in wastewater treatment using artificial neu- Engineering and Technology (MJCET), 5(1),
ral network. Computers and Chemical Engi- 8–28.
neering, 149, 107308. [18] Sundui, B., Ramirez Calderon, O. A., Abdel-
[12] Newhart, K. B., Holloway, R. W., Hering, dayem, O. M., Lázaro-Gil, J., Rene, E. R.,
A. S., and Cath, T. Y. (2019). Data-driven and Sambuu, U. (2021). Applications of
performance analyses of wastewater treat- machine learning algorithms for biological
ment plants: A review. Water research, 157, wastewater treatment: updates and perspec-
498–513. tives. Clean Technologies and Environmen-
[13] Ning, S., and Hong, S. (2022). Programma- tal Policy, 23, 127–143.
ble logic controller-based automatic control [19] Wang, D., Thunéll, S., Lindberg, U., Jiang,
for municipal wastewater treatment plant L., Trygg, J., Tysklind, M., and Souihi, N.
optimization. Water Practice and Technol- (2021). A machine learning framework to
ogy, 17(1), 378–384. improve effluent quality control in waste-
[14] Nourani, V., Asghari, P., and Sharghi, E. water treatment plants. Science of the total
(2021). Artificial intelligence-based ensem- environment, 784, 147138.
ble modeling of wastewater treatment plant [20] Wodecka, B., Drewnowski, J., Białek, A.,
using jittered data. Journal of Cleaner Pro- Łazuka, E., and Szulżyk-Cieplak, J. (2022).
duction, 291, 125772. Prediction of wastewater quality at a waste-
[15] Salles, R., Mendes, J., Araújo, R., Melo, C., water treatment plant inlet using a system
and Moura, P. (2022, July). Prediction of based on machine learning methods. Pro-
key variables in wastewater treatment plants cesses, 10(1), 85.
2 Green SaaS adoption in cloud
computing: An investigation into
environment sustainability
Bhanu Priyaa, Bhupal Aryab, K. M. Pandeyc, and Prashant Vermad
Department of Computer Science and Engineering, Roorkee Institute of Technology, Roorkee, India
Abstract: The advent of cloud computing has advanced tremendously in the realm of information technology,
resulting in a transformative influence on the IT sector. Cloud computing provides the provision and delivery of
services, software, and computing over the Internet. Due to the utilization of resources such as the virtualiza-
tion of servers, networks, and storage, as well as workload distribution through load-balancing processes, it
consumes a lot of power and energy. In this paper, we have defined the “Green IT component” which refers to
the utilization of energy-saving and environmentally friendly techniques within data centers and their compo-
nents as processing, storage, and network elements. This paper gives the concept of a “greener cloud” and its
implications for sustainability. We have examined various research papers discussing the challenges and oppor-
tunities associated with green cloud computing, with the application of green software-as-a-service models such
as desktop as a service, integration as a service, prognostic, and health management as a service.. Green cloud
technology ensures high performance and productivity through an energy-efficient approach, enabling busi-
nesses while also promoting a sustainable future for the planet. In future work, we will propose a new architec-
ture for a greener cloud for the utilization of different software-as-a-service models of delivery.
Keywords: Sustainability, energy efficiency, resource optimization, carbon footprint, green IT, energy-aware
design
1. Introduction until 2025, indicating a rising demand for
data center deployment in the country in the
The rapid growth of digital services has
foreseeable future. It was predicted that the
resulted in an extensive surge in worldwide
volume of the data center trade would dou-
internet traffic, growing 20 times since 2010
ble from 400 megawatts to 1008 megawatts
and accompanied by a doubling of internet
users. As a result, there is a pressing necessity by 2023. Hence the operation of servers in
to prioritize energy efficiency measures to alle- data centers requires a significant amount of
viate the considerable power consumption of energy, resulting in a corresponding demand
data centers. Google and Microsoft have set for power for resource components such
targets for 2030 to procure and utilize zero- as CPU, hard disc, and memory elements.
carbon electricity around the clock in every Moreover, this energy consumption generates
grid where their operations are present. A ris- excess heat that must be efficiently cooled
ing number of organizations are also striving and managed.
to achieve continuous access to carbon-free According to the International Energy
energy 24/7. The data center market in India Agency CO2 emission report 2022, the largest
is projected to experience a steady growth sectorial rise in emissions during 2022 was
rate of 5% compound annual growth rate) observed in the electricity and heat generation
[email protected]; [email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-2
14 Innovations in Computational Intelligence: Trends and Technologies
sector, with a 1.8% increase or 261 Mt. application businesses can eliminate the need
Notably cloud infrastructure is designed to for on-premises hardware and infrastructure,
minimize expenses and maximize resource resulting in lower energy consumption and
utilization, making it more environmentally reduced carbon footprint. PaaS platforms
friendly compared to traditional data centers frequently provide frameworks and tools
(IEA, 2022). In terms of service delivery mod- that encourage effective coding techniques,
els, cloud computing offers infrastructure as optimal resource utilization, and scalability,
a service (IaaS), platform as a service (PaaS), which results in energy savings and decreased
and software as a service (SaaS). Through the waste.
internet, virtualized computer resources are All spheres of life are starting to adopt
made available by IaaS, giving users control cloud computing technologies, particularly in
and management over their infrastructure. the area of data storage. Increasingly, Con-
PaaS provides a framework for creating, sumers are okay with endpoint devices having
testing, and deploying applications with- less storage since they prefer using the cloud.
out requiring the management of the foun- While this has been happening, some com-
dational infrastructure. SaaS eliminates the mercial cloud development firms, like Google,
requirement for local installation and main- Amazon, and IBM Cloud, have offered a vari-
tenance by delivering software programs via ety of services, including infrastructure and
the internet (Gulati, no date). Green cloud services for scientific computing.
solutions offer a means to reduce operational
expenses, minimize greenhouse gas emissions, 2.2. Green IT
and conserve energy resources, thereby fos-
tering a positive impact on the environment, The Environmental Protection Agency initi-
all while ensuring the maintenance of high- ated the Energy Star program in the United
quality service levels (QoS) (Raghavendran et States back in 1992. The program specifi-
al., 2017). cally focuses on enhancing energy efficiency
in computer devices and related technologies.
The term “green computing” gained popular-
2. Background ity subsequent to the inception of the Energy
Star program (Brown, Webber and Koomey,
2.1. Cloud computing 2002)a voluntary labeling program designed
A fundamental transformation in the utili- to identify and promote energy-efficient prod-
zation and management of hardware and ucts, in order to reduce carbon dioxide emis-
software resources are used and managed is sions. Since then, the EPA, now in partnership
brought about by cloud computing. Robust with the US Department of Energy (DOE.
data replication and failover mechanisms Around the same period, TCO Devel-
ensure high reliability and mitigate the risks opment, an organization based in Sweden,
of data loss and downtime. In the context introduced the TCO certification program
of the energy-efficient cloud model, IaaS (Joumaa and Kadry, 2012). This initiative
allows businesses to optimize their resource was aimed at promoting and certifying a
utilization such as provisioning, mapping, wide range of IT equipment include net-
adaptation, discovery, brokering, estimation, work accessories, computer systems, video
and modeling. It emphasizes the benefits of adapters, monitors, keyboards, and even
resource management in IaaS, including scal- mobile phones with a primary focus on ful-
ability, quality of service, cost-effectiveness, filling distinct environmental and sustainabil-
and simplified interfaces (Mary, 2013). SaaS ity standards. The use of energy-saving and
refers to cloud-based software applications environmentally friendly techniques in data
that are accessed through the internet. In SaaS centers, with a focus on the CPU, memory,
Green SaaS adoption in cloud computing: An investigation into environment sustainability 15
computing. The middle layer, which includes
“cloud service” and “green IT,” likely repre-
sents the components and technologies that
enable environmentally sustainable cloud
services.
3.1. Virtual machine management
When we talk about virtual machine (VM)
AQ1: management, to handle cloud services,
Please
provide VM migration is crucial. In the practice of
source VM migration (Jain and Inderveer, 2016),
for all the Figure 2.1. Green IT components.
figures
both live and non-live patterns are utilized
Source: to transfer complete virtual servers across
Author's own resource
physical machines within a data center.
Through the migration process, our objec-
storage, and network components, is referred tive is to balance the VM load, enhance power
to as a “green IT component,” as shown in efficiency, and streamline system maintenance
Figure 2.1. and fault tolerance capability.
Green IT in cloud computers uses tech-
nology like solid-state drives, which use
less power than conventional spinning hard 3.2. Workload consolidation
disc drives, to increase storage efficiency. By With the growing demand for computational
removing superfluous data and optimizing power, to achieving energy and power effi-
data location, reduplication, compression, ciency involves consolidating active VMs to
and data tiring techniques further mini- reduce the server and transitioning dormant
mize storage energy use. green IT focuses servers into sleep or shut down mode consid-
on power-saving networking hardware and ering lower power states.VM consolidation
protocols that preserve high-speed and low- methods can be categorized into two vari-
latency connections while using less energy. ants, namely static and dynamic, based on
3. Greener cloud: challenges and
opportunities
The term “greener cloud” encompasses the
environmentally friendly nature of cloud com-
puting and its efficient service delivery. Green
cloud serves as a strategic plan to reduce
operational expenses by leveraging virtual-
ization, multi-tenancy for resource optimiza-
tion, load balancing, and maintaining quality
of service (QoS) and manageability.
In the context of cloud, “green cloud
placement” mentions to the strategic place-
ment of cloud services to optimize energy effi-
ciency and minimize environmental impact. Figure 2.2. Greener cloud placement in cloud
Figure 2.1. depicts a visual representation of computing technology.
the different layers involved in green cloud Source: Author's own resource
16 Innovations in Computational Intelligence: Trends and Technologies
the initial condition of the data centers they resources, storage, network usage, data trans-
start with. These techniques aim to minimize fer, load balancing, and additional services.
relocation overhead and maximize consolida- By considering factors such as instance usage,
tion by incorporating it into their modeling data storage, network traffic, and support
process (Arshad et al., 2022). options, the calculator provides an estimation
of cloud service costs.
3.3. Efficient task scheduling By adopting these practices, cloud service
providers (CSPs) contribute to a more sus-
In cloud data centers, an effective task sched- tainable IT industry and offer environmen-
uler is responsible for efficiently utilizing the tally responsible services to their customers.
resources available in the data center for Greener cloud projects offer businesses a
task execution. The efficiency of a scheduling variety of chances to cut costs, support the
algorithm is typically evaluated based on two environment, innovate, and stand out from
key factors: first is makespan which measures the competition while promoting a greener
how long it takes to finish every work on the future as:
cloud, and second is energy consumption,
which concentrates on using energy as effi- 3.6. Environment sustainability
ciently as possible when carrying out activi-
ties (Zambuk et al., 2021)such as security, the In order to reduce the negative effects on the
efficient allocation of resources, which in turn environment, cloud computing sustainabil-
results in the waste of resources. Researchers ity, especially in SaaS offerings, entails uti-
have explored a number of approaches over lizing resource-efficient and environmentally
the past decade to overcome these challenges. friendly practices such as server virtualiza-
The main objective of this research is to tion, resource sharing, and dynamic scaling
explore the task scheduling of cloud comput- also implementing a system to monitor and
ing using multi-objective hybrid Ant Colony report on energy consumption.
Optimization (ACO).
3.7. Innovation technological
3.4. Power and energy calculator advancements
Calculating carbon emissions in data centers The pursuit of energy-efficient cloud com-
involves considering various factors such as puting drives innovation in the development
AQ2:
Please
PUE as it quantifies the ratio of the overall of new technologies, software, and hardware
define energy consumed by a data center compo- solutions. This fosters technological advance-
PUE. nent to the energy consumed exclusively by ments, such as improved virtualization tech-
the IT infrastructure within that facility. A niques, energy-efficient data center designs,
PUE: PUE value of 1.0 represents ideal efficiency, and intelligent workload management systems.
Power indicating that total power utilization by the
Usage data center components is utilized by the IT 3.8. Government incentives and
Effectiveness
hardware and infrastructure. regulations
Governments worldwide are recognizing the
3.5. Cost calculator importance of energy efficiency and environ-
A data center cost calculator estimates mental sustainability. They may offer incen-
expenses including infrastructure, energy, tives, grants, or tax benefits to organizations
maintenance, cooling, personnel, and miscel- adopting green cloud practices. Compliance
laneous costs. Cloud services cost calculator with green regulations can also open doors to
involves estimating expenses for compute new markets and partnerships.
Green SaaS adoption in cloud computing: An investigation into environment sustainability 17
3.9. Collaboration and knowledge field, present patterns, and upcoming research
sharing issues. Katal (Katal, Dahiya and Choudhury,
2022) explain energy efficiency in containers
Green cloud initiatives promote collabora- and present a solution to reduce the power
tion among industry stakeholders, including consumption in data centers. Instead of
CSPs, researchers, and policymakers. Sharing solely showcasing new possibilities in green
best practices, research findings, and experi- cloud computing, this paper emphasizes
ences can accelerate the adoption of sustain- the importance of long-term technological
able cloud technologies and foster collective advancement and directs attention toward
efforts toward a greener future. new technologies applicable at the operating
system, software and virtualization and also
4. Study of the state-of-the-art application levels.
CSPs offer various IT infrastructure and
techniques and research services over the internet, governed by service
Zambuk (Zambuk et al., 2021)such as secu- level agreements (SLAs) between CSPs and
rity, the efficient allocation of resources, customers/clients. Gebreslassie et. al (Gebre-
which in turn results in the waste of resources. slassie and Sharma, no date) aims to address
Researchers have explored a number of metering and billing system as client-side for
approaches over the past decade to overcome cloud services, empowering customers to
these challenges. The main objective of this verify measured and billed services, including
research is to explore the task scheduling of quality attributes like greenness.
cloud computing using multi-objective hybrid Park (Park, Han and Lee, 2023) investi-
Ant Colony Optimization (ACO explores gates the impact of cloud computing on users’
proposes a multi-objective hybrid approach energy efficiency using a novel industry-level
called ant colony optimization with bacterial measure. Findings show that IT cloud ser-
foraging (ACOBF). ACOBF achieves this by vices give the efficient result on energy con-
classifying requests based on protocol sen- sumption in the cloud, particularly after
sitivity, scheduling tasks considering dead- 2006. Chaurasia (Chaurasia et al., 2023)
lines, and creating a VM cluster to minimize proposes strategies for achieving green cloud
energy consumption. This research contrib- computing through efficient replica manage-
utes to improving the efficiency of internet ment, virtualization, resource scheduling, and
services by enhancing cloud computing task power scaling. It explores the utilization of
scheduling. sustainable resources, such as thermal-heat
Stoyanova (Stoyanova et al., 2020) explain recovery and free cooling systems, to reduce
in the context of IoT-based investigations, the environmental impact of high energy
consumers’ privacy rights and forensics suc- consumption.
cess must be balanced. It investigates the IoT- Public health issues and epidemics, such
based cloud security challenges in legal and as obesity, diabetes, cardiovascular illnesses,
private terms. The paper offers a safeguard cancer, osteoporosis, and dental diseases,
of privacy during data extraction and ensures pose significant challenges to human well-
the integrity of evidence through dispersed being. Priya (Priya et al., 2020) provide a
blockchain-based solutions and also explores promising approach for intelligent disease
the forensics-as-a-service to reduce the data diagnosis by machine learning and artificial
and use of forensics intelligence techniques. intelligence. The proposed paper introduces a AQ3:
Please
Patil (Patil and Patil, 2019) presents a thor- novel hybrid algorithm and prediction model define
ough examination of green cloud computing, that combines the features of KNN and CNN KNN and
covering its attributes, prior successes in the algorithms. CNN.
KNN: K-Nearest Neighbors
CNN: Convolutional Neural Network
18 Innovations in Computational Intelligence: Trends and Technologies
Azambuja (Azambuja et al., 2013) provides valuable insights to researchers
addressed the issue of inefficient construction working on energy consumption, highlight-
management, where delayed feedback leads ing the efficiency of the GreenCloud simu-
to resource wastage. They proposed a solu- lator compared to other simulators’ system
tion by utilizing cloud-based tools to partici- (Atiewi and Yussof, 2014).
pate with suppliers and the genuine demand Wang (Wang, 2021) analyse the fusion
at the construction site, thereby reducing the technology as the system’s subject, object,
accumulation of large inventories and mini- and means of evaluating green investments.
mizing material wastage on-site. The study demonstrates the significant impact
In greener cloud methodology the author of cloud computing and information fusion
evaluates three popular simulators: Cloud- technology on the green investment evalu-
Sim, CloudAnalyst, and GreenCloud, which ation system. Ahmad et. al (Ahmad et al.,
all measure energy consumption based on 2015) explain the use of virtualization and
their respective capabilities. This survey VM migration in modern cloud data centers
Table 2.1. State-of-the-art of green cloud
Author/Year Approach Technique Performance metric
AQ4: Zambuk et. al ACOBF (ant colony Classification of Convergence, solution
Please
(Zambuk et al.,
provide optimization with protocol sensitivity, diversity
2021)
citation bacterial foraging) scheduling, vm cluster
for Table
Stoyanova et. al
2.1. and Privacy-preserving data Theoretical past and Data reduction and
source(Stoyanova et al., extraction, current models in digital forensics intelligence
info 2020) forensics
Patil et. al (Patil Green cloud computing Green cloud computing, Developmental stage and
and Patil, 2019) and its characteristics cutting-edge solutions implementation barriers
Katal et. al (Katal, Problem-solving Dynamic, power Clear measures at each
Dahiya and approaches to reduce modeling at OS level level for efficient power
Choudhury, 2022) power consumption and application level
Gebreslassie
AUTHOR'S et. al Introducing a client-side
SELF GENERATED DPS-Yemane (simulated Performance, service
(Gebreslassie and metering and billing model) for client-side outage, updates,
Sharma, no date) system SLAs violation
Park et. al (Park, Environmental impact of Novel industry-level Estimated energy cost
Han and Lee, cloud computing measure savings for users in the
2023) us economy
Chaurasia et. al Emphasized for network Greedy-based Thermal-heat recovery
(Chaurasia et al., traffic management scheduling, load and free cooling systems
2023) balancing
Priya et. al(Priya et Machine learning and Leveraging cloud Uncovering hidden
al., 2020) artificial intelligence storage for high-speed patterns in extensive
prediction analysis medical data
Azambuja et. al Utilizing cloud Cloud-based technology Material wastage on-
(Azambuja et al., computing in the integration of suppliers site, accumulation of
2013) construction industry and demand large inventories
Atiewi et. Empirical qualitative, CloudSim, cloud analyst, Analysis of green
al(Atiewi and quantitative analysis, and Greencloud attributes
Yussof, 2014) data integration,
Source:
Green SaaS adoption in cloud computing: An investigation into environment sustainability 19
to efficiently manage resources and reduce
computational cost and energy consumption.
Yan (Yan et al., 2022) represents a detailed
design and implementation of a deep rein-
forcement learning approach to address the
real-time job allocation problem in cloud
environments. The primary objective is to
optimize the consumption of energy while
ensuring a high QoS for users.
5. Green cloud software-as-a- Figure 2.3. Green cloud service model of
software as a service.
service: A provider’s perspective
Source:
AUTHOR'S SELF GENERATED
Cloud computing refers to the storage,
administration, and data processing carried DAAS helps to unravel many worries such
out using an internet-connected network of as data security, low computational cost, and
distant computers as an alternative to a local efficient use of storage and technology, and
server or personal computer. The green cloud to build a green environment (Charul Jag-
software-as-a-service (SaaS) model which is tap et al., 2020)we have integrated Cloud
shown in Figure 2.2 illustrates environmen- Computing with Desktop Virtualization. In
tally sustainable practices and energy-efficient our research and project, we have deployed
strategies within cloud computing. a cloud using openstack along with Desk-
top Virtualization. This paper aims for the
5.1. Information as a service security and cost reduction for institutional
As a service, information (IAS) is a cloud ser- purpose. I.INTRODUCTION As IT sectors
vice model that provides data to its users. The and Institutions are expanding rapidly day
information is given in an enterprise-friendly by day, need for computer systems are also
and user-friendly style. IAS Green technol- increased. new machines are being purchased.
ogy focuses on giving perspectives based on The IT companies can easily afford buying
analyzed and processed data, as well as infor- new machines as per their requirements. But
mation based on knowledge or a trustworthy what about the institutions and educational
source (Taryana, Fajar and Utama, 2018). sectors? Hence an effective way should be
Validation and search of zip codes and determined for the usage. N-Computing i.e.
addresses, payment processing, and data vali- Desktop Virtualization with Cloud provides
dation and completion services are all exam- us with the solution. It helps to resolve many
ples of Information as a Service. concerns such as Cost reduction, Data Security
and Efficient use of storage and technology.
Cloud Computing helps to provide maxi-
5.2. Desktop as a service mum storage with low computational cost.
Desktop as a service (DAAS) providers man- Hence the Institutions find this method more
age tightly closed desktop services that help affordable and promising. which reduces the
businesses to set up a whole far-off computer total hardware with software.Enabling tech-
surroundings for their employees. Using cloud nology of cloud computing, virtualization
desktop virtualization, we are providing cus- provides a realistic solution for resource con-
tomized software from any device in secure solidation and simplifying the management.
storage with flexible time and also providing On the other hand it helps an organization
anywhere in the world. to build a green computing environment with
20 Innovations in Computational Intelligence: Trends and Technologies
the increasing popularity of cloud computing real-time guarding of data, and restoration
with virtualization technology. By using soft- of vulnerabilities.
ware as a service(SAAS.
5.6. Security as a service
5.3. Database as a service SaaS model of cloud computing faces many
Database as a service (DBaaS) model is a troubles when we build the cloud infrastruc-
cloud computing managed service that allows ture because cloud providers generally are
users to line up, operate, manage, and scale not concerned about the security and pri-
with some style of access to information while vacy of the tenant or customers. Hence the
not the necessity to set up the physical hard- tenant uses specific protection gear together
ware, putting in code, or configuring it for with anti-virus and host-based intrusion
performance and heading for by themselves detection structures to stabilize their digital
(Al Shehri, 2013). Green DBaaS in enterprise machines additionally providers afford flex-
manager is applied all the way through 4 ibility to tenants to determine how a great
options which are: VM based, joint cluster, deal manage they want to have over their
collective installation, and shared database. very own VM.
payments are per-usage and software propri- SaaS provider model based on the base-
etors can get entry to their utility statistics line security framework (Varadharajan and
according to their need. Tupakula, 2014) to secure its own infra-
structure and make sure that malicious ten-
ants aren’t attacking the cloud infrastructure
5.4. Integration as a service or maybe the web website hosting malicious
Integration as a service (INaaS) is a machine software.
of equipment and applied sciences that con-
nects a number of applications, systems, 5.7. Prognostics and health
repositories, and IT environments for the management as a service
real-time trade of facts and processes. Many
industries prefer the SAAS for most consum- Prognostics and Health Management (PHM)
ers because they do not want more machines. service of cloud computing represents a great
So the center of attention is to invoke the chance to notice coming failures by predicting
functionalities and develop the legacy appli- future behavior. There are many challenges to
cation in heterogeneous cloud services such predicting the integrating behavior such as
as web service (Kumar and Bari, 2017). complex data of industries, large volumes of
data stored in different locations, high com-
putational resources, and continuous moni-
5.5. Compliance as a service toring of logistics and infrastructure. This
Compliance as a service (CaaS) is a cloud management service provides PHM solutions
SLA whose principal intent is to regulate and to have the ability of fault detection, fault iso-
the guard consumers’ privatizes and furnish lation, and estimation of remaining useful life
security by imposing greener attributes such (Terrissa et al., 2016).
as confidentiality, integrity, availability, and Table 2.2 Provides a list of critical consid-
accountability (CIAA) (Yimam and Fernan- erations for developing and overseeing envi-
dez, 2016). ronmentally sustainable SaaS and provides
CaaS offers partitioning of duties, annual some comparison and recognition of the ser-
risk evaluation, management of differ- vice’s environmental impact, especially with
ent applications, facts and figures detec- regard to carbon emissions and the ecosys-
tion, policy creation and enforcement, tem’s overall footprint.
Green SaaS adoption in cloud computing: An investigation into environment sustainability 21
Table 2.2. Green cloud SAAS performance metric
Green SaaS Green Software as a service affecting factors
Service Energy efficiency Optimize resource Carbon footprint Environmental impact
IAS High Variable Moderate Depends on data usage
DAAS High High Low Reduced e-waste
DBAAS High High Low Reduced hardware
INAAS High High Low Reduced resource usage
CAAS High High Low Streamline processes
PHM High High Low Efficiently maintained
Source:
Author'sOwn Generated
6. Conclusion and future directions the environmental impact of SaaS solutions.
This paper concludes that the green cloud
Green cloud computing has emerged as an SaaS model aligns with the principles of sus-
important area of research and development tainability by promoting energy efficiency,
in recent years. Green data centers provide a resource optimization, and carbon footprint
wide array of advantages, such as decreased reduction in services.
ongoing operational expenses, reduced space
demands, lower carbon footprints, mini-
mized water usage, decreased waste gen- Acknowledgment
eration, diminished electricity consumption,
and a strong commitment to sustainable and This paper and the research behind it would
renewable data center methodologies. Vari- not have been possible without the excep-
ous strategies and techniques have been pro- tional support of my colleague, Bhupal Arya.
posed and implemented to achieve greener I am also grateful for the insightful comments
clouds, including virtualization, workload offered by the anonymous peer reviewers at
consolidation and task scheduling. this conference. The generosity and exper-
By optimizing resource utilization and tise of one and all have improved this study
improving energy efficiency, green IT can in innumerable ways and saved me from
measure and reduce the carbon footprint many errors; those that inevitably remain are
of data centers and contribute to the over- entirely my own responsibility.
all sustainability of the IT industry as it can
help lower operational costs for businesses References
and enable more scalable and flexible cloud AQ5: As
services. This paper discusses the various [1] Ahmad RW, Gani A, Hamid SHA, Shiraz M, per style,
references
parameters of green cloud placement in cloud Xia F, Han J. A survey on virtual machine and its
services and surveyed various technique and migration and server consolidation frame- citations
works for cloud data centers. J Netw Com- should be
measure their performance metric used in num-
put Appl. 2015;52:11-25. doi:10.1016/j.
cloud environment. jnca.2015.02.002
bered.
Please
While significant progress has been made [2] Arshad U, Zia MA, Siddique K, et al. Utiliz- check and
in green cloud computing, there are still sev- ing power consumption and SLA violations confirm
eral avenues for future research and develop- using dynamic VM consolidation in cloud
for
clarity.
ment such as developing energy-aware SaaS data centers. Renewable Sustainable Energy
architectures and design patterns, and devel- Rev. 2022;167:112782. doi:10.1016/j.
oping regular metrics and tools to evaluate rser.2022.112782
CHECKED AND CONFIRMED
22 Innovations in Computational Intelligence: Trends and Technologies
[3] Atiewi S, Yussof S. Comparison between Cluster Comput. Published online 2022.
cloud sim and green cloud in measuring doi:10.1007/s10586-022-03713-0
energy consumption in a cloud environ- [14] Kumar M, Bari P. Integration of
ment. In: Proceedings - 3rd International cloud based services. Published online
Conference on Advanced Computer Sci- 2017;2(4):101-105.
ence Applications and Technologies, ACSAT [15] Mary MA. Survey on resource management
2014. IEEE; 2014:9-14. doi:10.1109/ technique in cloud computing. Published
ACSAT.2014.9 online 2013;2(12):1723-1731.
[4] Azambuja M, Camargo M, Filho LM, et al. [16] Park J, Han K, Lee B. Green cloud? An empir-
Enabling lean supply with a cloud comput- ical analysis of cloud computing and energy
ing platform - An exploratory case study. In: efficiency. Manag Sci. 2023;69(3):1639-
21st Annual Conference of the International 1664. doi:10.1287/mnsc.2022.4442
Group for Lean Construction 2013, IGLC [17] Patil A, Patil DR. An analysis report on
2013. 2013:815-824. green cloud computing current trends and
[5] Brown R, Webber C, Koomey JG. Status future research challenges. SSRN Electron
and future directions of the ENERGY STAR J. Published online 2019. doi:10.2139/
program. Energy. 2002;27(5):505-520. ssrn.3355151
doi:10.1016/S0360-5442(02)00004-X [18] Priya L, Nallakannu G, Abirami C, Kumar
[6] Jagtap C, Rathore K, Patel P, et al. Desk- SV, Pavithra A. A novel intelligent diagno-
top virtualization with cloud computing. sis and disease prediction algorithm in green
Int J Eng Res Technol. 2020;9(1):255-257. cloud using machine learning approach. J
doi:10.17577/ijertv9is010157 Green Eng. 2020;10(7):3421-3433.
[7] Chaurasia N, Chauhan R, Agarwal H, [19] Raghavendran CV, Guhan P, Balasubrama-
Kumar N, Jaiswal RK. Shifting from cloud nian P, Sathyakumar M. A study on cloud
computing to green cloud and edge comput- computing services. Int J Eng Res Technol.
ing. SSRN Electron J. Published online 2023. 2017;4(34):1-7.
doi:10.2139/ssrn.4387781 [20] Al Shehri W. Cloud database database as a ser-
[8] Gebreslassie Y, Sharma DP. DPS-Yemane- vice. Int J Database Manag Syst. 2013;5(2):1-
Shareme CSMM model for client-side SLA 12. doi:10.5121/ijdms.2013.5201
of green cloud services measuring and [21] Stoyanova M, Nikoloudakis Y, Panagiotakis
monitoring. S, Pallis E, Markakis EK. A survey on the
[9] Gulati R. A systematic review on the various internet of things (IoT) forensics: challenges,
approaches used for achieving energy con- approaches, and open issues. IEEE Com-
sumption in cloud. Published online. mun Surv Tutor. 2020;22(2):1191-1221.
[10] IEA. CO2 emissions in 2022. Glob doi:10.1109/COMST.2019.2962586
Energy. 2022;62(10):20-21. Available at: [22] Taryana U, Fajar AN, Utama DN. Informa-
https://www.iea.org/news/global-co2-emis- tion as a service on cloud computing tech-
sions-rebounded-to-their-highest-level-in- nology: a review. In: 2018 International
history-in-2021. Seminar on Research of Information Tech-
[11] Jain N, Inderveer K. Energy-aware virtual nology and Intelligent Systems, ISRITI
machine migration for cloud computing - a 2018. IEEE; 2018:39-42. doi:10.1109/
firefly optimization approach. J Grid Com- ISRITI.2018.8864468
put. Published online 2016. doi:10.1007/ [23] Terrissa LS, Sahnoun M, Zghal M, et
s10723-016-9364-0 al. A new approach of PHM as a ser-
[12] Joumaa C, Kadry S. Green IT: Case studies. vice in cloud computing. In: Colloquium
Energy Procedia. 2012;16(PART B):1052- in Information Science and Technology,
1058. doi:10.1016/j.egypro.2012.01.168 CIST. IEEE; 2016:610-614. doi:10.1109/
[13] Katal A, Dahiya S, Choudhury T. Energy CIST.2016.7804958
efficiency in cloud computing data cent- [24] Varadharajan V, Tupakula U. Security as a
ers: a survey on software technologies. service model for cloud environment. IEEE
Green SaaS adoption in cloud computing: An investigation into environment sustainability 23
Trans Netw Serv Manag. 2014;11(1):60-75. [27] Yimam D, Fernandez EB. A survey of com-
doi:10.1109/TNSM.2014.041614.120394 pliance issues in cloud computing. J Inter-
[25] Wang P. Application of cloud computing net Serv Appl. 2016;7(1). doi:10.1186/
and information fusion technology in green s13174-016-0046-8
investment evaluation system. J Sensors. [28] Zambuk FU, Anwar A, Zangeneh F, et al.
2021. doi:10.1155/2021/2292267 Efficient task scheduling in cloud comput-
[26] Yan J, Zhang S, Yu W, Yuan Y. Energy-aware ing using multi-objective hybrid ant colony
systems for real-time job scheduling in cloud optimization algorithm for energy efficiency.
data centers: a deep reinforcement learning Int J Adv Comput Sci Appl. 2021;12(3):450-
approach. Comput Electr Eng. 2022;99:107688. 456. doi:10.14569/IJACSA.2021.0120353
doi:10.1016/j.compeleceng.2022.107688
3 Anomaly detection in wireless sensor
network
Mayank Shuklaa, Sneha Yadavb, Abhay Pratap Singhc, Fizza Rizvid,
and Surya Vikram Singhe
Department Computer Science and Engineering, Babu Banarasi Das Institute of Technology and
Management, Lucknow, Uttar Pradesh, India
Abstract: With the recent emergence of the notion of the “Internet of Things,” wireless sensor networks (WSNs),
or WLANs, are critical and indispensable platforms for the future. They are employed in the military, health-
care, industry, and habitat for the purposes of tracking, monitoring, and regulating numerous applications.
However, anomalies that arise for a variety of causes, including node failures, reading mistakes, odd events, and
malicious attacks, have an impact on the quality of data that sensor nodes collect. Thus, before sensor data is
used to inform choices, anomaly detection is a crucial step in ensuring its quality.
We outline the difficulties in detecting anomalies in WSNs and the conditions that must be met in order to
create models that are both successful and efficient. Next, we examine the most recent developments in data
anomaly detection research in WSNs and group existing detection techniques into five major classes according
to the detection techniques used to create these techniques. Variations of the most advanced models in each
class are discussed, and their drawbacks are emphasized to give insight into possible future research directions.
In addition, the methods under assessment are contrasted and assessed according to how well they fulfill the
specified criteria. In conclusion, the overall constraints of existing methodologies are referenced and potential
avenues for future research are proposed and deliberated.
Keywords: IoT, anomaly detection, wireless sensor network, machine learning, neural network
1. Introduction reliability of WSNs. It empowers these net-
works to identify unusual patterns, irregulari-
Wireless sensor networks (WSNs) have ties, and deviations from expected behavior,
emerged as a cornerstone of modern infor- which can stem from various sources – envi-
mation and communication technologies, ronmental changes, equipment malfunctions,
enabling the collection of data from remote or malicious intrusions.
and often challenging environments. From
monitoring environmental conditions and
industrial processes to facilitating health-
2. Methodology
care and smart city applications, WSNs have
become pervasive. However, their ubiquity
2.1. Define research objectives
also exposes them to a wide range of vulner- The computer science project aims to address
abilities and threats. This is where the critical the problem of anomaly detection in WSNs
field of Anomaly Detection in WSNs comes by designing and implementing efficient algo-
into play. rithms that can accurately identify abnor-
Anomaly detection is a fundamental com- mal events while considering the limitations
ponent of ensuring the integrity, security, and of resource-constrained sensor nodes. The
a
[email protected]; [email protected]; [email protected];dRizvif952@
gmail.com; [email protected]
DOI: 10.1201/9781003606635-3
Anomaly detection in wireless sensor network 25
specific objectives and approaches in this pro- These applications may include early detec-
ject include: tion of environmental incidents, intrusion
detection in security applications, equipment
2.2. Algorithm selection and fault monitoring in industrial settings, and
more.
development
The project involves a comprehensive review
of existing anomaly detection algorithms, 3. Thematic Overview
both from the domains of machine learn- Anomaly detection in WSNs is a project topic
ing and statistical analysis. After evaluating that involves the identification of unusual
these methods, the project aims to develop a patterns or behaviors within the data col-
customized anomaly detection algorithm tai- lected by sensors in a wireless network. The
lored to the unique characteristics of WSNs. overarching goal is to develop techniques and
algorithms that can effectively distinguish
2.3. Resource efficiency between normal and anomalous activities,
helping to enhance the overall security, reli-
Given the resource constraints of sensor ability, and efficiency of the wireless sensor
nodes, the project focuses on creating an algo- network.
rithm that minimizes the computational and Here’s a thematic overview of such a
memory requirements. This ensures that the system:
anomaly detection process can be executed on
sensor nodes with limited resources without
compromising the network’s overall efficiency. 3.1. Sensor data characteristics
Analyzing the characteristics of sensor data,
2.4. Real-time detection such as signal strength, frequency, and tem-
poral patterns, to identify anomalies that
An important aspect of the project is to deviate from expected norms.
achieve real-time anomaly detection, allow-
ing the network to respond promptly to
abnormal events. This involves optimizing 3.2. Machine learning models
the algorithm for quick decision-making and Utilizing machine learning algorithms to
minimizing communication overhead, as well detect anomalies in user behavior based on
as energy consumption. historical data. This may involve supervised,
unsupervised, or semi-supervised learning
2.5. Evaluation and validation approaches.
The project involves extensive experimen-
tation using both synthetic and real-world 3.3. Pattern recognition
datasets. It aims to assess the performance Identifying patterns of normal user behavior
of the developed algorithm in terms of accu- and flagging deviations from these patterns as
racy, false-positive rate, detection time, and potential anomalies. This can involve statisti-
resource utilization. The algorithm will be cal methods or advanced pattern recognition
compared against existing methods to dem- techniques.
onstrate its advantages.
3.4. User profiling
2.6. Practical applications Creating user profiles based on their typical
The research project also explores practical interactions with the wireless sensor network.
applications of anomaly detection in WSNs. Anomalies are then detected when a user’s
26 Innovations in Computational Intelligence: Trends and Technologies
behavior deviates significantly from their 3.11. Integration with access control
established profile. systems
Integrating anomaly detection with access
3.5. Context-aware anomaly
control mechanisms to dynamically adjust
detection user privileges based on detected anomalies,
Considering contextual information, such as enhancing network security.
the physical location of users, time of day,
or specific network conditions, enhances the 3.12. User feedback and interaction
accuracy of anomaly detection and reduces
false positives. Incorporating user feedback and interaction
data to improve the accuracy of anomaly
detection models and reduce false positives
3.6. Energy-efficient anomaly by considering user-specific preferences and
detection habits.
Developing anomaly detection methods that are
energy-efficient, considering the resource con- 3.13. Scalability and robustness
straints of sensor nodes in wireless networks.
Designing anomaly detection systems that
are scalable to large wireless sensor networks
3.7. Real-time detection and response and robust against various types of anomalies
Implementing real-time anomaly detec- and environmental changes.
tion mechanisms to promptly identify and
respond to abnormal user behavior, minimiz- 3.14. Data fusion techniques
ing potential security threats or disruptions.
Employing data fusion techniques to com-
bine information from multiple sensors and
3.8. Collaborative anomaly detection sources for a more comprehensive under-
Enabling collaboration between sensor nodes standing of user behavior and improved
to collectively analyze user behavior pat- anomaly detection.
terns and detect anomalies more accurately
through information sharing. 3.15. Threat intelligence integration
Integrating threat intelligence feeds to
3.9. Adaptive models
enhance anomaly detection by incorporating
Creating adaptive anomaly detection mod- information about known threats and attack
els that can continuously learn and evolve patterns.
based on changing user behavior and network
conditions.
4. Critical Analysis
3.10. Security and privacy
4.1. Accuracy and false positives
considerations
Addressing the security and privacy impli-
4.1.1. Strength
cations of user anomaly detection, ensuring Many anomaly detection algorithms can
that sensitive information is protected and achieve high accuracy in identifying unu-
that the detection process itself is resistant to sual patterns. However, achieving a balance
malicious attacks. between sensitivity and specificity is crucial
Anomaly detection in wireless sensor network 27
to minimize false positives, as these can lead between efficiency and accuracy is essential,
to unnecessary alarms and resource wastage. especially in applications where energy con-
servation is critical.
4.1.2. Weaknesses
4.4. Security considerations
Achieving high accuracy may come at the
cost of increased false positives, especially in 4.4.1. Strength
dynamic and noisy environments. Striking the
Anomaly detection is a fundamental com-
right balance is challenging, and the trade-off
ponent of security systems in wireless sensor
between sensitivity and specificity needs care-
networks. It helps identify malicious activities
ful consideration.
and potential threats.
4.2. Adaptability to dynamic 4.4.2. Weaknesses
environments
Anomaly detection itself can be vulner-
4.2.1. Strength able to evasion and adversarial attacks.
Effective anomaly detection systems should Adversaries may manipulate sensor data
adapt to dynamic changes in the wireless sen- to evade detection, highlighting the need
sor network environment. Machine learning for robust security measures and continu-
models, particularly those based on unsuper- ous improvement of anomaly detection
vised learning, can learn, and adapt to new techniques.
patterns over time.
4.5. Scalability
4.2.2. Weaknesses 4.5.1. Strength
Adapting to rapid changes can be challeng- Scalability is essential for large-scale wire-
ing for some algorithms. Sudden variations less sensor networks. Efficient algorithms and
in sensor data due to environmental changes distributed approaches can enable effective
or network conditions may lead to delayed anomaly detection in networks with numer-
or inaccurate anomaly detection. Ensuring ous sensor nodes.
real-time adaptability remains a significant
challenge. 4.5.2. Weaknesses
As the network size increases, the complex-
4.3. Resource efficiency ity of anomaly detection also grows. Ensuring
4.3.1. Strength scalability without compromising accuracy
and response time requires careful system
Energy-efficient anomaly detection is crucial
design.
for wireless sensor networks with resource-
constrained nodes. Certain lightweight algo-
rithms and edge computing approaches can 4.6. Context awareness
minimize energy consumption. 4.6.1. Strength
Context-aware anomaly detection enhances
4.3.2. Weaknesses the accuracy of identifying abnormal behav-
Some resource-efficient algorithms may sac- ior by considering environmental factors and
rifice detection accuracy. Striking a balance user context.
28 Innovations in Computational Intelligence: Trends and Technologies
4.6.2. Weaknesses 4.9.2. Weaknesses
Implementing context awareness can be com- In highly dynamic environments, distinguish-
plex and may introduce additional computa- ing between anomalous behavior and normal
tional overhead. The challenge lies in defining
relevant context parameters and integrating 4.10. Gaps, inconsistencies, and
them into the anomaly detection process controversies
effectively.
Variations (noise) remain challenging. Potential areas for future research and devel-
Robustness to noise is crucial for avoiding opment in anomaly detection for WSNs:
false positives and maintaining the reliability
of the detection system. 4.10.1. Deep learning and AI
techniques
4.7. Privacy concerns: To improve the precision and effectiveness of
4.7.1. Strength anomaly identification in WSNs, deep learn-
ing models—such as deep neural networks
Privacy-preserving anomaly detection meth- (DNNs), convolutional neural networks
ods aim to protect sensitive information while (CNNs), recurrent neural networks (RNNs),
still identifying abnormal behavior. or hybrid architectures—are included. use
of pre-trained models and transfer learning
4.7.2. Weaknesses to enhance anomaly detection performance,
Balancing privacy and accuracy is a delicate particularly in situations with a shortage of
task. Some privacy-preserving techniques labeled training data.
may result in a loss of detection efficacy, espe- Investigate the deployment of anomaly
cially when limited information is available detection algorithms directly on edge devices
for analysis. within the WSN, enabling real-time process-
ing and reducing the need for transmitting
raw sensor data to a central server. Develop
4.8. Human-computer interaction lightweight anomaly detection models opti-
4.8.1. Strength mized for edge devices with constrained com-
putational resources and energy.
Involving user feedback and interaction can
enhance the accuracy of anomaly detection
models. 4.10.2. Federated learning and
collaborative anomaly detection
4.8.2. Weaknesses Research on federated learning approaches
where local WSN nodes collaborate to collec-
Depending solely on user feedback may intro-
tively train a global anomaly detection model
duce biases and subjective judgments. Strik-
without sharing raw data.
ing a balance between automated detection
Centrally, ensuring privacy and efficiency.
and user input is crucial for a reliable system.
Study how to aggregate and utilize insights from
multiple WSNs for improved anomaly detection
4.9. Robustness to noise in large-scale, distributed IoT environments.
4.9.1. Strength
Anomaly detection algorithms often include 4.10.3. Robustness and security
mechanisms to handle noise and outliers in Investigation of adversarial attacks and tech-
sensor data. niques to improve the robustness of anomaly
Anomaly detection in wireless sensor network 29
detection models against malicious adver- integrity and confidentiality of data
saries attempting to deceive the system. transmitted across the network.
Development of anomaly detection meth-
3. Real-time responsiveness:
ods that can detect anomalies resulting from • Synthesis: Anomaly detection systems
cyber-physical attacks and security breaches that operate in real-time enable swift
in WSNs. identification and response to abnor-
mal events, minimizing the impact
4.10.4. Human-in-the-loop anomaly of disruptions and ensuring timely
detection interventions.
• Implications: In scenarios where rapid
Study the integration of human expertise and response is crucial, such as emergency
feedback into the anomaly detection process, management or critical infrastructure
leveraging human knowledge to enhance monitoring, real-time anomaly detec-
the interpretation and validation of detected tion enhances the effectiveness of the
anomalies. Design interactive and interpret- wireless sensor network in providing
able anomaly detection systems that allow actionable insights.
domain experts to understand, validate, and
refine anomaly detection results.
6. Conclusion
5. Synthesis and implication Concluding a computer science project on
“Anomaly Detection in WSN” should effec-
5.1. Patterns, trends, and tively summarize the key findings, outcomes,
relationships and the significance of your research. Here’s a
sample conclusion for such a project:
1. Improved network reliability: In conclusion, the research and implementa-
• Synthesis: In wireless sensor networks, tion of anomaly detection in WSN have yielded
anomaly detection helps to maintain several important insights and contributions.
overall network reliability by spotting This project aimed to address the critical issue
and addressing anomalous activity of detecting abnormal behavior or events in
that could jeopardize data integrity or WSNs, which is vital for ensuring the reliability
interfere with communication. and security of these networks in various appli-
• Implications: Enhanced network cations, including environmental monitoring,
reliability leads to improved trust- surveillance, and industrial automation.
worthiness in data collection and dis- Key findings and contributions of this pro-
semination, critical for applications ject are as follows:
such as environmental monitoring,
healthcare, and industrial automation. 1. Algorithm evaluation: Through an in-
2. Enhanced security: depth analysis of existing anomaly
• Synthesis: Anomaly detection acts as detection algorithms, we assessed their
a frontline defense against malicious performance in the context of WSNs. Our
activities and security breaches in project involved the evaluation of various
wireless sensor networks, helping to machine learning and statistical methods,
identify and respond to unauthorized such as Isolation Forest, One-Class SVM,
access or tampering. and K-means clustering, to identify their
• Implications: By bolstering network suitability for WSN anomaly detection.
security, anomaly detection systems 2. Customized solution: We developed a
safeguard sensitive information and customized anomaly detection algorithm
prevent potential threats, ensuring the tailored to the specific characteristics
30 Innovations in Computational Intelligence: Trends and Technologies
of WSNs. This algorithm considers the technology. Here are some potential areas for
constraints of limited resources, wire- future research and development in anomaly
less communication, and distributed data detection for WSNs:
gathering in WSNs, making it a practical
choice for real-world deployments.
7.1. Deep learning and AI techniques
3. Performance evaluation: Extensive exper-
imentation was conducted on both syn- To improve the precision and effectiveness of
thetic and real-world datasets. Our results anomaly identification in WSNs, deep learn-
demonstrated that the proposed algo- ing models—such as DNNs, CNNs, RNNs,
rithm outperformed existing methods in or hybrid architectures—are included. use
terms of accuracy, false-positive rate, and of pre-trained models and transfer learning
resource efficiency. It exhibited a high to enhance anomaly detection performance,
detection rate while minimizing energy particularly in situations with a shortage of
consumption and communication over- labeled training data.
head, which is crucial in resource-con- Investigate the deployment of anomaly
strained WSNs. detection algorithms directly on edge devices
within the WSN, enabling real-time process-
In conclusion, the successful development ing and reducing the need for transmitting
and evaluation of an efficient anomaly detec- raw sensor data to a central server. Develop
tion algorithm for Wireless Sensor Networks lightweight anomaly detection models opti-
have the potential to enhance the reliability mized for edge devices with constrained com-
and security of WSNs in various applications. putational resources and energy.
This project contributes to the ongoing efforts
to make WSNs more resilient and practical for
real-world deployments. As technology contin- References
ues to evolve, so will the challenges and solu- [1] Akyildiz I. Wireless sensor networks: A sur-
tions in the field of anomaly detection in WSNs. vey. Comput Netw. 2002;38(4):393–422.
[2] Barreca A, Curto R, Malavasi G, and
7. Future Work Rolando D. Energy retrofitting for the
modern heritage enhancement in weak real
Future work in anomaly detection for WSNs estate markets: The Olivetti housing stock
is likely to focus on addressing emerging in Ivrea. Sustainability. 2022;14(6):3507.
challenges and leveraging advancements in doi:10.3390/su14063507.
4 Revolutionizing object recognition
beyond CNN and YOLO with deep
learning breakthrough
G. S. S. S. S. V. Krishna Mohan1,a, Mahammad Firose Shaik2,b,
G. Usandra Babu3,c, R. G. V Prasanna1,d, Pinagadi Venkateswara
Rao4,e, and Inakoti Ramesh Raja5,f
1
Department of Electronics and Communication Engineering, Aditya Institute of Technology and
management, Tekkali, India
2
Department of Electronics and Instrumentation Engineering V R Siddhartha Engineering College
Vijayawada, India
3
Department of Electronics and Communication Engineering, Aditya Engineering College,
Surampalem, India
4
Department of AI and ML, School of Engineering, Malla Reddy University, Hyderabad, Telangana,
India
5
Department of Electronics and Communication Engineering Aditya College of Engineering and
Technology, Surampalem, India
Abstract: In the world of computer vision and object detection, there’s a strong connection between these areas.
Object recognition is all about recognizing specific things in pictures and videos. But this paper goes a step fur-
ther than the usual methods of finding objects. It tries to understand images in more detail, like how our eyes
do. The work starts by looking at deep learning and well- known object detection systems like CNN, R-CNN,
and YOLO. These systems can typically find only a few objects in a picture, and they work best at distance of
5–6 meters. However, our new model is much better at this task and has an interesting feature it can even tell
you the names of the objects it sees using Google Translate. This is especially helpful for people with vision
problems because it helps them understand what’s around them better. In summary, the research combines
computer vision, deep learning, and real-time object recognition to enhance visual perception and offer valuable
assistance to individuals with visual impairments.
Keywords: Deep learning, CNN, YOLO, RCNN, visual impairs, computer vision
1. Introduction over 220 million people throughout the
world would experience visual impairment,
The human eye, a vital component of our visual presenting substantial daily challenges [2].
system, is not only integral to perceiving the While the traditional white cane remains a
world but also susceptible to external factors, widely used tool for the visually impaired [3],
including potential contact with pathogens. its limitations and safety concerns propel this
Eye infections and other visual impairments work towards a clear and compelling objec-
may result from some viruses that enter the tive: Developing an affordable, yet incred-
body via the eyes [1]. By 2019, the World ibly useful device that would enable the blind
Health Organization (WHO) predicted that and visually impaired to move about their
a
[email protected]; [email protected]; [email protected]; drgvprasanna@
gmail.com; [email protected]; [email protected]
DOI: 10.1201/9781003606635-4
32 Innovations in Computational Intelligence: Trends and Technologies
environment more quickly, easily, confidently, systems that perform optimally up to 5–6
and with a distinct sense of mobility. Visual meters, our innovative model excels in rec-
impairment extends beyond mere eyesight ognizing various objects, even articulat-
issues, stemming from various causes like ing their names using Google Translate.
trouble with eyesight, cataract development, This distinctive feature proves particularly
issues in the rear side of the eye, or problems advantageous for individuals facing vision
with the optic nerve [4]. People facing visual challenges, enhancing their comprehen-
impairment encounter numerous challenges sion of the surrounding environment. By
in daily life, from navigating roads safely to integrating computer vision, deep learn-
discerning traffic light signals or avoiding ing, and real-time object recognition, our
obstacles. Pursuing education often involves work strives to advance visual perception.
learning Braille, a system of raised dots felt Ultimately, our goal is to foster inclusivity
with the fingers. Basic activities, like visiting in technology, offering valuable support to
the doctor, become challenging as individuals individuals with visual impairments for a
rely on touch and hearing to navigate their more profound understanding of the world
world, introducing an element of motion around them.
awareness.
In the contemporary landscape, artificial
intelligence (AI) stands out as a transforma- 2. Literature Review
tive force, shaping our future with its smart Cheng et al. proposed a comprehensive
capabilities [5]. Advancements in AI, fueled by approach to enhance visual localization for
abundant data and powerful computers, find individuals with visual impairments [6].
applications in diverse fields like retail and Their system comprises multiple components,
delivery services. AI facilitates efficient inven- including a deep descriptor network, 2D–3D
tory management in retail, making shopping geometric validation, and online sequence
more convenient by tracking stock and offer- comparison. In this workflow, a Dual Descrip-
ing personalized product recommendations tor network is integrated with RGB, Infrared,
based on preferences. In the delivery realm, and depth images, enabling the generation
AI plans optimal routes to transport items of robust, context-rich descriptors and local
seamlessly, introducing a dynamic element of features.
motion. Researchers at Springer-Verilog presented
AI’s transformative influence extends to a LiDAR-based technique for predicting
education, enabling personalized support by distances and effectively measuring obsta-
helping teachers understand student progress. cles [7]. They utilize an EfficientDet-LiteV4
In security, AI aids law enforcement in iden- lightweight deep learning model for obsta-
tifying suspects through image processing in cle detection and calculate anticipated dis-
surveillance videos and monitors buildings, tances using LiDAR-generated depth maps.
alerting authorities to any unusual activities. To test their approach, they implemented it
Even in healthcare, when doctors conduct on a Raspberry Pi4 platform integrated with
thorough eye examinations using a contact LiDAR. The results of their experiments dem-
fundus examination tool, they capture not onstrate high accuracy in both obstacle detec-
only static images but also dynamic insights tion and distance estimation.
into motion, aiding in the diagnosis and treat- Shen et al. proposed a two-stage aggregate
ment of various eye conditions.” learning model called VRTMM, designed for
Our research explores a smarter approach interpreting remote sensing images [8]. This
to understanding images beyond merely model combines a variational autoencoder as
spotting objects. In contrast to traditional well as reinforcement learning. Convolutional
Revolutionizing object recognition beyond CNN and YOLO with deep learning breakthrough 33
Neural Networks (CNNs) are first tuned A helpful system has been developed for
using a variational autoencoder. In the later individuals with partial vision using a Rasp-
stage, a transformer-based model that com- berry Pi 3 and optical character recognition
bines semantic and geographical data gener- (OCR) technology, as discussed in [13]. This
ates textual descriptions. system assists people with limited vision by
Rahman et al. introduced an Internet of converting text from documents into spo-
Things (IoT)-enabled system for automated ken words. However, one drawback is that
object recognition in assisting individuals the complexity increases as a computer is
with visual impairments [9]. Their system required to process the images.
facilitates the recognition of obstacles and
enable audio feedback to the user when nec-
essary. It includes four laser sensors employed 3. Technologies
for object recognition and currency denomi- The contemporary atmosphere has increased
nation identification in real-time scenarios, the risk of inhaling disorders, namely asthma,
both indoors and outdoors. To accomplish chronic obstructive pulmonary disease
these objectives, the system makes use of Ten- (COPD), and also few other disorders. That
sorFlow Lite, MobileNet, and a Single-Shot risk has increased due to increase in air pol-
Detector (SSD). lutants like PM2.5, PM10, etc., a respirator
A tool that reads text from images and can be utilized as an immediate countermeas-
turns it into spoken words is created using ure on an individual level safety measure as
a method called Optical Character Recogni- bringing down pollution levels require much
tion (OCR). In this method [10], a specific longer time than the severity of the problem
algorithm called MSER is used to make sure is allowing.
it’s accurate. The setup involves connecting a
Raspberry Pi to a camera for taking pictures,
and then the MSER algorithm is used to pull 3.1. Deep learning
out the words. On average, this system gets it The neural network in the brain is repli-
right about 85–87% of the time. cated by a specific type of computer program
An advanced audio assistance system is known as deep learning [14]. It is frequently
designed to aid blind individuals by reading referred to as ‘deep learning’ because it
letters or text from documents and identify- employs deep neural networks, which are a
ing objects in their surroundings. This sys- subset of machine learning. Connected lay-
tem, as described in [11], uses technology to ers are utilized in the construction process of
detect both text in documents and obstacles deep learning systems.
nearby. Once detected, the system produces a The surface layer is addressed as the input
voice-based audio output, providing valuable layer, while the base layer is addressed as the
information to help individuals with visual output layer. “Hidden layers” refers to the
impairments navigate their environment more intermediary phases. The term ‘deep’ refers to
effectively. A special tool has been created for neural networks that connect neurons across
people who can’t see well [12]. This tool uses more than two layers. Each hidden layer com-
smart technology to help them read books. prises neurons with interconnections between
It works with a small computer called Rasp- them.
berry Pi 2 and is programmed using Python. The input signal undergoes processing
There’s a monitor to check how the program by the neuron before being sent to the layer
is doing, and it’s especially helpful for visually above it. The weight, bias, and activation
impaired students. However, it can read only function that are applied dictate the inten-
one language. sity of the signal that is sent into the neuron
34 Innovations in Computational Intelligence: Trends and Technologies
which lies in the layer below. This very net- Regression analysis, which yields the
work consumes large volumes of input data, greatest likelihood photos, is how YOLO
processing them through several layers and formulates the object identification process.
learning more intricate characteristics of the YOLO utilizes convolutional neural net-
data at each layer. works (CNN) to swiftly diagnose entities. As
Neural networks can be grouped into the name indicates, the approach recognizes
shallow and deep types. In a shallow neu- objects using one forward propagation over
ral network, there is a solitary hidden layer a neural network. This indicates that the pre-
connecting both the input and output layers. diction procedure will be performed through-
Deep learning networks, on the other hand, out the whole image with a single algorithm
include multiple layers, such as the 22 layers run. CNN anticipates numerous class prob-
in the Google LeNet model for recognizing abilities and bounding boxes in a single go.
images. Nowadays, deep learning is utilized The YOLO algorithm comes in a wide
in various applications, ranging from TVs variety, with Tiny YOLO and YOLOv5 being
and autonomous vehicles to cell phones and a couple of the more well-known versions.
Google’s fraudulent activity tracking. The YOLO algorithm functions by employ-
ing three distinct strategies:
3.2. YOLO (You Only Look Once)
• Residual blocks
algorithm • Bounding box regression
You Only Look Once (YOLO) is a cutting- • Intersection Over Union (IOU)
edge problem-solving object identification
technique that was originally published in the 3.3. Region-based convolutional
esteemed academic publication “You Only neural network (RCNN)
Look Once: Unified, Real-Time Object identi-
fication” in 2015. The authors of this work are The multifaceted field of computer vision
Joseph Redmon, Santosh Divvala, Ross Gir- surged in popularity with the advent of
shick, and Ali Farhadi, as shown in Figure 4.1. Convolutional Neural Networks (CNNs),
This real-time technique can be employed to particularly in the context of self-driving
identify various objects in images. automobiles taking the forefront. Within
AQ1:
Please
provide
source
for all the Figure 4.1. YOLO architecture.
figures
Source:
AUTHOR'S SELF GENERATED
Revolutionizing object recognition beyond CNN and YOLO with deep learning breakthrough 35
computer vision, object detection plays a piv- Sophisticated neural networks often employ
otal role. This technique simplifies tasks such COCO sets, a kind of rigorous, better visual
as pose estimation, vehicle identification, sur- dataset. As an illustration, COCO has been
veillance, and more. used extensively to evaluate and compare real-
In contrast to classification algorithms, time object identification skills. The formats
object identification techniques use a bound- of the COCO dataset are quickly interpreted
ing box to encircle the object of interest by sophisticated neural network frameworks.
inside an image. Predicting the exact number The various attributes of COCO are
of bounding boxes in an object identification
scenario that accurately represent the numer- • Segmentation of objects with comprehen-
ous places of interest may be challenging. The sive prototype annotations
variable length of the output layer, reflect- • Contextualizing apprehension
ing the changing number of items of interest, • Super pixel stuff segmentation
makes it impractical to address this issue by • More than 200’000 images of the total
simply adding a fully connected layer on top 330’000 are labelled
of a conventional convolutional network. • 1.5 Mio object prototype
To overcome these challenges, a solution • 80 object assorts, the “COCO classes”,
involves identifying specific regions of interest which comprises “things” for which indi-
and applying a CNN algorithm to determine vidual instances may be promptly labelled
whether an object is located in the specified (human-being, automobile, any kind of
region. The initial approach of creating a furniture etc.,)
large number of areas could lead to computa- Let’s utilize the image from the previous
tional challenges. As a result, algorithms like section as input for an object detection pro-
R-CNN, YOLO, and others have been devel- cess, which encompasses two tasks [8]: object
oped to swiftly identify instances of interest. localization and image classification. Initially,
object localization draws bounding boxes
4. Proposed Method around one or more objects in an image, as
illustrated in Figure 4.2. Subsequently, image
The following arguments make the YOLO classification predicts the type or class of an
algorithm crucial: object by comparing it with a predefined
dataset and assigning a class label to the
1. Speed: This technique is able to predict
image. Consequently, the YOLO algorithm
objects in real time, which speeds up
is employed for both object localization and
detection.
image classification.
2. High accuracy: With minute background
error, the YOLO forecast approach yields • The advantages of this approach include:
decisive results. • Facilitating the correct identification of
3. Learning capability: Due to the algo- objects through speech.
rithm’s superior learning capabilities, it is
capable of distinguishing object represen-
tations and use them for detecting
4.1. Data set
The term COCO, which stands for Common
Objects in Context, refers to the original goal Figure 4.2. Feature extracting process by RCNN
of the picture collection, which was to improve methodology.
visual understanding. Computer vision Source:
AUTHOR'S SELF GENERATED
36 Innovations in Computational Intelligence: Trends and Technologies
• Contributing to decision-making processes.
• Mitigating human risk.
• Eliminating the need for batteries
implanted within the body.
4.2. Software packages used
• NumPy: A versatile package designed for
array handling is known as NumPy. It
provides a multidimensional array object
with exceptional speed and features for
efficient collaboration with these arrays.
Figure 4.3. Block diagram for text to speech
• OpenCV: An open-source computer
conversion.
vision library, OpenCV, enables machines
Source:
to identify faces and other objects. AUTHOR'S SELF GENERATED
• pyttsx3: A Python library for text-to-
with image processing methods such
speech conversion, pyttsx3 is compatible
as noise removal, followed by feature
with both Python 2 and 3 and operates
extraction to identify lines, regions, and
offline, distinguishing it from interactive
potentially emphasize certain textures.
libraries.
The comprehension and extraction of
• PyWin32: For utilizing the Win32 appli-
objects in an image or video are achiev-
cation programming interface (API) func-
able through computer vision technology
tionality with Python on Windows, the
known as object detection. Specifically,
PyWin32 library offers valuable extensions.
object detection employs RCNN technol-
• OpenCV- In addition to the OpenCV
ogy to create bounding boxes around the
module found in the open-source library,
identified objects.
the system requires OpenCV contrib
• Object Tracking: Using a camera as well as
to effectively implement SURF feature
blob detection, a moving object is tracked
descriptions.
over time. Blob detection involves search-
• Imutuls: Basic image processing opera-
ing for areas in a digital image that exhibit
tions such as boundary detection, sort-
distinct characteristics from the surround-
ing contours, skeletonization, translation,
ing areas, such as differences in brightness
rotation, and scaling are made easier
or color. In a broader sense, a blob is a
using OpenCV. Python 2.7 and Python 3
distinct section of an image. In the con-
are supported.
text of object recognition, these identified
• winsound: The winsound module provides
zones may signify the presence of objects
access to primary sound-playing com-
or portions of objects in the visual domain.
ponents offered by Windows systems. It
The combination of YOLO with RCNN is
includes various constants and functions,
a widely adopted strategy for blob detec-
such as the ability to beep the PC’s speaker.
tion. Because it can produce forecasts for
the whole image in a single algorithm run,
4.3. Text to speech conversion YOLO is particularly helpful in this regard.
Primarily, an image or video can be provided • Deep Neural Network (DNN): Deep Neu-
as input, as illustrated in Figure 4.3. ral Networks simulate the neuronal net-
work in the brain. Deep neural networks
• Object Detection: The process of detecting are a part of the “deep learning” disci-
objects in any image typically commences pline of machine learning. Deep learning
Revolutionizing object recognition beyond CNN and YOLO with deep learning breakthrough 37
algorithms consist of interconnected lay- given time, the camera will identify and label
ers. The hidden layers are found in the 5–6 objects, as illustrated in Figure 4.4.
front-end layer, which is addressed as the These steps offer an overview of the key
input layer, and the back-end layer, that procedures involved in running your Python
which is addressed as the output layer. program, starting from importing libraries
The term ‘deep’ indicates a network that to the continuous video processing loop, and
connects neurons across more than two concluding with the program’s termination.
layers. Each Hidden layer comprises neu- Keep in mind that executing the program
rons that are activated during the process. requires running it within a suitable Python
The network consumes large volumes development environment or IDE.
of input data, processing them through
several layers, and progressively learns 1. Import necessary libraries (cv2, imutils,
intricate characteristics of the data at numpy, pyttsx3).
each layer. This iterative learning process 2. Define a text-to-speech (TTS) function.
occurs at each stage of the layer, allowing • Initialize the TTS engine.
the network to understand progressively • Adjust the TTS engine’s properties.
complex features of the input data. • Use TTS to speak labels.
• Trained Data: In this work, a pre-defined 3. Load the YOLO model and class names.
dataset has been imported, consisting of a 4. Open a camera feed for video capture.
network trained on more than 50 images. 5. Start a loop to continuously process
This trained network excels in classify- frames from the camera.
ing photos into various object categories, • Resize the frame for processing.
encompassing a wide range, from animals to • Detect objects in the frame using
everyday objects like keyboards, mice, and YOLO.
various stationery items. The pre-trained • Filter and display the detected objects.
model utilized in this context is COCO • Use TTS to announce the labels of
(Common Objects in Context). Leverag- detected objects.
ing the OpenCV-contrib-python package, • Continue processing frames until the
the recognized objects are compared to a user presses ‘q’ to quit.
pre-trained database. The identified images 6. Release the video capture when done.
are then labeled through a tagging process,
translating the visual information into
text. Subsequently, the spelled-out items
are announced using the winsound pack-
age, and the text is converted into speech
through the pyttsx3 package.
5. Results and Discussion
The pre-trained database, named ‘coco,’ is
sourced from the internet and includes pre-
trained objects. Alternatively, another data-
base can be chosen from the internet.
Upon running the code, the camera mod-
ule will open using the OpenCV package, Figure 4.4. Detection of object by camera
detecting objects in front of the camera. The module.
detected objects will be labeled, and at any Source:
AUTHOR'S SELF GENERATED
38 Innovations in Computational Intelligence: Trends and Technologies
After detecting objects, the identified References
objects will be spelled out using the winsound
[1] Pablo Argüeso. 2022. Human ocular
package. The winsound module provides
mucins: The endowed guardians of sight,
access to the fundamental sound-playing Advanced Drug Delivery Reviews. Volume
capabilities offered by Windows platforms, 180. 114074. https://doi.org/10.1016/j.
including various functions and constants. addr.2021.114074.
Additionally, the detected objects can be [2] Boyce P. Light. 2022. Lighting and
logged into a separate file, as depicted in human health. Lighting Research and
Figure 4.5. Technology. 54(2): 101–144. doi:
10.1177/14771535211010267.
[3] V. Adusumilli, M. F. Shaik, N. Kolavennu, L.
B. M. T. Adepu, P. A. V., and I. R. Raja. 2023.
Reading Aid and Translator with Raspberry
Pi for Blind people. 9th International Con-
ference on Advanced Computing and Com-
munication Systems (ICACCS. Coimbatore,
India, pp. 327–331).
[4] Bi, Siguo, Cong Wang, Jilong Zhang, Wutao
Huang, Bochun Wu, Yi Gong, and Wei Ni.
2022. A Survey on Artificial Intelligence
Aided Internet-of-Things Technologies in
Emerging Smart Libraries. Sensors 22(8):
2991. https://doi.org/10.3390/ s22082991.
[5] S. M. Zekavat, et al. 2022. Deep Learning of the
Retina Enables Phenome- and Genome-Wide
Figure 4.5. List of detected objects.
Analyses of the Microvasculature, Circula-
Source: tion 145:134–150. https://doi.org/10.1161/
AUTHOR'S SELF GENERATED
CIRCULATIONAHA.121.057709.
6. Conclusion [6] Ruiqi Cheng, Weijian Hu, Hao Chen,
Yicheng Fang, Kaiwei Wang, Zhijie Xu,
In conclusion, YOLO, a speedy and versatile and Jian Bai. 2021. Hierarchical visual
object detection method, efficiently identi- localization for visually impaired people
fies a range of objects in diverse scenarios. using multimodal images. Expert Systems
It transforms images into text and spoken with Applications, Volume 165. https://doi.
words, aiding individuals with visual impair- org/10.1016/j.eswa.2020.113743.
ments in comprehending visual content. The [7] Kuriakose, B., Shrestha, R., Sandnes, F.
study underscores YOLO’s rapid object rec- E. 2022. LiDAR- Based Obstacle Detec-
ognition and its role in enhancing accessibil- tion and Distance Estimation in Naviga-
ity and understanding of visual information, tion Assistance for Visually Impaired. In:
particularly for those with vision challenges. Antona, M. and Stephanidis, C. (eds), Uni-
By amalgamating various technologies, the versal Access in Human- Computer Inter-
action. User and Context Diversity. HCII
research strives to boost the speed of object
2022. Lecture Notes in Computer Science,
identification and improve the overall com-
vol 13309. Springer, Cham. https://doi.
prehension of visual content. Essentially, org/10.1007/978-3-031-05039-8_35.
YOLO emerges as a crucial player in this [8] Xiangqing Shen, Bing Liu, Yong Zhou, Jiaqi
endeavor, showcasing its effectiveness beyond Zhao, and Mingming Liu. 2020. Remote
conventional methods and contributing sig- sensing image captioning via Variational
nificantly to a more inclusive approach to Autoencoder and Reinforcement Learn-
interpreting the visual world. ing. Knowledge-Based Systems. Volume
Revolutionizing object recognition beyond CNN and YOLO with deep learning breakthrough 39
203. 105920, https://doi.org/10.1016/j. Technologies in Electronics, Computing and
knosys.2020.105920. Communication (ICETECC): 1–6.
[9] Md. Atikur Rahman and Muhammad Sheikh [12] Harum, N. B., Zakaria, N. A., Emran, N.
Sadi. 2021. IoT Enabled Automated Object A., Ayop, Z., and Anawar, S. 2019. Smart
Recognition for the Visually Impaired. Com- Book Reader for Visual Impairment Person
puter Methods and Programs in Biomedicine using IoT Device. International Journal of
Update, Volume 1, https://doi.org/10.1016/j. Advanced Computer Science and Applica-
cmpbup.2021.100015. tions. 10(2): 251–259.
[10] Ganesh, A. 2019. OCR based Image Pro- [13] Vinaya Phutak, Richa Kamble, Sharmila
cessing with Audio Output for Visually Gore, Minal Alave, and R. R. Kulkarni. 2019.
Challenged People. International Journal Text to Speech Conversion using Raspberry –
for Research in Applied Science and Engi- PI. International Journal of Innovative Science
neering Technology. 7. 599–604. 10.22214/ and Research Technology. 4(2): 291–293.
ijraset.2019.3104. [14] Alzubaidi, L., Zhang, J., Humaidi, A. J. et al.
[11] Sarwar, S., Turab, M., Channa, D., Chan- 2021. Review of deep learning: concepts, CNN
dio, A., Sohu, M. U., and Kumar, V. 2022. architectures, challenges, applications, future
Advanced Audio Aid for Blind People. 2022 directions. J Big Data 8, 53 (2021). https://
International Conference on Emerging doi.org/10.1186/s40537-021-00444-8.
5 Investigating efficacy of emerging ICTs
for the Indian construction sector
Sanjay Pandeya and G. T. Thampib
Thadomal Shahani Engineering College Bandra(w), Mumbai, India
Abstract: The construction sector, being one of the least digitized and major contributors in developed or
developing economies faces lots of challenges. We see an opportunity for reorienting the sector with cutting-
edge information and communication technologies (ICTs) integration. The construction sector challenges
namely poor productivity, lower efficiencies, safety, and sustainability can be addressed by ICT adaptation. The
research objective is to investigate the efficacy of utilizing ICTs as a technological intervention for the Indian
construction sector. The methodology adopted in this research study encompasses a hybrid approach, combin-
ing qualitative interviews of industry professionals and quantitative analytics of data gathered from question-
naire survey of construction projects. The findings of the questionnaire survey highlight that we can harness
emerging digital technology in various construction tasks to overcome challenges of the sector, and also identify
numerous potential benefits of utilizing ICTs. Additionally, the study also explores the challenges and barriers
faced by stakeholders in the adoption and implementation of cutting-edge ICTs within construction companies.
Keywords: Construction sector, cutting-edge ICTs, efficacy, productivity, efficiency, safety, cost-effectiveness
1. Introduction possible with the conventional approach.
With this fresh insight into the projects,
Recently, significant growth has been seen in stakeholders can make better decisions more
the construction sector due to urbanization quickly and intelligently which results reduc-
and infrastructure development. However, tion in time, cost, and risks, thus improving
the sector is facing challenges with efficiency, the performance of the overall project [1–3].
cost-effectiveness, and safety. Driven by rapid In this paper, the effectiveness and impact of
urbanization, expanding infrastructure, and emerging ICTs on the Indian construction
growing demand for innovative construction sector will be thoroughly examined. In the
technologies and solutions, the Indian con- conventional approach of construction, tasks
struction sector needs to undergo a dynamic are very slow and time-consuming as they are
transformation. So, by keeping those chal- labor and paper-based resulting in slippage
lenges in mind, the inclusion of emerging of schedule and budget overrun which ulti-
ICTs such as drones, IoT, 5G, Cloud, AI/ mately force projects to fall into the catego-
ML, etc. has become an important aspect of ries of failed or challenged projects. On the
transforming the conventional construction other hand, the ICT-enabled smart approach
approach into a smarter approach. Emerg- of construction where tasks are carried out
ing ICTs can change the way construction at a reasonably faster speed with optimum
projects are designed, monitored, and exe- scale as they are digitally enabled, results
cuted. For instance, ICTs have the capability in the completion of tasks on schedule and
of fetching real-time data that can provide budget which in turn enhances the productiv-
insight into projects that were previously not ity and efficiency of the projects. Therefore,
[email protected]; [email protected]
a
DOI: 10.1201/9781003606635-5
Investigating efficacy of emerging ICTs for the Indian construction sector 41
ICT adaptation in different construction extract meaningful insight from it, which ulti-
tasks proliferates the probability of success- mately will help in better decision-making.
ful completion of construction projects. In Doing so all stakeholders can benefit from it
this research work, we have identified various and coordination and communication will be
construction tasks where ICTs can be utilized, in a very efficient and effective manner. This
numerous advantages of using ICTs in con- sharing of information will curtail the time
struction companies, and different hindering required for data gathering, processing, and
factors coming in the way of ICT adaptation decision-making [8, 9]. Researchers have also
by the construction sector [4-6]. So, to carry highlighted that the utilization of wireless
out our research work we have conducted a networking omits the need for fellow work-
series of interviews with industry profession- ers, not to be present physically at the site,
als and recorded their valuable opinions and at the same moment. Web-based companies
suggestions. We have also prepared a ques- can get an edge in their operations [10]. Few
tionnaire, distributed it among the industry researchers have highlighted virtual modeling
expert, obtained their inputs, and lastly ana- for building design. It includes all relevant
lyzed the data obtained. The utilization of design information needed for every con-
emerging ICTs in Indian construction compa- struction phase during the project life cycle.
nies is a technological paradigm shift that has Any modification in the models gets updated,
the potential not only to change the way the and any design-related information may get
industry is working but also to enhance pro- extracted from this database, automatically
ductivity and efficiency, resulting in a higher [11, 12]. Some researchers are also talking AQ1:
Please
success rate of the projects. about the formation of a virtual team for col- provide
laborative purposes to improve communica- reference
tion, coordination and sharing of information citation
for refer-
2. Literature Review of ICT among stakeholders, teams, and enterprises. ences 13,
Adaptation in the Construction The ideation and creation of virtual teams for 14, 20,
21.
construction companies have been advised
Companies by the researchers [15]. ICTs may support
After the review process of the literature, we the virtual team for routine operations, but
have found that Some research papers high- unfortunately, construction enterprises are
lighted that the industry is using some basic much behind in adopting them.
ICTs like MS Office, AutoCAD, Wi-Fi, etc. [6, The present scenario of ICTs utilized in
7] for construction tasks. Many authors con- different construction company activities is
cluded that the utilization of ICTs can aid in discussed in this section. We have studied sev-
enhancing the present situation of construction eral conference papers, journal papers, blogs,
companies, apart from the fact that construc- and articles on different websites, discussing
tion companies are digitally divided, they need old and current trends of the different ICTs
to embrace ICTs to maximize the productivity for construction tasks or applications. It has
and efficiency of the construction enterprises. been noticed that many construction activities
Transfer of data and information among harnessed BIM, AutoCAD, mobile/web-based
the stakeholders are very slow and limited tools and technologies, virtual conferencing,
during and after the construction work as and information sharing systems, etc. [16,
many professionals are not interested in shar- 17]. Few articles, white papers, and bloggers
ing information and keeping it in their silos. have mentioned adaptation of emerging ICTs
So, with the advent of these newer commu- viz. drone, IoT, 5G, Cloud, Blockchain, and
nication technologies in the construction sec- AI/ML, etc. may be good options for different
tor, companies can get this precious data and construction activities/tasks. Deployment of
42 Innovations in Computational Intelligence: Trends and Technologies
various ICTs for construction activities is in 3. Methodology
the early stage and good attention is needed
by the researcher to grow the present situa- 3.1. Research objectives
tion of construction companies [18, 19]. It has
been noticed that many construction activi- Before carrying out research work, objectives
ties are using AutoCAD, mobile/web-based were finalized. The prime objective of the
applications, BIM, information systems, research is to estimate the efficacy of emerg-
video conferencing, etc. [16, 17] for their day- ing ICTs for the Indian construction sec-
to-day activities to enhance their productivity tor. Quantifying the overall performance of
[22]. According to many research, increasing construction companies planning to harness
productivity and building efficiencies may be ICTs, is another goal of the research project.
possible with the ICT adaptation. Planning,
designing, and implementation of a project 3.2. Research questionnaire
need effective knowledge sharing among var-
Based on the evaluation of the present sce-
ious stakeholders. It can be obtained through
nario, the research questionnaire was formu-
proper adaptation of ICT by the construction
lated for the Indian construction companies.
companies [23, 24]. Construction industries,
whether they are small, medium, or large • What is the impact of harnessing emerg-
can use emerging ICTs to enhance efficiency ing ICTs for construction sector tasks?
and productivity by minimizing dependabil- • What are the different benefits of utilizing
ity. The utilization of ICTs by construction ICTs in construction firms?
companies has various advantages, including • What are various hindering factors for
enhanced productivity, cost reduction, time ICT adaptation in the Indian construction
savings, information sharing, waste manage- sector?
ment, and many more [25]. Construction site
employees with limited technical proficiency,
fear of virus attacks, and incompetence when 3.3. Data collection and analysis
handling computing devices have limited their To collect industry data, questionnaires were
ability to use them properly [26, 27]. Also, designed and distributed. During the early
professionals think that ICT embracement is phase of the survey, we created a Google
a positive initiative to enhance the efficiency form and floated the questionnaire electroni-
and productivity of construction companies. cally among the industry professionals with
Construction firms have limited success the humble request to participate in research
due to the low penetration of emerging ICT work, but unfortunately received negligi-
solutions. We did an exhaustive literature ble responses. Therefore, we decided to go
survey where we noticed many hurdles to the offline survey and approach industry profes-
embracement of ICTs in construction compa- sionals personally. Many professionals have
nies [28, 29]. ICT adoption is confined by the not shown their interest in participating in
complex structure of the construction sector, the survey but a few interested professionals
low productivity, professionals with relatively seriously contributed by providing genuine
low work experience, schedule overruns, and responses to the questionnaire and also will-
budget overruns. A good no of researchers ingly participated in the interview process to
has done their research to assess the embrace- provide important insight about the activities
ment and utilization of ICT for construction which helped me for a meaningful contribu-
work and tried to earmark various reasons tion to the existing body of knowledge. Here
that are acting as a hurdle for construction data were obtained from primary sources of
works [30–32]. information including questionnaires and
Investigating efficacy of emerging ICTs for the Indian construction sector 43
interviews. In 223 questionnaires were dis- middle management and 3rd cadre is lower
tributed for data collection and we obtained management. The participation of the 1st
118 responses. cadre is 37% whereas 41% of respondents
A Likert scale of 3-points has been used for are playing their role under the 2nd cadre and
analysis of data, where the scale ranges from belonging in the middle management catego-
1 to 3. The scale of “Agree” was 3, “Disagree” ries. The lower management category share
was scored 2, and “Neutral” was scored 1, is 22%.
The statistical tool used for data analysis and
visual representation of this research study
was Microsoft Excel. 4. Results and Discussion
After the questionnaire survey of the con-
3.4. Types of construction companies struction companies, it has been noticed
that emerging ICTs viz. drone, IoT, and 5G
that contributed to the study
technology and cloud, etc. can be a power-
Figure 5.1 shows the types of construction ful digital solution for the construction sector
enterprises that participated in the research to improve productivity and efficiency with
study. The contribution of large-scale enter- the enhanced safety of the personnel. These
prises is 33% whereas 41% of enterprises are cutting-edge ICTs can be used for various
in medium-scale categories, collectively these construction tasks and it has the potential to
two categories make it nearly three-fourths change the ways construction tasks are per-
of the participating organization. Small-scale formed through conventional approaches.
and microscale enterprise shares are 15% and
11% respectively. 4.1. ICTs can be utilized to perform
different construction sector’s
3.5. Cadre-based classification tasks
(CBC) of respondents
4.1.1. Topographic mapping and land
Figure 5.2 shows the Cadre-based classifica-
tion of the respondents and this cadre depends survey
upon the type of role they performed in their Topographic mapping and land survey-
enterprises. There are three cadre of respond- ing using ICT, particularly drones have
ents that participated in the research study. become increasingly popular due to their
1st cadre is higher management, 2nd cadre is time-efficient, cost-effective and ability to fly
AQ2:
Please
provide
source
for all the Figure 5.1. Types of construction companies Figure 5.2. Cadre-based classification of
figures contributed to the study. respondents.
and tables
Source: Source:
AUTHOR'S SELF GENERATED
AUTHOR'S SELF GENERATED
44 Innovations in Computational Intelligence: Trends and Technologies
hard-to-access terrain. 82% of respondents 4.1.5. Personnel safety
are willing to adopt drones for topographic
IoT-enabled wearables can monitor health
mapping and land surveys in their enterprises
parameters like body temperature, pressure
as shown in Figure 5.3.
even ECG of workers. Construction work-
ers frequently have to climb unsteady heights
4.1.2. Equipment tracking and navigate around hazardous conditions.
Industry professionals can effectively use ICTs, particularly drones can be used to
ICTs for equipment tracking, functioning replace workers in such a dangerous situa-
status of equipment, enhancing security, etc., tion and mitigate the risk. According to 73%
which can reduce downtime, and improve of respondents, worker safety in the construc-
asset/equipment visibility and overall man- tion industry is of utmost importance since it
agement, and this is supported by 67% of can seriously harm personnel and the indus-
respondents who like to adopt ICTs for try as well, as shown in Figure 5.3.
tracking equipment in their organizations as
shown in F igure 5.3. 4.1.6. Structural inspection
Emerging ICTs, particularly drones fitted
4.1.3. Real-time progress reporting with sensors, cameras, and other devices can
Harnessing ICTs can fetch real-time data from examine the fine details and stability of struc-
construction sites and equipments, allowing tures. Drones equipped with thermal sensors
all stakeholders (Clients, contractors, design- can detect temperature variations, which can
ers project managers, etc.) to keep track of the reveal hidden problems such as electrical,
project’s progress and identify any deviations water, or insulation problems. In the long run,
from the original plan of action. This enables this quality assurance facilitates the develop-
timely adjustments and helps in meeting pro- ment of client relationships. 58% of respond-
ject deadlines, thus increasing the possibili- ents support the use of ICTs for structural
ties of successful project completion and this inspection. as shown in Figure 5.3.
real-time collected data can be used by spe-
cialized software to generate visual reports. 4.1.7. Supply replenishment
As per our survey findings, 80% of respond- Counting of materials becomes easy when
ents are willing to embrace ICTs for real-time supply units are hooked up with IoT-ena-
progress reporting of the project as shown in bled tags. As soon as the system notices that
Figure 5.3. the counting of units drops below a certain
threshold level it automatically generates
4.1.4. Security surveillance order requests, thus waiting time goes down,
and the probability of completion of projects
The ICTs-driven security surveillance pro-
increases. This also reduced the cost incurred
vides an efficient solution for monitoring large
with the project. As shown in Figure 5.3, 75
areas, enhancing situational awareness, and
% of respondents are likely to adopt ICTs for
responding to security incidents. The sensors
supply replenishment purposes.
of drones can ensure whether the equipment
is in a secure enough location or not. Drone
surveillance cameras can also ensure unau- 4.1.8. Concrete curing
thorized access to the site. our survey findings Sensors deployed in concrete can transmit
reflect that 72% of professionals would like temperature, strength, and humidity data to
to adopt it for security surveillance as shown phones or tablets via a 5 G-enabled mod-
in Figure 5.3. ule, and then managers can plan curing
Investigating efficacy of emerging ICTs for the Indian construction sector 45
Table 5.1. Respondent’s opinion ranking of
construction tasks performed by ICTs
Sr Construction Respondent’s
No. industry tasks Opinion
Performed by Percentage Ranking
ICTs (%)
1 Topographic map 82 1
and land survey
Figure 5.3. Harnessing ICTs for construction
industry tasks. 2 Equipment 67 8
tracking
Source:
3 Real-time 80 2
AUTHOR'S SELF GENERATED
progress reporting
4 Security 72 6
accordingly. As depicted in Figure 5.3, 76% surveillance
of the respondents are willing to use emerg-
5 Personnel safety 73 5
ing ICTs for concrete curing.
6 Structural 58 9
inspection
4.1.9. Waste management 7 Supply 74 4
Sensor-enabled smart bins or vehicles help in replenishment
handling waste properly, doing so penalties 8 Concrete curing 76 3
for the various contractors can be avoided. 9 Waste 69 7
Figure 5.3 shows that 69% of respondents management
are interested in adopting ICT-enabled waste
Source:
management for their enterprises AUTHOR'S SELF GENERATED
4.2. Respondent’s opinion ranking of 4.3. Perceived benefits of using ICTs
construction tasks performed by for construction companies
ICTs
The integration of ICTs in construction com-
Respondent’s opinion ranking of construc- panies brings forth a range of substantial
tion tasks performed by ICTs has been shown benefits that positively impact various aspects
in Table 5.1. As per this ranking table, 82% of the construction process. Numerous ben-
of respondents are willing to use ICTs, par- efits of using ICTs for performing different
ticularly drones for topographic mapping and tasks in the construction industry have been
land surveys of their enterprises, and in the perceived.
ranking table, this task is ranked as 1. Real-
time progress reporting is having rank 2 with
80% response. Concrete curing has 3rd rank 4.3.1. Time saving
in the rank table with 76%, supply replace- ICTs have the potential to enhance time effi-
ment is ranked 4th with 74%, personnel safety ciency in the construction industry by con-
is ranked 5th with 73% and Security Surveil- tributing in several ways whether it is through
lance is ranked 6th with 72% of responses, surveying and mapping, site inspection, mon-
waste management, and equipment tracking itoring construction progress, safety inspec-
having 69% and 67% responses respectively, tions, Concrete curing, etc. As per our finding
lastly, structural inspection has 9th rank with 88% of respondents were willing to use ICTs
58% positive responses. for time saving as shown in Figure 5.4.
46 Innovations in Computational Intelligence: Trends and Technologies
4.3.2. Budget controlling
ICTs can streamline several construction
activities like topographic mapping and land
surveys, real-time progress monitoring, etc.
which can lead to substantial cost and time
savings. Our survey reflects that 82 % of
respondents are looking to embrace digital
solutions for overall budget control of pro-
jects as depicted in Figure 5.4. Figure 5.4. Perceived benefits of using ICTs for
construction companies.
Source:
4.3.3. Enhance safety and security AUTHOR'S SELF GENERATED
Drones equipped with sensors can improve
the security of the construction site by pro-
4.4. Ranking of the benefits of using
viding 24/7 monitoring and surveillance. Any ICTs as per respondent’s opinion
attempt made by the trespasser or unauthor- Table 5.2 depicts a rank matrix of various
ized personnel can be detected immediately advantages of using ICTs for different con-
and a response to the security threats shall be struction tasks. One of the significant bene-
sent. As per Figure 5.4, survey results reflect fits is time-saving, having rank 1 in the rank
that 80% of respondents feel that ICTs may matrix is supported by 88% of respondents.
uplift the present safety and security scenario 86% of respondents support the improve-
of construction works and sites. ment of productivity which is ranked 2nd in
the matrix. Budget controlling is having 3rd
4.3.4. Reduced paperwork rank with the 82% responses. Enhancement
of safety and security has 80% support and
We have identified multiple ways in which is ranked 4th. 69% of respondents support a
ICTs can facilitate the minimization of reduction in the paper which will contribute
paperwork such as digitally enabled sur- to green construction and having 5th rank.
veys and mapping, progress tracking and
reporting, inventory management, etc. In
our survey, 69% of respondents feel that Table 5.2. Ranking of the benefits as per
ICTs can digitize the construction process respondent’s opinion
hence reducing paperwork, as shown in Sr Perceived benefits Respondent’s Opinion
Figure 5.4. This also promotes sustainable No. of adopting ICTs Percentage Ranking
construction practices by reducing carbon for construction (%)
footprint. firms
1 Time-saving 88 1
4.3.5. Improved productivity 2 Budget 82 3
controlling
Incorporating ICTs in the construction indus-
try can help to improve productivity in several 3 Enhanced safety 80 4
and security
ways such as streamlining various processes,
enhancing communication among the stake- 4 Reduced 69 5
holders, fetching real-time data, etc..86% of paperwork
respondents would like to adopt ICTs for 5 Improved 86 2
various tasks to enhance the productivity of productivity
enterprise as shown in Figure 5.4. Source:
AUTHOR'S SELF GENERATED
Investigating efficacy of emerging ICTs for the Indian construction sector 47
4.5. Hindering factors for ICT management support is ranked 2nd with 81%
adaptation in the indian of respondent’s opinion. 79% of Respondents
construction sector are in the notion that the training of employ-
ees to use ICTs is also a serious hurdle and
As depicted in Figure 5.5, the primary barrier ranked as 3rd. Fear of job security is also a
to ICT adoption in the construction sector prime concern for the respondents with 71%,
is the higher cost of purchasing and main- ranked as 4th. Government support for ICT
taining ICTs. 85% of respondents say that integration is at 5th place with 68% and lastly,
the biggest concern is the higher cost of ICT legal and regulatory compliance have the last
adoption. 81% of respondents feel manage- position in the matrix with 64%. Apart from
ment support is 2nd biggest concern. 79% of the above hindering factors respondents also
responders rate training on ICT as a serious highlighted the limited skill of personnel with
concern. 71% of respondents also think that low expertise in managing/handling software
job security is a major hurdle for ICT adapta- solutions and digital devices, data analysis,
tion in construction companies and concerns and its interpretation.
regarding legal and regulatory compliance
are shared by 64% of respondents. Addition-
4.7. Overall performance of
ally, according to 68% of respondents, govt.
support poses a significant obstacle to ICT construction companies after
adoption in the construction industry. ICT adaptation
The utilization of emerging ICTs for construc-
4.6. Ranking of hindering factors tion companies is in the nascent stage and
for ICT adaptation as per has the potential to enhance the overall per-
respondent’s opinion formance of the enterprises. The majority of
The following ranking Table 5.3 depicts the
ranking of the different hindering factors for Table 5.3. Ranking of hindering factors for ICT
ICT adaptation. One of the biggest obstacles adaptation
in the adaptation of ICTs in Indian construc- Sr ICT hindering Respondent’s opinion
tion firms is the “higher cost of ICT adapta- No. factors Percentage Ranking
tion” in the industry which is ranked as 1, in (%)
rank Table 5.3 with 85% of responses. Top 1 Higher cost of 85 1
adaptation
2 Training of 79 3
employees to use
ICTs
3 Government 68 5
support
4 Management 81 2
support
5 Fear of job 71 4
security
6 Legal and 64 6
Figure 5.5. Hindering factors for ICT regulatory
adaptation in the Indian construction sector. compliance
Source: Source:
AUTHOR'S SELF GENERATED
AUTHOR'S SELF GENERATED
48 Innovations in Computational Intelligence: Trends and Technologies
Regulatory compliance remains a significant
hurdle, necessitating ongoing dialog between
industry stakeholders and regulatory bodies
to establish clear guidelines. This quantified
investigation reaffirms the strategic embrace-
ment of emerging ICTs for Indian construc-
tion sectors and holds immense promise
for revolutionizing traditional practices of
accomplishing numerous construction tasks.
Figure 5.6. Overall performance of construction The benefits of enhanced efficiency, safety,
companies after ICT adaptation. cost-effectiveness, and sustainability have the
Source: AUTHOR'S SELF GENERATED
potential to reshape the construction sector.
However, the study also sheds light on the
respondents agree that overall performance challenges that accompany the integration of
will be enhanced by 85%, after the adapta- emerging ICTs in Indian construction compa-
tion of cutting-edge ICTs by the construction nies. This research work is of good use for the
companies, as shown in Figure 5.6. construction sector professionals, research-
ers, and academic fraternity as ICTs reflect its
usefulness for a group of construction tasks
5. Conclusion with multiple benefits.
In the dynamic landscape of Indian construc-
tion enterprises, the integration of ICTs has References
emerged as a pivotal catalyst for construc-
tion sector transformation. The study find- [1] Dixit S, Stefan A, Musiuk A, and Singh P.
ings affirm that the utilization of ICTs in Study of enabling factors affecting the adop-
construction projects has demonstrated sig- tion of ICT in the Indian built environment
sector. Ain Shams Eng J. 2021;12:2313-2319.
nificant improvements in various critical
[2] Bello SA, Oyedele LO, and Akinade OO.
areas. The ability of ICTs to conduct precise Cloud computing in the construction indus-
site surveys, monitor progress in real-time, try: Use cases, benefits and challenges.
collect accurate data, supply replenishment, Autom Constr. 2021;122:103441.
waste management, etc., has enhanced pro- [3] Sawant RS, Ravikar A, Bagdi N, and Bel-
ject productivity, efficiency, and decision- lar V. Drone technology in the construction
making processes. Additionally, the improved industry in the Indian construction industry.
safety measures through remote inspections VIIRJ. 2020;Special Issue:2319-4979.
and early hazard detection contribute sub- [4] Mahajan G. Applications of drone technol-
stantially to the overall well-being of the ogy in construction industry: A study 2012-
workforce. The time and cost savings realized 2021. Int J Eng Adv Technol. 2021;11(1).
from streamlined processes, reduced rework, [5] Pandey S, and Thampi GT. Literature review
enhanced safety and security, and optimized on efficacy of ICT in realm of construc-
tion industry. J Emerg Technol Innov Res.
resource allocation can help to enhance the
2022;9(8).
project’s performance. The financial impli- [6] Gamil Y. Internet of things in construction
cations of ICT implementation cannot be industry revolution 4.0. Department of
understated and initial investment costs for Civil and Environmental Engineering, Uni-
ICT may be a serious concern. Skill develop- versiti Tun Hussein Malaysia, Batu Pahat,
ment is another critical aspect, requiring con- Malaysia.
certed efforts in training and upskilling the [7] Onyegiri I, Nwachukwu CC, and Onyegiri J.
workforce to maximize the potential of ICTs. Information and communication technology
Investigating efficacy of emerging ICTs for the Indian construction sector 49
in the construction industry. Am J Sci Ind [20] Chien HJ, and Barthorpe S. The current state
Res. 2011. of information and communication technol-
[8] Information and communications technol- ogy usage by small and medium Taiwanese
ogy in construction. Designingbuildings. construction companies. J Inf Technol Con-
co.uk. Last edited March 30, 2022. Accessed str. 2014;15(5):75–85.
September 28, 2024. [21] Mesaros P, and Mandiak T. Impact of ICT on
[9] Paudyal G. Role of ICT in construction. In: performance of construction companies in
National Students Conference on Informa- Slovakia. IOP Conf Ser Mater Sci Eng. 2017.
tion Technology (NaSCoIT). Nepal; 2016. [22] Griffis FH, Hogan DB, and Li W. An analy-
[10] Orihuelaa P, Orihuelab J, and Pachec S. sis of the impact of using three-dimensional
Information and communications technol- computer models in the management of con-
ogy in construction: A proposal for pro- struction. Constr Ind Inst (CII) Austin, TX,
duction control. In: Creative Construction USA. 1995;Research Report 106–11.
Conference 2016, CCC 2016; 2017:25–28. [23] Kang Y, O’Brien WJ, Thomas S, and Chap-
[11] Sharma DK. Digital revolution 4.0: Informa- man RE. Impact of information technologies
tion and communication technology (ICT) on performance: Cross-study comparison. J
in construction industry. JK Lakshmipat Constr Eng Manag. 2008;134:852–863.
University, Jaipur; July 16, 2020. [24] Owolabi OSB, and Olufemi ODK. Effect of
[12] Hosseini MR, Chileshe N, Zuo J, and the use of ICT in the Nigerian construction
Baroudi B. Approaches of implementing ICT industry. Int J Eng Sci. 2018;7(5):71–76.
technologies within the construction indus- [25] Mu’azu DA. The role of the professional
try. Univ South Aust. 2013. builder in the Nigerian construction industry.
[13] Chinowsky PS, and Rojas E. A guide to suc- ATBU J Environ Technol. 2015;1(1):29–31.
cessful implementation: A report on con- [26] Yisa SO. An investigation into the use of
struction industry institute. Univ Texas ICT in the Nigerian construction industry.
Austin, Constr Ind Inst. 2012. Unpublished HND Project, Department of
[14] Brandon PS, Kocaturk T, and Foundation R. Building Technology, The Federal Polytech-
Virtual futures for design, construction and nic, Ado Ekiti. 2014.
procurement. Malden, Ma: Blackwell Pub; [27] Oladapo AA. The impact of ICT on pro-
2018. fessional practice in the Nigerian construc-
[15] Zhang Y, Gregory M, and Shi Y. Global engi- tion industry. Electron J Inf Syst Dev Ctries.
neering networks (Gen): Drivers, evolution, 2016;24(2):1–19.
configuration, performance and key patterns. [28] Leverage on ICT in construction industry.
J Manuf Technol Manag. 2008;19:299-314. New Strait Times. May 13, 2018.
[16] Chen C, and Messner JI. A recommended [29] Al-shammary S, and Ali AA. ICT hindering
practices system for a global virtual engi- factors applied in Jordan construction pro-
neering team. Archit Eng Des Manag. jects. Civ Eng Archit. 2017;5(3):83–88.
2014;6:207–221. [30] Hassan A, and Afrah MHK. Identification
[17] Adwan EJ, and Al-Soufi A. A review of ICT and analysis of hindering factors of ICT
technology in construction. Int J Manag Inf adoption in project management in Iraq. J
Technol. 2016;8(3/4). Univ Babylon Eng Sci. 2019;27(3).
[18] Sarshar M, and Isikdag U. A survey of ICT [31] Shah K, Soni N, and Shah Z. Scrutinizing
use in the Turkish construction industry. Eng attributes influencing role of information
Constr Archit Manag. 2015;11(4):238-247. communication technology in building con-
[19] Mutesi N. Application of ICT in the con- struction. Int Res J Eng Technol. 2020;7(3).
struction industry in Kampala. In: Advances [32] Vasista TG, and Abone A. Benefits, barri-
in Engineering and Technology: Proceedings ers and applications of information com-
of the Second International Conference on munication technology in construction
Advances in Engineering and Technology, industry: A contemporary study. Int J Eng.
Kampala. 2013. 2016;105:933–939.
6 AgriTech: Empowering agriculture
through integrated technology
Pankaj Kunekara, Anushka Popalghatb, Onkar Borudec, Vishal
Gavalid, Dnyanesh Gholape, and Sarvesh Hadolef
Department of Information Technology Vishwakarma Institute of Technology, Pune, India
Abstract: Agriculture plays a pivotal role in ensuring global food security and economic stability. In an era of
rapidly advancing technology, harnessing the power of data and innovation is vital to address the multifaceted
challenges faced by farmers. AgriTech is a comprehensive agricultural technology system designed to empower
farmers at every farming stage. Leveraging a range of technologies, including Full Stack Web Development,
Ethereum, Alan AI and Various Machine learning algorithms, this multifaceted solution offers features such as
nearby soil testing laboratory location, crop recommendations, gamified plant growth mentoring, plant disease
detection, crop price prediction, expert consultations, an integrated e-commerce marketplace, and Realtime
weather monitoring with location-based alerts. This paper provides an overview of AgriTech’s architecture,
emphasizing its technology integration, machine learning applications, and potential impact on modern farm-
ing practices, including enhanced productivity, sustainability, and resource efficiency. AgriTech represents a
pioneering approach to digital agriculture, bridging the gap between traditional farming and the digital era.
Keywords: Agriculture technology, precision farming, crop price prediction, soil testing, disease detection, crop
recommendation
1. Introduction By harnessing the power of machine learn-
ing and real-time data analysis, it furnishes
Traditionally, agriculture relied on time-
farmers with personalized insights, data-
tested practices passed down through genera-
driven recommendations, and a platform to
tions. However, the contemporary farming
connect with agricultural experts, transcend-
landscape demands a departure from conven-
tional methods. AgriTech represents a pivotal ing geographical boundaries and traditional
response to this demand, offering farmers an constraints. Moreover, its gamified approach
integrated and technologically advanced ally to plant growth mentoring and innovative
in their quest for sustainable and productive reward mechanisms promotes knowledge
farming practices. acquisition and sustainable farming practices.
AgriTech seamlessly integrates critical This paper delves into the architecture,
aspects of farming, from soil quality analysis methodologies, and potential impact of
to crop selection, growth management, dis- AgriTech, as it endeavors to bridge the gap
ease detection, price forecasting, expert con- between age-old agricultural practices and
sultations, and real-time weather monitoring. the digital age. In so doing, it unlocks a new
The genesis of AgriTech is rooted in a pro- era characterized by enhanced productivity,
found understanding of the intricate interplay resource efficiency, and sustainability in farm-
between data, technology, and agriculture. ing. AgriTech embodies not just technological
a
[email protected]; [email protected]; [email protected]; dvishal.gavali21@vit.
edu; [email protected]; [email protected]
DOI: 10.1201/9781003606635-6
AgriTech: Empowering agriculture through integrated technology 51
advancement but also a testament to the syn- field which encompasses a wide range of tech-
ergy between human ingenuity and the relent- nologies, including precision farming, robot-
less pursuit of food security and ecological ics, artificial intelligence, and biotechnology,
harmony in an ever-changing world. which have already begun transforming the
agricultural landscape [8]. Discuss the impact
of Internet of Things on farming, including
2. Literature Review using sensors for soil, water, light, humidity,
The authors provide a comprehensive review and temperature management [9]. The paper
of recent advances in image-processing tech- proposes a Blockchain-based framework for
niques for automating leaf pests and disease data quality management in the context of
recognition. They discuss the various stages COVID-19, aiming to ensure a high level of
involved in the image-processing pipeline, data trust, security, and privacy. It explores
from image acquisition to feature extrac- the use of innovative technologies such as
tion and classification [1]. The study Intro- Blockchain, Artificial Intelligence, and Big
duces a framework for crop price forecasting Data in tracking infectious disease outbreaks
in emerging economies, emphasizing the and improving reporting accuracy [10].
importance of accurate price predictions for
farmers and policymakers. The proposed 3. Methodology
framework includes data quality assessment,
In response to the evolving agricultural land-
feature selection, and modeling, offering
scape, the AgriTech project emerges as a com-
advantages such as improved data quality
prehensive agricultural technology platform.
and feature selection techniques for enhanced
This support ranges from initial crop selec-
accuracy [2]. The system’s two-tier architec-
tion to the monitoring of crop growth and
ture and various sensors offer advantages like
culminates in facilitating marketing and sales.
real-time monitoring and portability, while
The platform leverages an amalgamation of
challenges include data quality and real-time
cutting-edge technologies and data-driven
performance [3]. This research Focuses on
methodologies to provide a holistic and User-
online price forecasting for cryptocurrencies
friendly agricultural support system.
like Bitcoin, Ethereum, and Ripple, utilizing
Frontend technologies and functionalities:
artificial intelligence techniques. The model’s
Frontend stack: The front of the AgriTech
advantages include periodic predictions and
platform is built using React, a renowned
customization options.[4].
JavaScript library for developing user inter-
The authors address crop price predic-
faces. This choice of technology ensures a
tion in developing nations, using decision tree
User-friendly and interactive experience.
regression techniques [5]. Authors Delves into
User interface: The frontend features a
crop yield and price forecasting using machine
user-friendly map-based interface, designed
learning, employing the decision tree algo-
to enhance accessibility and usability for
rithm. This research enhances crop yield pre-
farmers.
diction based on weather and price trends [6].
Features:
The authors introduce a neighborhood-
based strategy to optimize parameters and
obtain maximum yield, expanding knowl- 3.1. Soil testing labs map
edge through population exploration. The Farmers can easily locate nearby soil testing
paper presents a method to increase soil fertil- laboratories, simplifying the crucial process
ity performance by providing nutrient recom- of soil quality assessment. This feature facili-
mendations for optimal crop development [7]. tates data-driven decisions about crop selec-
The study examines the Modern agricultural tion and soil management.
52 Innovations in Computational Intelligence: Trends and Technologies
3.3. Plant disease detection and
AQ1:
Citations pesticide recommendation
for all fig-
ures are Utilizing image recognition technology, Agr-
missing in iTech can accurately identify plant diseases
the text,
please
from images of leaves. In addition to disease
provide detection, it recommends appropriate pesti-
the cides, aiding in timely disease management
appro-
priate and prevention.
citations Figure 6.1. Soil testing lab locations.
AQ2: Source: AUTHOR'S SELF GENERATED
Please
provide
source
3.2. Personalized crop recommendations
for all the
figures
The platform employs machine learning algo-
rithms to provide personalized crop recom-
mendations. [11,12]
Figure 6.4. Plant disease prediction.
Source: AUTHOR'S SELF GENERATED
3.4. Coin-based e-commerce
marketplace
AgriTech implements a coin-based reward
system during the plant growth mentoring
process. These earned coins can be redeemed
within an integrated e-commerce market-
Figure 6.2. Heatmap for crop recommendation. place, fostering user engagement and partici-
Source:AUTHOR'S SELF GENERATED
pation.[16,17]
By analyzing soil data and real-time
weather conditions, it assists farmers in
choosing the most suitable crops for their
specific geographic locations[13-15]. This pro-
motes optimal resource utilization and crop
yield improvement.
Figure 6.5. Blockchain-based marketplace.
Source: AUTHOR'S SELF GENERATED
3.5. Plant price prediction
The platform includes a sophisticated price
prediction functionality, leveraging cutting-
Figure 6.3. Crop recommendation. edge technologies to provide farmers with
Source: valuable insights into future crop prices.
AUTHOR'S SELF GENERATED
AgriTech: Empowering agriculture through integrated technology 53
Machine learning algorithms, specifically 3.6. Real-time weather monitoring
decision tree regression techniques, are and alerts
employed to analyze historical data and fore-
cast crop prices [18]. The platform offers real-time weather moni-
These algorithms consider factors such as toring, including temperature, humidity, and
market trends, historical price fluctuations, rainfall data. It provides farmers with essential
and relevant variables to generate accurate weather insights, aiding in informed decision-
price predictions. By harnessing the power making regarding planting and harvesting.
of machine learning and a robust database
system, AgriTech empowers farmers to make 3.7. Gamified plant growth
informed decisions regarding crop selection In the context of plant height estimation from
and pricing strategies. images, one crucial step involves color space
conversion. Typically, the input image is rep-
resented in the BGR (Blue, Green, Red) color
space. Still, for precise color-based analysis,
conversion to the HSV (Hue, Saturation,
Value) color space using OpenCV’s `cv2.cvt-
Color` is a valuable technique. In HSV, the
image is separated into three components:
Hue, capturing color information; Satura-
tion, reflecting color intensity; and Value, sig-
nifying brightness.
Ultimately, the plant’s height is estimated
Figure 6.6. Crop price prediction. by calculating the vertical extent of the larg-
Source: AUTHOR'S SELF GENERATED est contour concerning the image’s overall
Figure 6.7. Real-time weather monitoring.
Source:
AUTHOR'S SELF GENERATED
54 Innovations in Computational Intelligence: Trends and Technologies
height. This is achieved by determining the
ratio of the contour’s vertical extent to the
image’s height and subsequently scaling it
based on a constant base height, typically in
specified units such as centimeters or pixels.
3.8. Algorithm implementation
In the context of agricultural decision-mak-
ing, the project employs a robust model
known as the support vector machine (SVM)
with a radial basis function (RBF) kernel. Figure 6.10. Accuracy chart.
This machine-learning model is harnessed to
Source:
provide crop recommendations based on a
AUTHOR'S SELF GENERATED
comprehensive set of seven crucial features,
including nitrogen levels, phosphorus lev-
els, potassium levels, pH levels, temperature,
humidity, and rainfall.
SVM with the RBF kernel is particularly
valuable for this task as it excels in captur-
ing non-linear relationships between these
Figure 6.11. Crop recommendation flow
diagram.
Source: AUTHOR'S SELF GENERATED
features and crop suitability. By leveraging
this combination, the project offers farmers
Figure 6.8. Gamified plant growth. data-driven insights into selecting the most
Source: AUTHOR'S SELF GENERATED appropriate crops for their specific soil and
environmental conditions, ultimately enhanc-
ing agricultural productivity and resource
optimization.
As seen in this diagram, the first step in
the environmental assessment process is to
gather and analyze crucial data on soil and
environmental conditions. This information
is used to make well-informed choices regard-
ing land use and resource management. Once
the data is collected, a thorough data pre-pro-
cessing is carried out to guarantee the quality
of the data. Sophisticated machine learn-
ing algorithms are utilized in the analysis to
unveil hidden insights and patterns. To cre-
Figure 6.9. Confusion matrix. ate a model with enhanced accuracy for fore-
Source: casting environmental changes and assessing
AUTHOR'S SELF GENERATED
AgriTech: Empowering agriculture through integrated technology 55
interventions, these insights are put into
practice. At the end of the day, the model’s
outcomes are used to deliver well-informed
suggestions, guiding decisionmakers in the
implementation of sustainable measures for
responsible land and resource management.
In the project, plant disease prediction
plays a crucial role in identifying and miti-
gating diseases affecting various crops. The Figure 6.12. Visualizing plant growth using
project employs the MobileNetV2 model, a image processing.
state-of-the-art deep learning architecture, for Source:
AUTHOR'S SELF GENERATED
this task. With access to a diverse dataset com-
prising 38 different plant diseases, sourced
from the Kaggle dataset repository (acces- As seen in this diagram, visualizing plant
sible at https://www.kaggle.com/vipoooool/ growth through the implementation of image
newplantdiseases-dataset), MobileNetV2 is processing comprises a holistic procedure
trained to recognize and classify these diseases comprising multiple essential components.
accurately[19]. By leveraging deep learning The process commences with the collection
and the extensive dataset, the model can rap- of images that document the various devel-
idly diagnose plant diseases from images of opmental phases of the plants. Once these
leaves, offering a swift and effective solution raw images are obtained, various enhance-
to disease management and crop protection ment techniques are applied to enhance their
in agriculture. overall quality and clarity. After the enhance-
ment, the application of image segmenta-
3.8.1. Image processing tion is used to differentiate the plants from
To estimate the height of a plant from the background and measure their growth
images, several image processing techniques patterns. Data visualization is done using
are applied. Initially, blurring is employed the segmented data to create graphical rep-
to reduce noise in the input image, enhanc- resentations of plant growth metrics, which
ing subsequent processing robustness. The offer valuable insights. Ultimately, the usage
image is then converted from the BGR color of image processing techniques is crucial in
space to the HSV color space to facilitate every step of this process. These techniques
color-based analysis. In the HSV space, a are essential for extracting valuable informa-
green color range is defined to create a mask tion from plant images and enhancing our
isolating the plant’s green parts. Morpho- comprehension of plant development and
logical operations, including opening and health.
closing, further refine the mask, eliminating
noise and connecting object parts. Contour
detection identifies the plant’s boundaries,
4. Data Collection
with the largest contour selected as the plant. The datasets sourced from GitHub reposito-
This largest contour is drawn on the origi- ries provide essential resources for agricul-
nal image to visualize the region of interest. tural research.
Finally, height estimation is calculated by The [Plant Disease Detection Dataset]
assessing the vertical extent of the largest (https://github.com/manthan89-py/PlantDis-
contour relative to the image’s height and easeDesection) features a diverse collection
scaling it based on a predefined base height of labeled plant leaf images afflicted by vari-
value. ous diseases, facilitating the development and
56 Innovations in Computational Intelligence: Trends and Technologies
evaluation of machine learning models for impact on modern farming practices. The
automated disease detection. results from table 6.1 and table 6.2 can be sum-
The [Crop Price Prediction Dataset] -marized as follows:
(https://github.com/posi2/Crop-Price-Pre- 1. Improved crop recommendations:
diction) offers historical data on crop prices, Through the utilization of the SVM with
market trends, and relevant factors, serv- an RBF kernel, AgriTech successfully pro-
ing as a foundation for crop price forecast- vided personalized crop recommenda-
ing models that benefit both farmers and tions to farmers.
policymakers. 2. Efficient disease detection: The implemen-
Meanwhile, the [Crop Growth Stage tation of the MobileNetV2 deep learning
Dataset] (https://github.com/jahaniam/Crop- model for plant disease detection exhib-
GrowthStage) provides valuable insights into ited remarkable accuracy in identifying
the growth stages of different crops, aiding and classifying various plant diseases
precision agriculture and crop management from images of leaves. This functionality
research. offered rapid disease diagnosis, allowing
These datasets collectively empower agri- farmers to take timely measures for dis-
cultural advancements, from disease identi- ease management and crop protection.
fication to crop price prediction and growth 3. Enhanced user engagement: The coin-
monitoring. based reward system within the integrated
e-commerce marketplace effectively
5. Results and Discussion increased user engagement. Farmers
actively participated in the plant growth
The AgriTech platform has undergone rigor- mentoring process to earn coins, which
ous testing and real-world implementation, could be redeemed for agricultural prod-
AQ3: resulting in several key findings and outcomes ucts and tools.
Cita- that demonstrate its efficacy and potential 4. Real-time weather insights: AgriTech’s
tions for
both the real-time weather monitoring and alerts
tables are Table 6.1. Algorithm and accuracy provide farmers with essential weather
missing in
the text, Algorithm Accuracy Functionality data, including temperature, humid-
please ity, and rainfall. Farmers reported that
provide SVM + RBF 98.72% Crop
the Kernel Recommendation these insights significantly contributed
appro- to informed decision-making regarding
priate MobileNetV2 98.9% Disease Prediction
CITATION
citations
ADDED
planting and harvesting, resulting in more
Random Forest 97.1% Crop Price efficient resource utilization and crop
AQ4: Regression Prediction management.
Please
provide
Source:
AUTHOR'S SELF GENERATED 5. Secure and transparent transactions: The
source integration of the Ethereum blockchain
for all the
Tables Table 6.2. Comparison with previous work into AgriTech’s e-commerce marketplace
ensured secure and transparent transac-
Algorithm Previous Current tions. Users appreciated the added layer
Accuracy Accuracy
of trust and transparency in their mar-
SVM 98.2% [14] 98.72% ketplace interactions, which is especially
CNN 95.05% [20] 98.9% critical in the agricultural sector.
6. Improved communication: Agora SDK’s
Random Forest 98% [21] 98%
video conferencing capabilities facili-
Regression
tated real-time video calls and chat func-
Source: tionality, enhancing communication and
AUTHOR'S SELF GENERATED
AgriTech: Empowering agriculture through integrated technology 57
support for farmers. This improved com- India’s Livelihoods. 2023:107. Terborgh J.
munication enabled farmers to seek expert Preservation of natural diversity. BioScience.
consultations and collaborate effectively 2009;24:715–22.
with peers. [3] Reyana A, Kautish S, Karthik PS, Al-Baltah
7. Conversational AI: The integration of IA, Jasser MB, and Mohamed AW. Acceler-
ating crop yield: multisensor data fusion and
Alan AI for voice commands enhanced
machine learning for agriculture text classifi-
the platform’s accessibility, allowing users
cation. IEEE Access. 2023;11:20795–20805.
to interact with AgriTech using natural [4] Sundaram P, and Suparna K. Crop recom-
language. This feature proved particu- mendation system using machine learning. J
larly useful for users who may have lim- Eng Sci. 2023;14(09).
ited typing abilities or prefer voice-based [5] Sharma P, Dadheech P, and Senthil ASK. AI-
interactions. enabled crop recommendation system based
on soil and weather patterns. In: Artificial
Intelligence Tools and Technologies for
6. Conclusion Smart Farming and Agriculture Practices.
In conclusion, the AgriTech project represents IGI Global; 2023:184–199.
a pioneering and comprehensive agricultural [6] Baishya DR. International Journal for
technology platform that bridges the gap Research in Applied Science and Engineer-
ing Technology. 2020;8(IV).
between traditional farming practices and
[7] Ahmed U, Lin JCW, Srivastava G, and Dje-
the digital age. AgriTech empowers farm-
nouri Y. A nutrient recommendation system
ers at every stage of the agricultural process, for soil fertilization based on evolution-
from crop selection to growth management, ary computation. Comput Electron Agric.
disease detection, and price forecasting. The 2021;189:106407.
platform’s machine learning capabilities pro- [8] Green S. Reshaping the future of agriculture
vide personalized insights and data-driven through agri-tech innovations. Int Multidis-
recommendations, fostering enhanced pro- cip J Sci Technol Bus. 2023;2(02):1–4.
ductivity, sustainability, and resource effi- [9] Rathod ML, Shivaputra A, Umadevi H,
ciency. The implementation of MobileNetV2 Nagamani K, and Periyasamy S. Cloud com-
for plant disease detection and the integration puting and networking for SmartFarm Agr-
of the Agora SDK for real-time video commu- iTech. J Nanomater. 2022.
nication further demonstrate the platform’s [10] Ezzine I, and Benhlima L. Technology against
COVID-19: a blockchain-based framework
commitment to offering holistic support to
for data quality. In: 2020 6th IEEE Con-
farmers. As digital agriculture continues to
gress on Information Science and Technol-
evolve, AgriTech serves as a testament to the ogy (CiSt). IEEE; 2021:84–89.
potential of technology to revolutionize and [11] Yu J, Yin H, Xia X, Chen T, Li J, and Huang
improve farming practices, making a signifi- Z. Self-supervised learning for recommender
AQ5:
cant contribution to the well-being of farm- systems: a survey. IEEE Trans Knowl Data
Refer- ing communities worldwide. Eng. 2023.
ences are [12] Bandara P, Weerasooriya T, Ruchirawya T,
missing
Nanayakkara W, Dimantha M, and Paba-
cita-
tions in
References sara M. Crop recommendation system. Int J
the text. [1] Ngugi LC, Abelwahab M, and Abo-Zahhad Comput Appl. 2020;975:8887.
Please
ensure
M. Recent advances in image processing [13] Iqbal M, Kumar V, and Sharma VK. Krishi
that all techniques for automated leaf pest and dis- Portal: web-based farmer help assistance. Int
references ease recognition: a review. Inf Process Agric. J Adv Sci Technol. 2007;29:4783–4786.
are cited
2021;8(1):27–51. [14] Bakthavatchalam K, Karthik B, Thiru-
and in
sequential [2] Kanitkar A, and Wadekar I. Digital Agri- vengadam V, Muthal S, Jose D, Kotecha
order. culture: Is this the Future of ‘New’. State of K, and Varadarajan V. IoT framework for
58 Innovations in Computational Intelligence: Trends and Technologies
measurement and precision agriculture: International Conference of Safe Produc-
predicting the crop using machine learning tion and Informatization (IICSPI). IEEE;
algorithms. Technologies. 2022;10(1):13. 2020:462–465.
[15] Kang M, Wang X, Wang H, Hua J, de Reffye [18] Mulla SA, and Quadri SA. Crop-yield and
P, and Wang FY. The development of AgriV- price forecasting using machine learning. Int J
erse: past, present, and future. IEEE Trans Anal Exp Modal Anal. 2020;12:1731–1737.
Syst Man Cybern Syst. 2023. [19] Kosamkar PK, Kulkarni VY, Mantri K,
[16] Fathi M, Haghi Kashani M, Jameii SM, and Rudrawar S, Salmpuria S, and Gadekar N.
Mahdipour E. Big data analytics in weather Leaf disease detection and recommenda-
forecasting: a systematic review. Arch Com- tion of pesticides using convolution neu-
put Methods Eng. 2022;29(2):1247–1275. ral network. In: 2018 Fourth International
[17] Su Z, Wang H, Wang H, and Shi X. A finan- Conference on Computing Communica-
cial data security sharing solution based tion Control and Automation (ICCUBEA).
on blockchain technology and proxy re- IEEE; 2018:1–4.
encryption technology. In: 2020 IEEE 3rd
7 Online and offline learning in
universities: A student’s dilemma
Ruchi Agarwala and Rahul Pradhanb
Department of Computer Engineering and Applications, GLA University, Mathura, Uttar Pradesh,
India
Abstract: The pandemic, heatwave brought life to a standstill, if not for older but for young ones and students
it brings things to a standstill, presenting a significant obstacle to humanity. However, online learning offers
flexibility, interaction, and cooperation. This article aims to compare and contrast the results of student surveys
and explore the differences between live online instruction and conventional classroom instruction, as well
as the future of education in universities and colleges. The study was conducted at several universities in the
Mathura region of Uttar Pradesh, India, with 150 participants selected from those enrolled in online and offline
university courses, aged between 16 and 40 years. The students were then asked to complete a detailed question-
naire. 58.95% of students mostly agreed with the question in the survey, while 36.69% disagreed. As per the
survey analysis, online learning seems to be more effective. It is concluded that traditional classroom teaching
has advantages for students, saves them time, and improves their academic performance in the current environ-
ment of the COVID-19 epidemic. Higher education institutions must collaborate to make these initiatives effec-
tive, especially by decentralizing educational teleconferences and granting access to surrounding institutions to
ensure that all students have reliable internet connectivity.
Keywords: COVID-19, E-learning, offline learning, online learning, students, survey
1. Introduction technology will revolutionize education,
teaching, and learning Lee (2006); Romeo
E-learning designs, delivers and manages for-
(2006). Integration of educational technolo-
mal and informal learning and knowledge
gies presents learning obstacles and problems.
exchange anywhere, anytime. Some courses
The COVID-19 pandemic has stopped
are delivered fully online, while others are
hybrid. Online learning offers flexible tim- most human activities, posing a huge obsta-
ings, self-paced, more interactivity, and sup- cle to humanity. Although in this era where
port Gedera, Williams, and Wright (2013). computers and the internet reach each indi-
With e-expansion, teachers and students can vidual’s pocket, this tiny microbe is devas-
AQ1: As
construct knowledge and enhance learning tating Akbar, Rashid, et al. (2018). Even the
per style,
the refer- outside the classroom. According to Shafaati school system was closed as everyone was
ence list
et al., 2022 Shafaati, Valizade, Rahimi, and affected. These stalled face-to-face classes dis-
and its
citation Panahi (2022), 4,600,000 college students in turbed student training. Access to technology
should be the US take at least one online class. By 2014, in rural areas causes barriers to learning. The
num-
bered, this figure will rise to 18,650,000 Shafaati et research literature validates students’ online
please al. (2022). Despite learning of the popular- needs and challenges Aboagye, Yawson, and
check and
confirm
ity of e-Commerce Bell and Federman (2013), Appiah (2021); Chase et al. (2018a); Chung,
for positive results are not guaranteed Alexander Subramaniam, and Dass (2020); Lorenzo
clarity. (2001). Some researchers are unsure whether (2017). Some people think differently about
CHECKED AND CONFIRMED
a
[email protected]; [email protected]
DOI: 10.1201/9781003606635-7
60 Innovations in Computational Intelligence: Trends and Technologies
Internet connections. In-efficient internet
facilities in universities do not affect internet
use Apuke and Iyendo (2018). Another study
highlighted the correlation between inter-
net connectivity and online learning in rural
areas Ahmed, Khan, Faisal, and Khan (2017).
In another survey, the majority of respond-
ents were ready for e-learning Muthuprasad,
Aiswarya, Aditya, and Jha (2021). In another
paper, Hossain Hossain and Rahman (2017)
and colleagues suggested in 2017 that uni- Figure 7.3. Subject distribution according to
versities provide students with internet facili- education qualification.
ties and an environment. Tarimo Tarimo and Source:
Kavishe (2017) and colleagues revealed in
2017 that 82% of children use the Internet
to go to school. A study showed negative atti-
tudes toward online learning management Learning tools or equipment are also
systems Serhan (2020). Taking into account important in this study. Although we live in
these concepts and facts. Internet connec- the age of technology, not everyone has. A
AQ2: tivity is a problem for students, faculty, and study in South Asia showed that respondents
Please institutions. had computers and internet at home and at
provide
in-text
college. In another study, the students owned
citation various smartphones. This is a possibility as
for Figure demand for phones increases and prices are
1 to 3
competitive. Research shows that students
rely on smartphones to learn and surf the
Internet. They found no correlation between
smartphone ownership and nomophobia
AQ3: in college students Gezgin (2017). Another
Please study found that mobile learning devices can
provide
improve student learning efficiency Chase et
source Figure 7.1. Subject distribution according to the
for all the ages. al. (2018b). The biggest problem is the resist-
figures ance of students to blended online learning.
Source:
AUTHOR'S SELF GENERATED In the Philippines, Surveys revealed mixed
sentiments toward internet connectivity and
device usage for online education. Lack of
reliable Internet connection is a barrier to
learning. Fabito, Trillanes, and Sarmiento
(2020). Authors noted that few pupils had
internet access, preventing them from using
the e-learning platform Casillano (2019).
Poor students lack laptops, desktops, and
Internet access, according to another sur-
vey Cleofas and Rocha (2021). They found
Figure 7.2. Subject distribution according to the mobile devices can be used for learning Jin
gender. and Junio-Sabio (2018). Another study found
Source: that owning devices plays a crucial role in
AUTHOR'S SELF GENERATED
Online and offline learning in universities: A student’s dilemma 61
learning readiness Estira (2020). Another and Solomon (2021). Atika Qazi et al.2020
state university survey found that students are Qazi et al. (2020) 320 students participated
ready for online lessons, but renting comput- in study about learning in different modes
ers and the Internet in cafes is a burden Yra, during the COVID-19 pandemic. The study
Castillo, Bautista, Camayang, and Camayang concluded that online education is the only
(2020). This study assesses students’ internet solution and effective way Qazi et al. (2020).
connection and learning devices for online or The vast majority (70.4%) of the survey
flexible learning. The study will show school respondents said they would advise use of
administrators how students and staff will e-learning material. Online learning scores
modify teaching-learning procedures. The 4/5 for effectiveness, while offline learning
result will be used to institutionalize flexible scores 3/5 Yang et al. (2022).
learning in the future.
3. Material and Method
2. Literature Review
Favale et al. 2020 Favale, Soro, Trevisan, 3.1. Study design
Drago, and Mellia (2020), The infrastructure The online survey is used to collect data and
for the internet is under pressure as a result respondents’ responses. A questionnaire was
of this sudden spike in usage. E-learning designed to capture respondents’ opinions.
tools like Microsoft Teams and Zoom were The purpose of this study is to see the mood of
quickly adopted. Students are more aware of students at the university level about learning
online learning. E-learning, however, may be techniques adopted by their higher education
extremely challenging for teachers Favale et institutions during the pandemic covid-19 in
al. (2020) who not habitual of making pow- Braj area of Uttar Pradesh which constitutes
erpoint ppts. Agra, Mathura, and some parts of Aligarh.
Padmalini Singh et al. 2021 Singh et al.
(2021), conducted a study at India, Indo- Table 7.1. Participants distribution on the basis
nesia, and Malaysia on learning with 100 of Age
respondents. they compare online and offline
Age Frequency Percentage AQ4:
learning. their study shows traditional teach- Please
ing is more effective than online teaching 16–20 40 26 provide
21–25 46 31 in-text
Singh et al. (2021). Elyassi et al. 2021 Elyassi citation
et al. (2021), various changes in the psychic 26–30 25 16.6 for Tables
of students due to lockdown and pandemics 31–35 19 13
1 to 3.
Elyassi et al. (2021). >35 20 13.3 AQ5:
Biswas 2021, study India for learning. they Please
Total 150 100 provide
concluded that the face-to-face offline class- source
room is better than online resources can be Source: for all the
AUTHOR'S SELF GENERATED
used as a supplement. After life returns to Tables
normal in the post-COVID-19 era, we need Table 7.2. Participants distribution on the basis
to return to traditional classroom teach- of Gender
ing Dalai, Dey, Chatterjee, Chakravorti, and Gender Frequency Percentage
Bhattacharya (2014). Sheba et al. 2021 sur- Female 85 56.6
veyed 450 students for effectiveness of online
Male 65 43.3
study material. students found the mate-
rial was effective when used with anima- Total 150 100
tions, PPT, and visuals Darius, Gundabattini, Source:
AUTHOR'S SELF GENERATED
62 Innovations in Computational Intelligence: Trends and Technologies
3.2. Investigation Participants According to the data above, as much as
36.6% of respondents disagreed compared to
For this study, we have selected 150 partici-
63.3% respondents chose to agree who chose
pants according to age between 16 and 40
to agree that “During the COVID-19 pandemic,
years, sex, and educational qualification reg-
students prefer online learning”. “Online Learn-
istered at universities for online and offline
ing saves time while studying” has a tie where
courses.
50% says they agree to the statement while
50% disagree to the statement. For the third
3.3. Investigation content question, 56.6% agree to be satisfied with the
Feedback on the learning material provided, leaning outcomes in online mode. Figure 7.4
instruction techniques, usage of ICT, and indicates the Student feedback graph.
mode of teaching was collected through
questionnaires. 4. Discussion
Table 7.3. Participant’s distribution based on Covid-19 impacted education and popular-
their educational background ized online education. This study aimed to find
Education
Frequency Percentage
Qualification
Bachelor’s Degree 45 30
Master 65 43.3
PhD 19 12.6
Others 21 14 Figure 7.4. Feedback– Distribution of
Total 150 100 Agreement and disagreement on above issues.
Source: AUTHOR'S SELF GENERATED Source:
AUTHOR'S SELF GENERATED
Table 7.4. Responses in favor of online mode of learning
S.No. Feedback Agree Disagree Agree (%) Disagree (%)
1 Online learning is preferred during pandemic 95 55 63 36.66
CoVID-19
2 Online Learning saves time while studying. 75 75 50 50
3 Satisfaction with the learning effect 85 65 56.6 43.33
4 Online mode of learning is beneficial for 100 50 66.66 33.33
students
5 Students prefer online lectures to physical 105 45 70 30
lectures.
6 Online class materials were useful and accurate 96 54 64 36
7 Exams were based on material covered in the 87 63 58 42
assignment and lectures
8 Communicated clearly 90 60 60 40
9 The teacher showed an interest in helping 87 63 58 42
students learn
10 The method of evaluating my work were fair 65 85 43.33 56.6
Total 58.95 36.69
Source:
AUTHOR'S SELF GENERATED
Online and offline learning in universities: A student’s dilemma 63
the effectiveness of new techniques adopted internet struggling province in the philip-
by various higher educational institutes and pines. Indian Journal of Science and Tech-
government guidelines during COVID-19 for nology, 12 (10), 1–4.
continued learning. The study was conducted [8] Chase, T. J., Julius, A., Chandan, J. S., Powell,
in the Indian state of Uttar Pradesh in the Brij E., Hall, C. S., Phillips, B. L., and Fernando,
B. (2018a). Mobile learning in medicine: an
region. A total of 150 participants partici-
evaluation of attitudes and behaviours of
pated and responded to questions. We con-
medical students. BMC medical education,
clude that in our study online education and 18 (1), 1–8.
techniques are effective and most students are [9] Chase, T. J., Julius, A., Chandan, J. S., Powell,
satisfied with it. E., Hall, C. S., Phillips, B. L., . . . Fernando, B.
(2018b). Mobile learning in medicine: an eval-
uation of attitudes and behaviours of medical
5. Conclusion students. BMC medical education, 18 (1), 1–8.
We can conclude based on the responses col- [10] Chung, E., Subramaniam, G., and Dass, L.
lected, upon analyzing the responses, online C. (2020). Online learning readiness among
learning could be adopted. Online learning university students in malaysia amidst covid-
19. Asian Journal of University Education,
could be more effective if devices and inter-
16 (2), 45–58.
net costs could be subsidized and reach rural
[11] Cleofas, J. V., and Rocha, I. C. N. (2021).
areas. Demographic, gadget and internet profiles
as determinants of disease and consequence
related covid-19 anxiety among filipino col-
References lege students. Education and Information
[1] Aboagye, E., Yawson, J. A., and Appiah, K. Technologies, 26 (6), 6771–6786.
N. (2021). Covid-19 and e-learning: The [12] Dalai, S., Dey, D., Chatterjee, B., Chakra-
challenges of students in tertiary institutions. vorti, S., and Bhattacharya, K. (2014). Cross-
Social Education Research, 1–8. spectrum analysis-based scheme for multiple
[2] Ahmed, S. S., Khan, E., Faisal, M., and Khan, power quality disturbance sensing device.
S. (2017). The potential and challenges of IEEE Sensors Journal, 15 (7), 3989–3997.
moocs in pakistan: a perspective of students [13] Darius, P. S. H., Gundabattini, E., and Solo-
and faculty. Asian Association of Open Uni- mon, D. G. (2021). A survey on the effective-
versities Journal, 12 (1), 94–105. ness of online teaching–learning methods for
[3] Akbar, M. A., Rashid, M. M., et al. (2018). university and college students. Journal of
Technology based learning system in inter- The Institution of Engineers (India): Series
net of things (iot) education. In 2018 7th B, 102 (6), 1325–1334.
international conference on computer and [14] Elyassi, H., et al. (2021). Economics of the
communication engineering (iccce) (pp. financial crisis: any lessons for the pandemic
192–197). downturn and beyond? Contemporary Eco-
[4] Alexander, S. (2001). E-learning develop- nomics, 15 (1), 100–121.
ments and experiences. Education+ Train- [15] Estira, K. L. A. (2020). Online distance
ing, 43 (4/5), 240–248. learning readiness of business administra-
[5] Apuke, O. D., and Iyendo, T. O. (2018). Uni- tion students in one state university in the
versity students’ usage of the internet resources philippines. Journal of Critical Reviews, 7
for research and learning: forms of access and (12), 826–832.
perceptions of utility. Heliyon, 4 (12). [16] Fabito, B. S., Trillanes, A. O., and Sarm-
[6] Bell, B. S., and Federman, J. E. (2013). iento, J. R. (2020). Barriers and challenges
E-learning in postsecondary education. The of computing students in an online learn-
future of children, 165–185. ing environment: Insights from one private
[7] Casillano, N. F. B. (2019). Challenges of university in the philippines. arXiv preprint
implementing an e-learning platform in an arXiv:2012.02121.
64 Innovations in Computational Intelligence: Trends and Technologies
[17] Favale, T., Soro, F., Trevisan, M., Drago, I., Conventional to online education during
and Mellia, M. (2020). Campus traffic and covid-19 pandemic: Do develop and under-
e-learning during covid-19 pandemic. Com- developed nations cope alike. Children and
puter networks, 176, 107290. Youth Services Review, 119, 105582.
[18] Gedera, D., Williams, P. J., and Wright, N. [26] Romeo, G. (2006). Engage, empower, enable:
(2013). An analysis of moodle in facilitating Developing a shared vision for technology in
asynchronous activities in a fully online uni- education. Engaged learning with emerging
versity course. International Journal of Sci- technologies, 149–175.
ence and Applied Information Technology [27] Serhan, D. (2020). Transitioning from face-
(IJSAIT), 2 (2), 6–10. to-face to remote learning: Students’ atti-
[19] Gezgin, D. M. (2017). Exploring the influ- tudes and perceptions of using zoom during
ence of the patterns of mobile internet use covid-19 pandemic. International Journal of
on university students’nomophobia levels. Technology in Education and Science, 4 (4),
European Journal of Education Studies. 335–342.
[20] Hossain, M. A., and Rahman, M. H. (2017). [28] Shafaati, A., Valizade, R., Rahimi, A., and
Comparative study of internet usage among Panahi, A. (2022). The role of smart urban
university students: A study of the university develop- ment in the development of the
of dhaka, bangladesh. European Scientific dilapidated areas of tabriz city: A case study
Journal, 13 (34), 134–150. of 8th district of tabriz. Geography and
[21] Jin, W., and Junio-Sabio, C. (2018). Poten- Environmental Planning, 33 (1), 113–132.
tial use of mobile devices in selected public [29] Singh, P., Sinha, R., Koay, W. L., Teoh, K.
senior high schools in the city of manila phil- B., Nayak, P., Lim, C. H., . . . Aryani, D. N.
ippines. International Journal of Learning, (2021). A comparative study on effective-
Teaching and Educational Research, 17 (4), ness of online and offline learning in higher
102–114. education. International Journal of Tourism
[22] Lee, K. T. (2006). Creating ict-enriched and Hospitality in Asia Pasific (IJTHAP), 4
learner-centred environments: Myths, gaps (3), 102– 114.
and challenges. Engaged learning with [30] Tarimo, R., and Kavishe, G. (2017). Inter-
emerging technologies, 203–223. net access and usage by secondary school
[23] Lorenzo, A. R. (2017). Comparative study students in morogoro municipality, tanza-
on the performance of bachelor of secondary nia. International Journal of Education and
education (bse) students in educational tech- Development using ICT, 13 (2).
nology using blended learning strategy and [31] Yang, S., Tan, C., Madsen, D. Ø., Xiang, H.,
traditional face-to-face instruction. Turkish Li, Y., Khan, I., and Choi, B. J. (2022). Com-
Online Journal of Educational Technology- parative analysis of routing schemes based
TOJET, 16 (3), 36–46. on machine learning. Mobile Information
[24] Muthuprasad, T., Aiswarya, S., Aditya, K. S., Systems, 2022.
and Jha, G. K. (2021). Students’ perception [32] Yra, J. F. P., Castillo, R., Bautista, R. G.,
and preference for online education in india Camayang, J. G., and Camayang, A. G. G.
during covid-19 pandemic. Social sciences (2020). Students’ online learning readiness
and humanities open, 3 (1), 100101. and internet connectivity: Bases for the cus-
[25] Qazi, A., Naseer, K., Qazi, J., AlSalman, H., tomization of qsu e-aral. American Journal
Naseem, U., Yang, S., . . . Gumaei, A. (2020). of Educational Research, 8 (11), 878–884.
8 Strengthening UAV network security:
Advanced authentication-based
intrusion detection systems
Altaf C1,a, Shanila Mahreen2,b, C Atheeq3,c, and Shaista Khanam2,d
1
Department of ECE, Lords Institute of Engineering and Technical, Hyderabad, India
2
Department of ECE, Nawab Shah Alam College of Engineering and Technical, Hyderabad, India
3
Department of CSE, GITAM University, Hyderabad, India
Abstract: This article delves into contemporary Intrusion Detection Systems (IDS) designed for Unmanned
Aerial Vehicle (UAV) networks, emphasizing authentication-based approaches. With UAVs playing pivotal roles
across sectors, securing their communications is paramount. Traditional IDS often falls short in UAV scenarios,
necessitating innovative strategies. The proposed IDS employs advanced authentication mechanisms, including
cryptography, biometrics, and dynamic key generation. Cryptographic techniques ensure secure key exchange,
while biometric authentication ties UAV activities to verified identities. Dynamic key generation, bolstered by
chaotic maps, enhances security, making it computationally challenging for attackers. Anomaly detection and
machine learning contribute to the IDS’s adaptability, distinguishing normal and malicious UAV behavior.
Simulated evaluations using tools like ns3 quantify system performance, with metrics like Packet Delivery
Ratio, Delay, and Overhead providing insights. This authentication-based IDS presents a robust solution for
safeguarding UAV network integrity, addressing unique security challenges in diverse operational scenarios.
Keywords: attack, UAV, authentication, chaotic maps, network, IDS
1. Introduction offering real-time data acquisition from van-
tage points that might be inaccessible or haz-
Unmanned Aerial Vehicle (UAV) networks ardous for human operators. This capability
have emerged as transformative technologies has proven invaluable in enhancing security,
with multifaceted applications across vari- border control, and monitoring large-scale
ous industries. These aerial platforms, com- events. In agriculture, UAVs equipped with
monly known as drones, are equipped with specialized sensors contribute to precision
diverse sensors and communication systems, farming by assessing crop health, optimizing
enabling them to perform tasks that were irrigation, and facilitating crop management
once deemed challenging or impractical for [2]. This not only enhances productivity but
traditional methods [1]. The significance of also minimizes resource utilization, aligning
UAV networks lies in their ability to offer with sustainable agricultural practices.
cost-effective, efficient, and flexible solutions UAV networks play a pivotal role in dis-
in domains such as surveillance, agriculture, aster management, where their swift deploy-
disaster management, and infrastructure ment can provide rapid aerial assessments
inspection. of affected areas, aiding in search and res-
In the realm of surveillance, UAVs pro- cue missions, damage evaluation, and plan-
vide a dynamic and adaptable approach, ning efficient relief efforts. Furthermore, in
[email protected]; [email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-8
66 Innovations in Computational Intelligence: Trends and Technologies
infrastructure inspection and maintenance, sensitive information but also in ensuring the
UAVs offer a safer alternative to manual uninterrupted and reliable operation of UAVs
inspection, particularly in complex and ele- across various applications. UAVs heavily rely
vated structures [3]. They can capture high- on seamless communication for tasks such as
resolution images and data, facilitating the navigation, data acquisition, and command
early detection of potential issues and reduc- execution. Any compromise in communica-
ing the risks associated with traditional tion integrity can lead to severe consequences,
inspection methods. The versatility of UAV including loss of control, unauthorized access
networks is underscored by their adaptabil- to mission-critical data, and potential threats
ity to diverse environments and tasks. Their to public safety.
compact size, agility, and ability to operate One primary consideration in securing
in challenging terrains make them indispen- UAV communication is the protection of
sable tools in scenarios where conventional mission data. In applications like surveil-
methods may fall short. However, this tech- lance, infrastructure inspection, or disaster
nological revolution is not without its chal- response, the data collected by UAVs often
lenges, particularly concerning the security contains sensitive and confidential informa-
and integrity of the communication systems tion [6]. Unauthorized interception or tam-
within UAV networks. pering of this data not only compromises
As UAVs rely heavily on wireless com- the mission’s objectives but can also lead to
munication for control, navigation, and data privacy breaches and misuse of information.
transfer, ensuring the confidentiality, integrity, Therefore, robust security measures are essen-
and authenticity of the transmitted informa- tial to safeguard the confidentiality and integ-
tion is paramount. Unauthorized access, data rity of the data exchanged between UAVs and
tampering, and interception pose significant ground stations.
threats to the secure operation of UAV net- Moreover, securing communication is
works [4]. Therefore, implementing robust integral to maintaining control over UAV
security measures, including Intrusion Detec- operations. In scenarios where UAVs operate
tion Systems (IDS), becomes imperative to autonomously or semi-autonomously, secure
safeguard these networks against evolving communication ensures that commands from
cyber threats. ground control stations reach the UAV accu-
In the face of these challenges, the inte- rately [7]. Any interference or manipulation
gration of authentication-based IDS emerges in these commands could result in the UAV
as a promising solution. By implementing deviating from its intended path or executing
advanced authentication mechanisms, such as unintended actions, posing risks to both the
dynamic key generation with chaotic maps, UAV and its surroundings. Security breaches
UAV networks can enhance their resistance in communication can also open avenues for
against malicious attacks. This article delves malicious actors to take unauthorized control
into the intricacies of these authentication- of UAVs, leading to potential safety hazards
based approaches, exploring their potential and misuse of the technology [8].
to fortify the security posture of UAV net- Beyond the immediate operational con-
works and ensure the continued advancement cerns, the importance of secure communica-
of this transformative technology [5]. tion extends to the public’s perception and
Securing communication in Unmanned regulatory compliance. As UAVs become
Aerial Vehicle (UAV) operations is of para- more integrated into civilian airspace and
mount importance, given the critical nature daily life, ensuring their secure operation is
of the tasks these aerial platforms undertake. crucial to building public trust and meeting
The significance lies not only in protecting regulatory standards. The potential for UAVs
Strengthening UAV network security: Advanced authentication-based intrusion detection systems 67
to be used for nefarious purposes, such as bandwidth management become paramount
spying or carrying out malicious attacks, fur- in ensuring seamless and secure UAV oper-
ther underscores the need for robust commu- ations [12]. The constraints on onboard
nication security measures. resources, including processing power and
energy, further complicate the implementa-
2. Literature Review tion of sophisticated security measures.
Securing UAV networks, therefore, is a
2.1. Discussion on the vulnerabilities multifaceted endeavor. It requires not only
addressing specific technical vulnerabilities
and challenges faced by UAV
but also considering the operational con-
networks text in which UAVs operate. The develop-
Unmanned Aerial Vehicle (UAV) networks, ment of adaptive communication protocols,
representing a technological marvel, bring encryption techniques resilient to signal
forth a spectrum of vulnerabilities and chal- interference, and real-time threat detection
lenges intrinsic to their operational character- mechanisms becomes imperative. Addition-
istics. The wireless communication paradigm, ally, collaborative efforts between industry,
a cornerstone of UAV networks, introduces regulators, and cybersecurity experts are
susceptibility to diverse cyber threats [9]. essential to establish standardized security
The reliance on radio frequency signals ren- practices that can be universally applied to
ders communication channels vulnerable to UAV networks [13,14].
interception, eavesdropping, and jamming
attempts. Given the varied applications of 2.2. Overview of traditional IDS
UAVs, spanning from surveillance to delivery and their limitations in UAV
services, the potential impact of a compro-
mised communication link is substantial. scenarios
Furthermore, the deployment of UAVs in Traditional Intrusion Detection Systems (IDS),
dynamic and diverse environments adds lay- stalwarts in safeguarding network integrity,
ers of complexity to their security landscape. encounter distinct challenges when applied
Urban areas, characterized by high signal to the dynamic and resource-constrained
interference and potential hacking threats, domain of UAV networks. While traditional
pose different challenges than remote regions IDS have proven effective in terrestrial set-
where the communication infrastructure may tings, their seamless integration into UAV
be limited. This diversity demands a nuanced operations requires a nuanced understanding
understanding of the threat landscape to of the unique characteristics defining aerial
devise adaptive security mechanisms. Cyber- missions.
physical threats, such as GPS spoofing or A fundamental limitation arises from the
manipulation of sensor data, underscore the reliance of traditional IDS on predefined pat-
need for robust security protocols to main- terns and signatures for identifying malicious
tain the integrity of UAV operations [10,11]. activities. In the fluid and dynamic airspace
Operational challenges intertwine with where UAVs operate, the ability to adapt to
technical vulnerabilities, shaping the unique emerging threats not captured by known pat-
security concerns of UAV networks. The terns becomes critical. Traditional IDS may
dynamic nature of UAV missions, often struggle to keep pace with rapidly evolving
involving rapid changes in altitude, speed, threats that do not conform to pre-established
and environmental conditions, demands signatures, necessitating the development of
communication links that can swiftly adapt. anomaly detection approaches tailored to
Issues such as signal reliability, latency, and UAV environments [15].
68 Innovations in Computational Intelligence: Trends and Technologies
The resource constraints inherent in UAVs Securing Data Transmission: Authentica-
pose another significant challenge. Traditional tion plays a pivotal role in securing the trans-
IDS often assume a stable network environ- mission of sensitive data between UAVs and
ment with ample computational resources. ground stations. By confirming the identity of
In contrast, UAVs operate with limited communicating entities, authentication miti-
onboard processing capabilities and energy gates the risk of data interception or manipu-
reservoirs. Implementing resource-intensive lation during transmission.
IDS solutions may compromise the overall Establishing Trust in UAV Operations: Trust
performance and endurance of UAVs [16]. is fundamental in UAV operations, especially
Achieving an optimal balance between the in applications like surveillance, reconnais-
computational demands of intrusion detec- sance, and critical infrastructure monitoring.
tion and the resource constraints of UAVs Authentication mechanisms build trust by ver-
becomes a paramount consideration. ifying the legitimacy of UAVs within the net-
Moreover, the decentralized and autono- work, fostering confidence in the reliability of
mous nature of UAV operations further com- data and commands exchanged.
plicates the application of traditional IDS. In Preventing Spoofing and Impersonation:
scenarios where continuous connectivity is UAVs are susceptible to various cyber threats,
not guaranteed, IDS solutions need to adapt to including spoofing and impersonation. Effec-
intermittent communication links and oper- tive authentication mechanisms act as a deter-
ate in a standalone fashion. This demands a rent against these threats, ensuring that UAVs
paradigm shift in IDS design, moving away only respond to legitimate commands and
from centralized models to distributed archi- communications from authenticated sources.
tectures that align with the decentralized
nature of UAV networks. 3.1.2. Challenges in UAV
authentication
3. Methodology While authentication is paramount, imple-
menting effective authentication in UAV
3.1. Authentication mechanisms in networks presents unique challenges due to
UAV networks the operational characteristics of these aerial
systems.
3.1.1. Importance of authentication
Limited Computational Resources: UAVs
in UAV networks often operate with constrained computa-
Authentication is a cornerstone of ensuring the tional resources, necessitating lightweight yet
security and integrity of communications within secure authentication protocols. Traditional
UAV networks. In the context of Unmanned methods designed for resource-rich environ-
Aerial Vehicles (UAVs), where these autono- ments may prove impractical, requiring the
mous systems often operate in sensitive and development of specialized solutions tailored
mission-critical environments, the significance to UAV constraints.
of robust authentication cannot be overstated. Dynamic and Ad-Hoc Network Topol-
Protection Against Unauthorized Access: ogies: UAV networks frequently exhibit
Authentication safeguards UAV networks dynamic and ad-hoc topologies, making it
from unauthorized access attempts. Ensuring challenging to establish and maintain con-
that only authenticated and authorized enti- tinuous authentication. Traditional methods
ties can participate in communication pro- relying on stable network structures encoun-
cesses is essential for preventing malicious ter difficulties in adapting to the dynamic
actors from compromising the system. nature of UAV communications.
Strengthening UAV network security: Advanced authentication-based intrusion detection systems 69
Vulnerability to Jamming and Interference: communicating parties to enable them to
UAV communication links are susceptible encrypt and decrypt messages. In UAV net-
to jamming and electromagnetic interfer- works, where data integrity is paramount,
ence. Authentication methods must contend robust key exchange protocols are essential.
with these challenges to ensure the reliable Biometric Authentication: Explore the
exchange of authentication signals between integration of biometric authentication to
UAVs and ground stations. establish a strong connection between UAV
Harsh Environmental Conditions: UAVs activities and verified identities.
operate in diverse environmental conditions, Step 1: Fingerprint = {P1, P2,..., Pn}: Col-
including extreme temperatures and varying lect a set of n fingerprint data points.
weather patterns. Authentication hardware Step 2: Template = f(Fingerprint): Process
and protocols must be resilient to these con- the fingerprint data to create a biometric
ditions to guarantee continuous and reliable template.
performance. Step 3: Database = {Template1, Template2,
In overcoming these challenges, the ..., Templatem}: Store the template securely in
development of authentication mechanisms a database.
tailored for UAV networks is imperative. Step 4: CapturedData=f(OperatorFingerp
Solutions should strike a balance between rint): Capture the operator’s biometric data
robust security measures and adaptability during UAV operation.
to the unique operational characteristics of Step 5: Match = g(CapturedData,Database):
UAVs. Match the captured data with stored
templates.
Step 6: If match successful, authorize
3.2. Advanced authentication
access: Grant access to the UAV operator.
techniques Dynamic Key Generation with Chaotic
Cryptography and Key Exchange: Cryptogra- Maps: Explain the use of dynamic key gen-
phy and key exchange form the cornerstone eration, enhanced by chaotic maps, to bolster
of secure communication systems, especially the security of UAV communications.
in the realm of Unmanned Aerial Vehicle Step 1: InitialConditions = {IC1, IC2, ...,
(UAV) networks where sensitive informa- ICk}: Set initial conditions for the chaotic map.
tion is transmitted. The primary objective Step 2: xn+1 = f(xn, InitialConditions): Iter-
of this process is to establish a secure chan- ate the chaotic map to generate a sequence.
nel for the exchange of cryptographic keys, Step 3: Kn = g(xn): Generate a crypto-
ensuring confidentiality and integrity during graphic key from the chaotic map.
communication. Cryptography involves the Step 4: Kn+1 = h(Kn): Update the crypto-
use of mathematical algorithms and cryp- graphic key for the next iteration.
tographic techniques to encode information Step 5: KeyDistribution = {K1, K2, ..., Km}:
in such a way that only authorized entities Distribute the generated keys securely.
can decipher it. In UAV networks, crypto- Step 6: EncryptedData = E(OriginalData,
graphic mechanisms play a pivotal role in Kn): Use the key for encryption/decryption.
securing data transmitted between UAVs and Step 7: Synchronization = i(Map1, Map2):
ground stations. The emphasis is on prevent- Synchronize chaotic maps for consistency.
ing unauthorized access, data tampering, and Step 8: SecurityAnalysis = j(Data,
eavesdropping. KeyDistribution): Perform security analysis on
Key exchange is a critical aspect of data and key distribution.
secure communication. It involves the secure The Biometric Authentication Algorithm
sharing of cryptographic keys between in UAV networks operates by first acquiring
70 Innovations in Computational Intelligence: Trends and Technologies
and processing unique biometric data, specifi- Biometric authentication, a key element in
cally fingerprints, to create templates stored UAV network security, establishes a robust
in a secure database. During UAV opera- connection between UAV activities and veri-
tion initiation, real-time fingerprint data is fied identities. The algorithm initiates by
captured, matched against stored templates, collecting a set of n fingerprint data points
and upon successful authentication, access (Fingerprint = {P1, P2, ..., Pn}). These data are
authorization is granted to the identified then processed to create a biometric template
operator, ensuring only authorized individu- (Template = f(Fingerprint)), securely stored
als control the UAV. This robust authentica- in a database (Database = {Template1, Tem-
tion mechanism links UAV activities with plate2, ..., Templatem}). During UAV operation,
verified operator identities, enhancing overall the operator’s biometric data (Captured-
network security. Data = f(OperatorFingerprint)) is captured
On the other hand, the Dynamic Key and matched against stored templates. If
Generation with Chaotic Maps Algorithm the match is successful, access is authorized,
focuses on fortifying UAV communications ensuring that only authorized operators con-
through the dynamic generation of crypto- trol the UAV. This process enhances overall
graphic keys. Utilizing chaotic maps with network security by linking UAV activities to
unpredictable iterations based on initial verified operator identities.
conditions, the algorithm generates crypto- The Dynamic Key Generation with Cha-
graphic keys that are dynamically updated. otic Maps Algorithm fortifies UAV commu-
These keys are securely distributed across nications through the dynamic generation
the UAV network, serving as the foundation of cryptographic keys. The algorithm begins
for encryption and decryption processes. by setting initial conditions for a chaotic
Chaotic map synchronization ensures con- map (InitialConditions = {IC1, IC2, ..., ICk}).
sistency among communicating entities, and Through iterations, a sequence is generated
a comprehensive security analysis evaluates (xn+1 = f(xn, InitialConditions)), and crypto-
the algorithm’s effectiveness in safeguard- graphic keys (Kn = g(xn)) are derived. The keys
ing UAV communications. This dynamic key are dynamically updated for subsequent iter-
generation approach adds an extra layer of ations (Kn+1 = h(Kn)) and securely distributed
security, mitigating the risk of unauthorized across the UAV network (KeyDistribution =
access in UAV networks. {K1, K2, ..., Km}). Encrypted data (Encrypted-
Cryptography and key exchange play Data = E(OriginalData, Kn)) utilizes the key
a pivotal role in securing communications for encryption/decryption. Chaotic map
within Unmanned Aerial Vehicle (UAV) net- synchronization (Synchronization = i(Map1,
works, where sensitive information transmis- Map2)) ensures consistency, and security
sion demands confidentiality and integrity. In analysis (SecurityAnalysis = j(Data, KeyDis-
UAV networks, cryptographic mechanisms tribution)) evaluates the algorithm’s efficacy
utilize mathematical algorithms to encode in safeguarding UAV communications. This
information, thwarting unauthorized access, dynamic key generation approach enhances
data tampering, and eavesdropping. Key security, mitigating the risk of unauthorized
exchange, a critical component of secure com- access in UAV networks.
munication, involves sharing cryptographic
keys between parties to facilitate encryption
and decryption. This process ensures data 4. Results and Discussions
integrity, a crucial aspect in UAV networks, Network Simulator 3 (ns-3) played a piv-
where robust key exchange protocols are otal role as a powerful tool in simulat-
imperative. ing and evaluating the proposed Chaotic
Strengthening UAV network security: Advanced authentication-based intrusion detection systems 71
Map-Based Authentication Algorithm 4.1. Packet delivery ratio
within a virtual Unmanned Aerial Vehicle
Packet Delivery Ratio (PDR) stands as a criti-
(UAV) network environment. Leveraging
cal performance metric that signifies the pro-
ns-3’s capabilities, we constructed a robust
portion of successfully transmitted packets in
framework for modeling complex network
relation to the total sent. This metric serves as
scenarios, allowing for a thorough assess-
a barometer for network reliability, where a
ment of the algorithm’s performance under
heightened PDR signifies efficient data deliv-
diverse conditions. By utilizing ns-3, we
ery. Conversely, a diminished ratio suggests
efficiently generated simulation results that
potential challenges such as packet loss or
facilitated a comprehensive examination of
disruptions in transmission. PDR thus pro-
key metrics, including packet delivery ratio,
vides a crucial insight into the overall robust-
delay, and overhead. This integration of
ness and effectiveness of the network’s data
ns-3 significantly enhanced the credibility
transmission capabilities.
of our research, providing valuable insights
The graphical representation in Figure 8.1
into the algorithm’s efficacy in ensuring
clearly illustrates the superior PDR achieved
secure communication among UAVs. More-
by the proposed CHAOS authentication sys-
over, the simulation allowed us to assess
tem, emphasizing its effectiveness in improv-
the algorithm’s resilience against poten-
ing data delivery rates. This highlights the
tial cybersecurity threats, particularly in
strength of chaotic maps in authentication,
the context of Intrusion Detection Systems
presenting a more dependable and efficient
(IDS) designed for UAV networks. Evalua-
approach when compared to existing meth-
tion parameters used in the proposed model
ods. The distinct difference in the PDR graph
is listed in Table 8.1.
accentuates the competitive advantage offered
by chaotic authentication in the domain of
secure communication within UAV networks.
Table 8.1. Simulation Parameters
Parameter Description 4.2. Delay
Mobility Models Random Waypoint Network delay, which includes propagation,
Number of UAVs 30 transmission, and processing delays, gauges
Communication 300 meters the time required for data packets to trav-
Range erse the network. The reduction of delay is
Altitude and Height 150 meters
Communication Directional Antennas
Models
Propagation Models Two-Ray Ground
Traffic Models CBR with 5 Mbps
Routing Protocols AODV
Propagation Delay 10 milliseconds
Link Quality SNR of 20 dB
AQ1: Data Packet Size 512 bytes
Please
provide MAC Protocols CSMA/CA
source Energy Models Energy Harvesting Figure 8.1. Comparison of proposed method
for all the
figures Model with existing with respect to PDR.
and table Source: Source:
AUTHOR'S SELF GENERATED AUTHOR'S SELF GENERATED
72 Innovations in Computational Intelligence: Trends and Technologies
paramount in guaranteeing swift and respon-
sive communication within the network.
Minimizing these delays enhances the overall
efficiency and responsiveness of the commu-
nication infrastructure.
The delay graph underscores the remark-
able efficiency of the proposed CHAOS
authentication system in minimizing com-
munication delays, as depicted in Figure 8.2.
This emphasizes the robustness of chaotic
maps in the authentication process, leading
Figure 8.3. Comparison of proposed method
to reduced delays and ensuring timely data with existing with respect to overhead.
transmission in UAV networks. The compari-
Source:
son with traditional methods highlights the AUTHOR'S SELF GENERATED
superior performance of CHAOS in optimiz-
ing network responsiveness. HT, and DKRP. This showcases CHAOS as
a promising solution for optimizing network
resources and enhancing overall system per-
4.3. Overhead
formance in UAV communication.
In the realm of networking, overhead sig-
nifies the supplementary data or resources
beyond the fundamental requirements of 5. Conclusion
transmission, encompassing error check- In conclusion, the contemporary landscape
ing and control information. While crucial, of Unmanned Aerial Vehicle (UAV) networks
an overabundance of overhead can result in demands sophisticated security measures,
inefficiencies, underscoring the importance of and the proposed Intrusion Detection System
a well-balanced approach to ensure optimal (IDS) significantly advances the field. By pri-
utilization of the network. oritizing authentication-based strategies and
In terms of overhead, Figure 8.3 illustrates recognizing the limitations of traditional IDS
CHAOS’s notable advantage in mitigating in UAV scenarios, the system incorporates
unnecessary burdens on the network. The use advanced mechanisms such as cryptogra-
of chaotic maps in authentication proves to phy, biometrics, and dynamic key generation.
be more streamlined, offering a significant The synergy of these elements, coupled with
reduction in overhead compared to QKD, the robustness introduced by chaotic maps,
not only ensures secure key exchange but
also poses computational challenges for
potential attackers. The adaptability of the
system, augmented by anomaly detection
and machine learning, further distinguishes
between normal and malicious UAV behav-
ior. Simulated evaluations utilizing tools like
ns3 provide tangible insights into the system’s
performance metrics, including Packet Deliv-
ery Ratio, Delay, and Overhead. This com-
prehensive approach culminates in a robust
Figure 8.2. Comparison of proposed method solution that safeguards UAV network integ-
with existing with respect to delay. rity, effectively addressing the unique security
Source: challenges prevalent in diverse operational
AUTHOR'S SELF GENERATED
Strengthening UAV network security: Advanced authentication-based intrusion detection systems 73
scenarios. The proposed authentication- flying ad-hoc networks”, Ad Hoc Networks,
based IDS emerges as a pivotal advancement, Vol 133, 2022.
promising heightened security in the dynamic [7] C Atheeq, M Munir Ahamed Rabbani,
realm of UAV communications. “CACK—A Counter Based Authenticated
ACK to Mitigate Misbehaving Nodes from
MANETs” Recent Advances in Computer
Acknowledgement Science and Communications Vol 14, no. 3,
Dr. Atheeq and Dr. Altaf provided essential pp. 837–847, 2021.
insights and guidance, shaping the theoretical [8] Pandey, Gaurav K., et al. “Security threats
and mitigation techniques in uav commu-
framework and methodology. Dr. Shanila’s
nications: A comprehensive survey.” IEEE
expertise significantly enriched the data anal-
Access 2022.
ysis and interpretation phases. Ms. Shaista [9] Li, B. et al., “Anti-honeypot enabled optimal
played a crucial role in the implementation attack strategy for industrial cyber-physical
and experimental setup. It is noteworthy that systems”, IEEE Open Journal of the Com-
this research was independently conducted puter Society, Vol 1, pp. 250–261,2020.
without external sponsors, highlighting [10] Alshaer, N. et al., “Reliability and security
our unwavering commitment to advancing analysis of an entanglement-based QKD
knowledge and fostering innovation through protocol in a dynamic ground-to-UAV FSO
both individual and collective efforts. communications system”, IEEE Access, Vol
9, pp. 168052–168067, 2021.
[11] C Atheeq, M Munir Ahamed Rabbani,
References “Secure Data transmission in integrated
[1] Lateef, M.A. et al., “Data Aegis Using Che- internet MANETs based on effective trusted
byshev Chaotic Map-Based Key Authentica- knowledge algorithm” Indian Journal of Sci-
tion Protocol”, In Intelligent Manufacturing ence and Technology. Vol. 9 No. 47, 2016.
and Energy Sustainability: Proceedings of [12] Conrad, A. et al., “Drone-based quantum
ICIMES, pp. 187–195, 2022. key distribution (QKD)”, In Free-space laser
[2] C Atheeq, M Munir Ahamed Rabbani, communications XXXIII Vol. 11678, pp.
“Mutually Authenticated Key Agreement 177–184, 2021.
Protocol based on Chaos Theory in Integra- [13] Assaf, Tasneem, et al. “High-Rate Secret Key
tion of Internet and MANET” International Generation Using Physical Layer Security
Journal of Computer Applications in Tech- and Physical Unclonable Functions.” IEEE
nology, Vol. 56, No. 4, pp.309–318, 2017. Open Journal of the Communications Soci-
[3] Rosch-Grace, D. and Straub, J., “Analysis of ety 4, pp. 209–225, 2023.
the likelihood of quantum computing pro- [14] Tu, Yu-Ju, and Selwyn Piramuthu. “Security
liferation”, Technology in Society, Vol 68, and privacy risks in drone-based last mile
pp.101880, 2020. delivery.” European Journal of Information
[4] Gómez Hernández, J.A. et al, “Crypto-Ran- Systems, pp. 1–14, 2023.
somware: A Revision of the State of the Art, [15] Calcara, Antonio, et al. “Drones and Offen-
Advances and Challenges”, Electronics, Vol sive Advantage: An Exchange–The Authors
12, no. 21, p.4494, 2023. Reply.” Security Studies 32.3, pp. 582–588,
[5] S. Syed Abdul Syed et al., “Enhanced Chaotic 2023.
Map Based Key Agreement Mitigate Packet [16] Omolara, Abiodun Esther, Moatsum
Dropping Attack from MANETs” Interna- Alawida, and Oludare Isaac Abiodun.
tional Journal of Electronic and Communi- “Drone cybersecurity issues, solutions, trend
cation Engineering, Vol 10, no 7, 2023. insights and future perspectives: a survey.”
[6] Tsao et al., “A survey of cyber security threats Neural Computing and Applications 35.31,
and solutions for UAV communications and pp. 23063–23101, 2023.
9 Comparative analysis of wireless
routing protocols for security threat—A
review
Bhupal Arya1,a, Amrita Kumari2,b, and Jogender Kumar3,c
1
Research Scholar, Department of Computer Science and Engineering, Quantum University
Uttarakhnad, Rorrkee, India
2
Assistant Professor, Department of Computer Science and Engineering, Quantum University
Uttarakhnad, Rorrkee, India
3
Assistant Professor, Department of Computer Science, GBPIET Ghurdauri Pauri Garhwal, Pauri,
India
Abstract: There are various security issues, challenges, and threats in wireless networks. In this paper, we review
recent work in this field that primarily addresses such issues and challenges. Different researchers have taken
various approaches to justify their research based on different parameters. Methods such as enhanced packet
delivery, residual energy, and link quality, on-demand link and energy aware dynamic multipath (O-LEADM),
Rate aware congestion control (RACC), Quadrature Amplitude Modulation (QAM), Quadrature Phase Shift
Keying (QPSK), Network function virtualization (NFV), Software-defined network (SDN), hidden Markov
model (HMM), IDS, scalable and energy-efficient cluster-based anomaly detection (SEECAD) etc. are used to
identify jamming signal problems in fuzzy logic. The article discusses various issues and approaches covered by
many researchers in last few years paying close attention to wireless network-related topics.
Keywords: PCA, RACC, O-LEADM, HMM Model
1. Introduction stands as a critical security concern, capable
of disrupting communication and network
Wireless networks, by their very nature, functionality. These attacks are particularly
introduce vulnerabilities that are distinct relevant in scenarios where devices form net-
from those of traditional wired networks. works without a fixed infrastructure, such as
The inherent openness of wireless commu- disaster recovery. Black Hole Attack involves
nication, coupled with the broadcast nature a malicious node within a wireless network
of radio signals, creates opportunities for that deceitfully attracts and intercepts data
eavesdropping, unauthorized access, and data packets from legitimate nodes [1]. By falsely
interception. The rapid expansion of wireless advertising itself as having the minimum and
infrastructures, including Wi-Fi, cellular net- shortest path to the destination, the malicious
works, and emerging technologies like Inter- node request to other nodes asking for route
net of Things (IoT) devices, has magnified request (R-Req) to route their traffic through
these security concerns, making it imperative it. However, instead of forwarding the data
to explore novel approaches for safeguard- packets to their final destinations, the mali-
ing wireless communications. In the realm cious node drops or hold them, resulting in
of wireless networks, Black Hole Attack data loss and breakdown of communication.
[email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-9
Comparative analysis of wireless routing protocols for security threat—A review 75
The process of a Black Hole Attack follows this research which inspire the researchers
these distinct stages mostly. in this field to propose methods to detect
Advertisement Phase: The malicious node the basic behavior of black-hole nodes and
claims that it has the shortest route for the the route request (RREQ) and RREP which
R-Req by the source and it generate the Route is mostly used by attackers to neutralize the
request reply (RREP) and send it to the target network. O-LEADM [2] is a method used to
node. detect black hole in MANET. It is essential
Attraction Stage: Legitimate nodes, relying to provide safe and secure communication in
on the falsified routing details, unwittingly wireless network, prevent from the network
direct their data packets toward the mali- damage and harm in small and large scale.
cious node. The purpose of this study is to showcase the
Packet Interception: Upon receiving these research conducted in this field and empha-
packets, the malicious node abstains from size the approaches taken to get around secu-
forwarding them to their rightful destina- rity vulnerabilities.
tions, opting to discard or consume them.
Impact and Consequences: As a result,
genuine nodes remain unaware of the com- 2. Theoretical Details
promised communication path, leading to In order to achieve wireless network architectural
data loss, network congestion, and potential complexity, various heterogeneous non-realistic and
service disruption. The detection and elimina- artificial approaches are offered. Communicating in
tion of the Black Hole Attacks present intri- various situations makes the deployment of security
cate challenges due to their deceptive nature. measures and regular packet delivery even more
Nevertheless, researchers have proposed complex. Figure 9.1 illustrates different types of
strategies to tackle this menace. jammers commonly encountered in cybersecurity
Due to highly dynamic and challenging and wireless security. Wireless networks have
open environment of wireless network there revolutionized the way we connect and
are many reasons for packet failure like bat- communicate, but they also come with a range of
challenges that need to be addressed to ensure their
tery drainage in IoT, signal interference and
reliability, security, and efficiency. Some of the key
malicious nodes. A malicious node or faulty
challenges of wireless networks include:
node can enter forcefully or without the prior
knowledge of victim node into the network • Signal Interference and Congestion: Wire-
and damage, disturb the network settings less networks work in shared frequency
and can decrease overall network throughput bands which can lead to potential interfer-
and performance. Black-hole node problem ence from other devices and networks. This
may try to route all the data packets towards interference can degrade the quality of the
itself by falsely claiming of having shortest signal and impact the overall transmission.
path and false unrealistic information. Source • Reduced Bandwidth: The bandwidth of
node chooses the path having higher RREP wireless networks is lower than that of
[2] destination sequence by simply disregard- wired networks. This restriction may
ing other routes and RREPs, comparing all result in slower data transfer in nations
RREPs being trapped in the attacker’s strat- having large population with numerous
egy, and compromising the node connected connected devices.
to the aggressor. The risk of packet drop • Coverage and Range: The range of wire-
exists for nodes that forward data packets less networks can be limited, especially in
along pathways that include rogue nodes, indoor environments or areas with obsta-
which reduces overall network performance. cles. Achieving consistent coverage across
This is the main motivational point behind larger areas can be challenging.
76 Innovations in Computational Intelligence: Trends and Technologies
• Data Security and Privacy: Wireless com-
munication is susceptible to eavesdrop-
ping and unauthorized access due to the
process of broadcast of signals. Ensuring
data security and user privacy is a signifi-
cant challenge.
• Authentication and Authorization: Prop-
erly authenticating and authorizing
devices and users on a wireless network is
essential to prevent unauthorized access.
However, doing so securely and efficiently
can be complex.
• Quality of Service (QoS): Maintaining AQ1:
Please
consistent QoS in wireless networks can be provide
difficult due to factors like signal strength figure
in-text
variations, interference, and network con- citation
gestion. This is particularly important Figure 9.1. Classification of Jammers in caption
for applications like video streaming and Cybersecurity and Wireless Security and
source.
online gaming. Source:
AUTHOR'S SELF GENERATED
• Mobility Management: Wireless networks spectrum management becomes critical.
often serve mobile devices that move Allocating and utilizing frequency bands
between different access points. Seam- effectively to prevent overcrowding and
less handovers from one access point to interference is a challenge.
another without disrupting the connec-
tion can be challenging, especially for 2.1. Security threats to wireless
real-time applications.
• Energy Efficiency: Many wireless devices network
used worldwide, now operate on battery • Traffic Analysis: In this type of problem an
power, and optimizing energy consump- attacker can access information related to
tion is crucial to extend the duration of the identification of activities on the net-
device battery life. Balancing performance work physical location of the user, traffic
with energy efficiency is a significant analysis of the user.
challenge. • Eavesdropping: In this type of attack an
• Heterogeneity: Wireless networks encom- attacker tries to listen to the conversation
pass various technologies such as Wi-Fi, of other person without his/her permis-
cellular, Bluetooth, and IoT protocols. sion. It can be categorized into active and
Managing and integrating these diverse passive forms.
technologies to ensure seamless connec- • Denial of Services (DoS): A Denial of Ser-
tivity can be complex. vice (DoS) attack is a type of cyber-attack
• Location Accuracy: Many applications in which the attacker tries to disturb the
rely on accurate location information. normal functioning of a target node where
Achieving precise location accuracy in the target node is not able to reach the
indoor and urban environments can be resource and the request for the resource
challenging due to signal reflections and is denied.
obstructions. • Brute Force Attack: Brute force attack is
• Spectrum Management: As demand for considered as a passive type of attack in
wireless communication grows, efficient which the attacker will try a number of
Comparative analysis of wireless routing protocols for security threat—A review 77
efforts or trial and error to encrypt the des-sequence [1] which claims to have the
user’s message. shortest path to deliver the data from the
• Black hole Attack: A Black Hole Attack is source to destination node in the network,
a type of security threat in wireless ad hoc which is not true and hence it will disturb
networks in which a false node will gener- the communication.
ate a request-response, req-sequence and
3. Comparative Analysis of Wireless Routing Protocols for Security
Threat Table 9.1. Comparative Analysis of Wireless Routing Protocols
S. Author Parameters/ Findings
No Method
1. Suma and link quality, • The behavior of any node is determined using control AQ2:
Please
Harsoor [1] residual energy messages destination-sequence (des-Seq) and reply- provide
Higher packet sequence (rep-Seq) table in-
Delivery decision • Every intermediary node within the network delivers text cita-
tion and
Sequence a des-Seq message to every node next to it, and each caption.
control messages node returns to the intermediate node through a rep-
destination- Seq message.
sequence and • If a node’s neighbors’ des-Seq and req-Seq diverge,
reply-sequence it may be considered malicious. Nodes choose
(rep-Seq) forwarding based on behavior and capability to
achieve QoS criteria.
2. Angurala [2] (des-Seq) • Authors differentiated between packet loss caused by
message congestion and malicious nodes.
destination- • The discussed Sheme utilizes an The energy-aware
sequence and dynamic multipath (O-LEADM) routing strategy for
(rep-Seq) reply- MANETs locates black-hole nodes using on-demand
sequence [2] links and a bait plan of action.
3. Grove [3] NS2 simulator • The RACC technique, which is applied at the transport
layer in the OSI model, improves congestion control by
regulating the rate of source nodes at specific hotspot
locations.
• RACC is used to evaluate different modulation
schemes, like Binary Phase Shift Keying (BPSK),
Quadrature Phase Shift Keying (QPSK), and 16
(Quadrature Amplitude Modulation, or QAM), in
order to identify the best modulation scheme for the
method that is being reviewed and suggested.
• The throughput parameter was improved by 17%,
the packet delivery ratio was improved by 8.35%, the
normalized routing overhead was improved by 0.56%,
the MAC overhead was improved by 0.64%, the
average end-to-end delay was improved by 2.04%
(continued)
78 Innovations in Computational Intelligence: Trends and Technologies
S. Author Parameters/ Findings
No Method
4. NFV and (SDN), NFV, • Connecting their devices to the Internet and other
SDN [4] DNS networks opens up a world of potential ramifications
for consumers.
• It is possible to get what we need without keeping it
permanently on our gadgets. Communication enables
us to coordinate and plan our actions with one another.
• These connected gadgets offer the foundation for the
systems that assist us in running our lives. However,
these links also make our data susceptible to theft and
our equipment defenseless to harm. Network and cyber
security deals with this issue. Many of the companies
spend millions of dollars a year to secure their
computer networks and safeguard their data. Cisco
provides hardware equipped with firewalls, modeling
tools, and many network security protocols; certain
software is required to operate them.
5. Michael IDS and ANN- • The PCA: The authors constructed an IDS system for
Savva [7] based IDS for man-in-the-middle attacks based on two approaches
simulation and compared the outcomes.
they used NS2 • ANN method was used in the second. While the ANN-
simulator and based IDS reported 100% accuracy and detection rate,
they found the the PCA-based IDS claimed 76% accuracy and 74%
accuracy of detection rate.
ANN based IDS • The simulation was conducted using the NS2 simulator.
is more accurate where the 76% accuracy rate and 74% detection rate
with a accuracy of the PCA-based IDS are achieved. When 100% of
of almost 100% detections are made in the case of ANN-based IDS
-based IDSs is accuracy and 100% of PCA- and ANN-based average
0.768 and 1.0, mean accuracy is achieved. and ANN.
respectively.
6. Savva, Three types of • Jammers that are proactive, persistent, deceptive, or
Michael [8] jammer attacks random. First-generation attacks are particularly easy
Proactive to identify
Jammers • Their primary goal is to disrupt and impair node
Reactive connectivity while also keeping the source channel
Jammers occupied. Identifiability of the second sort of attack is
Specific function more difficult than that of continual jammers since it
jammers. focuses on various approaches where valid packets are
provided instead of constant signal transmission.
• The third kind of attack can operate in two ways:
sometimes it acts as a deceptive jammer and other
times it acts as a constant jammer. reactive jammer for
RTS/CTS. In this kind of attack, the RTS request is
destroyed by jammers.
(continued)
Comparative analysis of wireless routing protocols for security threat—A review 79
S. Author Parameters/ Findings
No Method
7. Simulation AODV routing Channel Frequency 2.4 GHz No. of Nodes 100 nodes
based protocol Terrain File- Digital Elevation Model (DEM)
protocols can yield the Traffic Source- Constant Bit Rate (CBR) traffic load
[9-20] greatest results
in terms of
wireless network
congestion
avidity. They
have also
suggested a task
in which delivery
timing, and
optimizing data
throughput
8. Jeffrey, Cyber-physical The increased access to the public internet has made
Nicholas[21] systems (CPS) incidents targeting essential civilian infrastructure—like oil
are software- pipelines and electrical power grids—alarmingly common.
and hardware- This has increased CPS vulnerability. This article provides
integrated a thorough analysis of the body of research that looks at
integrated the most recent developments. Over the past ten years, CPS
systems. has grown quickly across a variety of industries, including
telemedicine, smart manufacturing, Internet of Things,
autonomous cars etc.
4. Conclusion control techniques in wireless sensor net-
works. International Journal on Recent and
Advancement in the field of computer Innovation Trends in Computing and Com-
and software technology had make a munication, 10(8), 47–54.
remark- able contribution in the growth [3] Grover, Amit, R. Mohan Kumar, Mohit
of wireless network but it has also many Angurala, Mehtab Singh, Anu Sheetal, and
serious issues and challenges as discussed R. Maheswar. (2022). Rate aware conges-
in this paper researcher has given tion control mechanism for wireless sensor
different approach and result in the field networks. Alexandria Engineering Journal,
of security in wireless networks. Table 9.1 61(6), 4765–4777.
presents comparative analysis of wireless routing [4] Suma, S. and Harsoor, B. (2016). An
protocols. However, in my view, the HMM approach to detect black hole attack for
model utilized in the Contiki-based congestion control utilizing mobile nodes in
operating system with the Cooja wireless sensor network. In: Editor, F., Edi-
simulator gives best result although some tor, S. (eds.) CONFERENCE 2016, LNCS,
chances of improvement are there. vol. 9999, pp. 1–13. Springer, Heidelberg. AQ3:
References [5] Kalkha, H., Satori, H. and Satori, Khalid. Please
note
[1] Suma, S. and Harsoor, B. (2022). An (2018). Preventing Black Hole Attack in that the
approach to detect black hole attack for Wireless Sensor Network Using HMM. references
congestion control utilizing mobile nodes in [6] Maria Hanif, Humaira Ashraf, ZakiaJalil, 5-6 is not
cited in
wireless sensor network. Materials Today: Noor Zaman Jhanjhi, MamoonaHumayun,
text.
Proceedings, 56, 2256–2260. Saqib Saeed, and Abdullah M Almuhaideb.
[2] Angurala, M., Bala, M. and Khullar, V. (2022). Aibased wormhole attack detection
(2022). A survey on various congestion
80 Innovations in Computational Intelligence: Trends and Technologies
techniques in wireless sensor networks. Elec- [14] Premkumar, M., and Sundararajan, T. V. P.
tronics, 11(15), 2324. (2021). Defense countermeasures for DoS
[7] Michael Savva, Iacovos Ioannou. (2022). attacks in WSNs using deep radial basis net-
Fuzzy-logic based IDS for detecting jam- works. Wireless Personal Communications,
ming attacks in wireless mesh IoT net- 120(4), 2545–2560.
works. 20th Mediterranean Communication [15] Prasse, P., Machlica, L., Pevný, T., Havelka,
and Computer Networking Conference J. and Scheffer, T. (2017). Malware detection
(MedComNet). by analyzing network traffic with neural
[8] Sumaiya Thaseen Ikram and Aswani Kumar networks. 2017 IEEE Security and Privacy
Cherukuri. (2016). Improving accuracy of Workshops (SPW), 205–210.
intrusion detection model using pca and [16] Krishnamurthy, M. and Rajashekara, H.M.
optimized svm. CIT, 24(2), 133–148. (2011). Current trends in wireless technolo-
AQ4:
[9] Naga RohitSamineni, Ferdous A. Barbhuiya, gies in academic libraries. DESIDOC Jour-
Please and Sukumar Nandi. (2012). Stealth and nal of Library and Information Technology,
note semi-stealth mitm attacks, detection and 31(1).
that the
references
defense in ipv4 networks. In 2012 2nd IEEE [17] Chen, Z., Zhou, W., Wu, S., and Cheng,
9-20 is International Conference on Parallel. Dis- L. (2020). An adaptive on-demand mul-
not cited tributed and Grid Computing. tipath routing protocol with QoS support
in text. [10] Farhan Ahmad, AsmaAdnane, Virginia Fran- for high-speed MANET. IEEE Access, 8,
queira, FatihKurugollu, and Lu Liu. (2018). 44760–44773.
Man-in-the-middle attacks in vehicular ad- [18] El-Semary, A. M. and Diab, H. (2019). BP-
ATION ADDEDhoc networks: Evaluating the impact of AODV: Blackhole protected AODV rout-
attackers’ strategies. Sensors. ing protocol for MANETs based on chaotic
[11] QuaziWarisha Ahmed, Shruti Garg, Amrita map. IEEE Access, 7, 95185–95199.
Rai, Manikandan Ramachandran, Noor [19] Thebiga, M., and SujiPramila, R. (2020).
Zaman Jhanjhi, MehediMasud, and Moham- A new mathematical and correlation coef-
med Baz. (2022). Ai-based resource alloca- ficient based approach to recognize and to
tion techniques in wireless sensor internet of obstruct the black hole attacks in manets
things networks in energy efficiency with data using DSR routing. Wireless Personal Com-
optimization. Electronics, 11(13), 2071. munications,, 114(2), 975–993.
[12] J. R. Renofio, M. E. Pellenz, E. Jamhour, [20] Singh, J., Goyal, S., Kumar Kaushal, R.,
A. Santin, M. C. Penna, and R. D. Souza. Kumar, N., and Singh Sehra, S. (Eds.).
(2016). On the dynamics of the RPL proto- (2024). Applied Data Science and Smart
col in AMI networks under jamming attacks. Systems (1st ed.). CRC Press. https://doi.
In 2016 IEEE International Conference on org/10.1201/9781003471059
Communications (ICC), pp. 1–6, IEEE. [21] Jeffrey, Nicholas, Qing Tan, and José R. Vil-
[13] David, J., and Thomas, C. (2019). Efficient lar. (2023). A review of anomaly detection
DDoS food attack detection using dynamic strategies to detect threats to cyber-physical
thresholding on flow-based network traffic. systems. Electronics, 12(15), 3283.
Computers and Security, 82, 284–295.
10 Cascadability analysis of free space
optical wireless data centers
Anand Kumar Dixit1,a, Meenakshi Srivastava2, b, and
Rajiv Srivastava3,c
1
Research Scholar, Amity Institute of Information Technology, Amity University Uttar Pradesh,
Lucknow, India
2
Assistant Professor, Amity Institute of Information Technology, Amity University Uttar Pradesh,
Lucknow, India
3
Ex-Faculty, Indian Institute of Technology, Jodhpur, Rajasthan, India
Abstract: Data centers having optical nature are the heart of modern day communication. The data is rising
continuously and thus increasing the demand for more bandwidth. The current data center relies on the elec-
tronics thus producing lot of heat and for connecting them huge amount of cabling is required. To reduce copper
cabling, use of fiber optic technology with fiber cable was proposed, and optical components produce lesser heat
as compared to electronic devices. However, fiber cabling is still an issue. To deal the cabling and heat dissipa-
tion, free space optical communication can be a better choice. This paper, presents the cascadability analysis
of the free space optical wireless data centers. In this work line of sight links are considered for cascadability
analysis and bit error rate is evaluated for various distances.
Keywords: Bit error rate, data center, free space optical communication, line of sight
1. Introduction
High speed communications switches making the wiring more complex [6]. In addi-
and networks are basically used in tion, we have more consumption of power
data-intensive computing platforms. with electrical switches than optical DCNs
These switches and network divides which also have an added advantage of less
the data into computing resources and expensive [7].
therefore the data is processed separately The issue of the limited frequency spec-
delivering output and scalability as trum in classical RF systems is becoming
per the data. A lot of researches are progressively more critical. However, by
going on to deal with these issues such as implementing optical communication free
introductionof new interconnect space technology, such a problem can be alle-
topologies for data center networks [1–4]. viated [8].
In this topology, we put twenty to fifty 60 GHz lines can support multi-Gbps
servers in a rack along with top of rates [9–11]. In contrast, to solve cabling AQ1:
rack (ToR) switch. Now, as a major issues, [1] proposes a totally wireless DC sys- Please
abbre-
portion of the traffic passes through tem based on 60 GHz technology. Despite its viate
some of the hot ToRs to other [5] which promise, 60 GHz technology has its demerit “DCN”,
makes the requirement of fast reconfiguration like, very high attenuation, low bandwidth “ToR”,
and
of Data Center Networks (DCNs). Generally, [10]. Because of the lack of atmospheric limi- “NLoS”
commodity electrical switches are used tations to free space optical (FSO) commu- in full
at first
bynormal DCNs to optimize nication in interior systems, we recommend mention.
the limited bandwidth. But, the problem
with this is the need of numerous links
and switches
a
[email protected]; [email protected]; [email protected]
DOI: 10.1201/9781003606635-10
82 Innovations in Computational Intelligence: Trends and Technologies
employing FSO in Data Center. Furthermore, The impulse response of LoS channel hLoS
light speed in FSO is approximately 1.5 can be expressed as [14, 15]:
times faster than fiber optics, it is ensure less
latency. As a result, we anticipate that FSO (1)
will result in high-performance, low-cost
DC infrastructure. This approach is useful where Ar is area of the photo-detector,
in situations where lines of optical fiber can- ϕ is radiance angle,
not be installed [12, 13]. It has the advan- ψ is light incidence angle,
tages of spectrum restriction and high-speed d is distance between LED and PD.
1
information transmission over other wireless m = - ln ( 2 ) / ln lim ( cos f1/2 ) is order of lam-
x ®¥ n
communication. Above mentioned system is bertian source emission and LED semiangle
predictable to swap other WC methods in half power is denoted by ϕ1/2.
various industries, including last-mile com-
munication. This technology, which is still in
its early phases, is broadly used in network. 3. Mathematical Modelling of
In the Figure 10.1, communication between Cascadability Analysis
the ToRs using line of sight (LoS) and
In this section, mathematical analysis of cacad-
Non-LoS links is shown. Considering four
ability analysis is presented. In Figure 10.2,
racks for an as example, in cluster 1, rack
BUS topology is shown where four racks are
1, and rack 2 can communicate using LoS
connected.
communication, similarly in cluster 2, rack
3, and rack 4 can communicate using LoS Power Calculation
model. This type of communication is The power received by the rack 2 when rack
known as intra racks com- munication. The 1 transmit is
rack 2 of cluster 1 and rack 3 of cluster 2 can
also communicate using the LoS model. The (2)
communication between the rack 1 and
racks 3, 4, will takes place using NLoS
model. The rack 1 can also communi- cate
to rack 3 via rack 2 using LoS model. In this
work LOS links are considered for cas-
cadability analysis and BER is evaluated
for various distances.
2. Free Space Optical
Communication AQ2:
Please
provide
In FSO communication systems, free space Figure 10.1. Communication between the ToRs source
serves as the communication channel between using LoS and NLoS links. for all
transceivers, playing a vital role in the effective Source:
figures.
AUTHOR'S SELF GENERATED
transmission of optical signals. The quality of
the received signal is significantly impacted by
the parameters of this channel. FSO enables
optical data transfer with potential data rates
of up to 100 Gbps, making it suitable for high-
speed optical networks. However, atmospheric
conditions can hinder FSO performance, limit- Figure 10.2. LoS communication between the
ing sensitivity and achievable data rates while cascaded racks (BUS topology).
maintaining an acceptable bit error rate (BER). Source:
AUTHOR'S SELF GENERATED
Cascadability analysis of free space optical wireless data centers 83
The power received by the rack 3 when BER Calculations
rack 1 transmit is In optical WCSNR is given by [14]
(8)
where, Pr(i) is the power received at the ‘ith’
(3) rack.
The shot noise variance will be given by
(4) (9)
The power received by the rack 4 when The thermal noise expression is as follows:
rack 1 transmit is
.(10)
(5)
Finally, BER can be evaluated using:
In Figure 10.3, 16 racks are connected, (11)
where distance among the nodes is shown.
The power received by the rack 14 when Where, Q is error function.
rack 1 transmit is
(6) 4. Results
Table 10.1 contains parameters utilized in the
The power received by the rack 16 when calculation. In the simulation, we used a sin-
rack 1 transmit is gle LED and a single photo detector (PD) [16].
(7) Table 10.1. Simulation parameters [14]
Parameter Value AQ3:
Please
Semi angle 70° provide
source for
LED transmission power 20 dBm
Table.
Photo-detector Area (A) 10−4 m2
Field of View 70°, 10°
Refractive Index 1.5
Room Size 10×10×10 m3
Responsivity (R) 1.28
Avalanche Factor (M) 20
Noise factor 1.5
I2 0.087
Background current (IB) 5 mA
Noise bandwidth factor(I3) 0.52
Fixed capacitance of photo(Cpd) 110 pF/cm2
Light speed 3×108 m/s
Figure 10.3. LoS communication between the Transconductance (gm) 30
racks (16 Racks topology). voltage gain in open loop (Gol) 10
Source: Source:
AUTHOR'S SELF GENERATED
AUTHOR'S SELF GENERATED
84 Innovations in Computational Intelligence: Trends and Technologies
The server room’s volume is estimated to be racks is 3 meter. FoV=70° is applicable for
10 × 10 × 5 m3. In the simulation transmitter the broadcast messages, while in case of point
and receiver pair placed opposite side of the to point connection FoV=10° is applicable.
racks. FSO rack having servers such that the In case of FoV=10°, the maximum received
same side of the rack having all of the serv- power is −3.36 dBm.
ers’ transmitters (receivers). Main aim is to In Figure 10.5, transmitter and receiver
use the intra/inter-rack option. Optical Beams distance is 3√2 meters with FoV=70° and
are scattered to all servers using a beam dis- FoV=10°. The Pma is −35.29 dBm, and the
tributer, allowing for switch-free intra-rack minimum received power is −45.80 dBm,
connectivity. when the transmitted power is 20 dBm. This
In Figure 10.4, transmitter and receiver result is applicable for all the connection
distance is three meters with FoV=70° and where transmitter and receiver racks are align
FoV=10°. The maximum received power to each other, and distance between the adja-
(Pmax) is −32.29 dBm, and the minimum cent racks is 3√2 meter. In case of FoV=10°,
received power is −47.15 dBm, when the the maximum received power is −6.24 dBm.
transmitted power is 20 dBm. This result is In Figure 10.6, transmitter and receiver dis-
applicable for all the connection where trans- tance is 6 meters with FoV=70° and FoV=10°.
mitter and receiver racks are align to each The Pma is −38.29 dBm, and The min power
other, and distance between the adjacent received is −45.18 dBm, when the transmitted
Figure 10.4. LoS communication between the Figure 10.5. LoS communication between the
racks is 3 meters (a) FOV=70° (b) FOV=10°. racks is 3√2 meters (a) FOV=70° (b) FOV=10°.
Source: Source:
AUTHOR'S SELF GENERATED AUTHOR'S SELF GENERATED
Cascadability analysis of free space optical wireless data centers 85
power is 20 dBm This result is applicable for
all the connection where transmitter and
receiver racks are align to each other, and dis-
tance between the adjacent racks is 6 meters.
In case of FoV=10°, the max power received
is −9.19 dBm.
In Figure 10.7, transmitter and receiver
distance is 6√2 meters with FoV=70° and
FoV=10° The Pma is −41.30 dBm, and The
min power received is −45.48 dBm, when the
transmitted power is 20 dBm This result is
applicable for all the connection where trans-
mitter and receiver racks are align to each
other, and distance between the adjacent
racks is 6√2 meters. In case of FoV=10°, the
max power received is −12.17 dBm.
In Figure 10.8, transmitter and receiver
distance is 9 meters with FoV=70° and
FoV=10°. The max power received is −41.81
Figure 10.7. LoS communication between the
racks is 6√2 meters (a) FOV=70° (b) FOV=10°
Source:
Figure 10.6. LoS communication between the Figure 10.8. LoS communication between the
racks is 6 meters (a) FOV=70° (b) FOV=10°. racks is 9 meters (a) FOV=70° (b) FOV=10°.
Source: Source:
AUTHOR'S SELF GENERATED AUTHOR'S SELF GENERATED
86 Innovations in Computational Intelligence: Trends and Technologies
dBm, when the transmitted power is 20 dBm.
This result is applicable for all the connection
where transmitter and receiver racks are align
to each other, and distance between the adja-
cent racks is 9 meters. In case of FoV=10°, the
max power received is −12.68 dBm.
In Figure 10.9, transmitter and receiver
distance is 9√2 meters with FoV=70° and
FoV=100. The Pma is −44.81 dBm, when the
transmitted power is 20 dBm. This result is
applicable for all the connection where trans-
mitter and receiver racks are align to each
other, and distance between the adjacent
racks is 9√2 meters. In case of FoV=10°, the
maximum received power is −15.67 dBm. Figure 10.10. Probability of Error for various
In Figure 10.10 shows probability error distances between the racks.
against various distant racks. Because the dis- Source:
tance between the racks across the room will AUTHOR'S SELF GENERATED
vary and referring Figure 10.3, the minimum
distance is 3 meter and the maximum distance
is 12.72 meters. In case of the distance of 3m
BER is very low and of the order of 10–45.
In case of diagonal racks with distance 3√2
meters the BER order is 10–13, and when dis-
tance between the racks is 6 meters the BER
order is 10–4 between the racks is distance 6
√2 the BER order is 10-2. The BER for the
distance of 9√2 is very high and of the order
of 0.2. Therefore, as the distance increases
the BER falls very rapidly, and thus to main-
tain low BER power must be increased. But
increase of power is not possible due to the
LED power limitations. The other possible
choice is decrement of FOV, an in case of
FOV=100, the BER is of the order of 10–200
up to distance of 15 meters.
5. Conclusion
Data centres are the core of modern day
communication, and as the amount of data
kept in DCs grows, the size of the data cen-
tre grows enormously, as does the amount of
heat released. As a result, it is now important
to develop a new communication paradigm
Figure 10.9. LoS communication between the where heat dissipation and amount of cabling
racks is 9√2 meters (a) FOV=70° (b) FOV=10°. required can be controlled. FSO is one such
Source: technology that can considerably reduce
AUTHOR'S SELF GENERATED
Cascadability analysis of free space optical wireless data centers 87
squandered power. The application of FSO Maxim Podlesny, Md Golam Rabbani, Qi
technology in DC is a relatively new concept, Zhang, and Mohamed Faten Zhani (2012),
with just a few preliminary investigations to Data center network virtualization: A s urvey.
date. In this work cascadability analysis of the IEEE Communications Surveys and Tutori-
DC racks is presented and BER is evaluated. als 15, no. 2, 909–928.
[8] H. Davidson et al., (2013), Data center with
It is found that using proper setting param-
free-space optical communications, US Pat-
eters the error free transmission is possible.
ent 8,483,569.
[9] J. Shin et al. (2012), On the feasibility of com-
pletely wireless datacenters. Proc. Archit. for
References Netw. and Commun. Sys, pp. 3–14,
[1] Srivastava R., Bhattacharya P., and Tiwari [10] K. Ramachandran et al. (2008), 60 ghz data-
A.K. (2019), Dual buffers optical based packet center netw.: wireless => worry less?, NEC
switch incorporating arrayed waveguide grat- Laboratories America, Tech. Rep.
ings, Journal of Engineering Research 7, no. 1. [11] S. Kandula et al. (2009), Flyways to de-
[2] Singh A., Tiwari A.K., and Srivastava R. congest data center networks, Proc. ACM
(2018): 19, Design and analysis of hybrid Hotnets-VIII.
optical and electronic buffer based optical [12] N. Riza and P. Marraccin I (2012), Power
packet switch, Sādhanā 43, no. 2. smart indoor optical wireless link applica-
[3] [3] Xu, Maotong, Chong Liu, and Suresh tions, Wireless Comm. and Mobile Comp.,
Subramaniam (2016) PODCA: A passive pp. 327–332N.
optical data center architecture, IEEE Inter- [13] Hamedazimi et al. (2013), Patch panels in
national Conference on Communications the sky: A case for free-space optics in data
(ICC), pp. 1–6. IEEE, 2016 centers, Proc. ACM Hotnets, pp. 23:1–23:7.
[4] Shukla, Manahar Prashant, and Rajiv Sriv- [14] Dixit A.K., Srivastava M., and Srivastava.
astava (2018), Arrayed Waveguide Grating R (2020) Physical layer analysis of opti-
and Re-Circulating Buffer Based Optical cal wireless data centers. Journal of Opti-
Packet Switc., Journal of Optical Communi- cal Communications, vol. 2020, pp. 1–10,
cations 1, no. ahead-of-print. https://doi.org/10.1515/joc-2020-0207
[5] Srivastava, Rajiv, and Yatindra Nath Singh [15] Dixit, Anand Kumar, Meenakshi Srivastava,
(2010) Feedback fiber delay lines and AWG and Rajiv Srivastava. (2023), Analysis of
based optical packet switch architecture, Opti- mimo optical wireless data center networks.
cal Switching and Networking 7, no. 2 75–84. International Journal of Information Tech-
[6] Srivastava, Rajiv, Rajat Kumar Singh, and nology 15, no. 1 (2023): 519–529, https://
Yatindra Nath Singh (2009), Design analy- doi.org/10.1007/s41870-022-01103-8
sis of optical loop memory. Journal of light- [16] Z. Ghassemlooy et al. (2012), Optical Wire-
wave technology 27, no. 21 4821–4831. less Communications: System and Channel
[7] Bari, Md Faizul, Raouf Boutaba, Rafael Modelling with MATLAB, 1st ed. Taylor
Esteves, Lisandro Zambenedetti Granville, and Francis.
11 Wireless network with machine learning
and deep learning techniques
Bhupal Arya1,a, Amrita Kumari2,b, and Jogendra Kumar3,c
1
Research Scholar,Quantum University Uttrakhand, Roorkee, India
2
Assistant Professor,Department of Computer science and Engineering, Quantum University
Uttarakhnad, Roorkee, India
3
Assistant Professsor,Department of computer science and Engineering,
GBPIET Ghurdauri pauri Garhwal,Uttarakhnad, India
Abstract: In the current scenario of wireless technology, we go through the many challenges and issues sur-
rounding the application and usage of tools, techniques of machine learning as well and deep learning algo-
rithms in wireless networks. In heterogeneous networks, designing the machine learning foundation routing
approach is challenging. At the beginning of this work, the fundamental basic concepts of (ML) machine learn-
ing and (DL)deep learning in different wireless networks are presented. Because of the constantly changing
network conditions present in many ad-hoc networks, preparing data collection and training with that data is
a particularly difficult process. In addition, this study gives an overview of several studies that used deep learn-
ing and machine learning techniques in a range of research areas, including networking, communications, and
dissipative settings. Finding possible issues and challenging tasks is the main objective of this survey activity.
Keywords: Deep learning; machine learning; routing algorithms; wireless sensor networks
1. Introduction ratio, etc. to figure out the network dynam-
ics, which include hotspots, incursion distri-
Machine learning in wireless networks, is a bution, congestion zones, traffic bottlenecks,
broader field that encompasses various tech- and spectrum availability.As a result, wireless
niques and algorithms for building models networks with numerous nodes and dynamic
which can trained and learn from data and connection quality may be analyzed using
make some decisions or predictions deep learning.
Deep learning can also be called as a sub-
part of machine learning that focuses on 1.1. Approaches for machine learning
neural networks with multiple layers (deep
neural networks).
in wireless networks
To handle accurate pattern detection i. Reinforcement Learning
from complicated raw data, deep learning ii. Deep Reinforcement Learning
is a promising machine learning technology. iii. K-Nearest Neighbors Algorithm
Deep learning makes use of several layers of iv. Bayesian Net/HMM
neural networks to extract fine, brain-like v. K-Means
properties from high-dimensional raw data. vi. Decision Tree
It is possible to use it to examine a variety of vii. Particle Swarming Routing
network measures, including latency, Apply viii. Deep Q-Learning
metrics like loss rate, link signal-to-noise ix. Q-Learning
[email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-11
Wireless network with machine learning and deep learning techniques 89
2. Machine Learning Techniques in 2D. Rate adaption problem-
Wireless Mesh Network a. Q-Learning
Depending on the suitability of a given b. LA (Learning Automata)
approach and a network issue, machine c. Bayesian
learning techniques may be utilized to handle
different management and design challenges. 2E. Solving the problem of fault
Most often, wireless mesh network design detection a) K-means 2F Routing
optimization issues may be solved using
reinforcement learning (RL). Three different Q-Learning—A machine learning technique
types of learning and their categories - called Q-learning makes use of reinforce-
ment learning to determine the worth of a
A. Supervised learning— given action in a certain state. The “Q” in
a. Support Vector Machine Q-learning stands for quality, which is how
b. ANN valuable an action is in maximizing future
c. DT (Decision Tree) rewards.
d. Perception
e. Bayesian 2F. Learning Automata
B. Unsupervised Learning—
a. K-means Learning automata-based processes have been
b. Principal Component Analysis C. applied to route optimization. The learning
Reinforcement Learning- automata principle is used by algorithms like
c. Q-Learning Steiner Connected Dominating Set, and LA
d. LA (Learning Automata) (Learning Automata)-based Multicast Rout-
e. MKP ing to provide an optional path from source
to destination. These problems include con-
Techniques used in solving problems in nection congestion, routing decision delays,
wireless mesh networks- and channel allocation.
2A. Solving the channel assignment 2G. Artificial Neural Network
problem An Artificial Neural Network (ANN), or
a. K-means simply a neural network, is a computational
b. LA (Learning Automata) model that is modeled after the architecture
and operation of biological neural networks
within the human mental state. Machine
2B. Solving the problem of routine
learning is a subset which involves neural net-
issue works.fundamental building block of deep
a. LA (Learning Automata) learning.
b. MDP
c. ANN
2H. Channel Allocation
2C. Improve the fairness in wireless Bayesian Learning- Using a testing item’s
input characteristics and the whole training
mesh network data set, Bayesian learning attempts to deter-
a. MDP (Markov Decision process) mine the posterior probability distribution of
b. Q-Learning the target attributes.
90 Innovations in Computational Intelligence: Trends and Technologies
2I. Network Deployment • Big data transport made easier by deep
learning
Mesh routers and mesh gateways (MGs)
• The difficult problem of employing deep
(MRs) must be placed properly during net-
learning for big data transports
work deployment to guarantee the network’s
• to build a thick routing pipe with the abil-
intended performance. The mesh routers and
ity to send a large number of packets per
gateways in wireless mesh networks may be
second.
deployed using Q- Learning and LA (Learn-
• Identifying the hop-by-hop connection
ing Automata) approaches.
failure
• MAC parameters selection to guarantee
2J. Anomaly/Intrusion Detection-[7] service delivery quality.
Decision Trees (DT)—The decision tree is a • Deep learning-based network swarming
well-liked machine learning technique for • to control node movement to achieve the
regression and classification. They’re a fun- appropriate swarming form and effective
damental part of supervised learning and communication architecture
have applications in various domains such • Management of nodes and cluster forma-
as finance, healthcare, and natural language tion in swarming networks
processing. Integrity and Faulty Detection- • Parameters that should be gathered from
design patterns
• Big data architecture selection
2.1. Deep Learning in the Wireless • Deep learning’s suitability in wireless net-
Network: Problems and works for particular Tasks coordination
Challenges among several CP controllers
• Optimization of the transport layer using • Implementation of distributed deep learn-
deep learning ing in wireless nodes
• the multi-queue evolution pattern detection • applying a particular deep learning algo-
• the multi-queue evolution pattern detection rithm component to a particular distrib-
• determining the length of the RED zone uted job
line • using MAC protocols to control wireless
• Mixing several protocols and schemes nodes so they may exchange input param-
with the congestion control scheme eters and output data
• Control of traffic along the whole route • Cross-layer design based on deep learning
from source to destination • To achieve cross-layer optimization
• The fundamental application layer for
deep learning
AQ1:
• Based on the performance objective of the
Please application layer, to design a less-com-
provide plexity deep learning model.
intext
citation • Deep learning leads to DRL
for Figure • applications for controlling cognitive
11.1 and
11.2
radio networks
• Deep learning-based swarming
AQ2: • to control node movement to achieve the
Please
provide appropriate swarming form and effective
source communication architecture
for all the Figure 11.1. Deep learning in wireless network.
figures
• Management of nodes and cluster forma-
Source: tion in swarming networks
AUTHOR'S SELF GENERATED
Wireless network with machine learning and deep learning techniques 91
• A fundamental application layer for deep suitability in wireless networks for par-
learning ticular tasks
• to specify a less-complexity deep learn- • Coordination among several CP controllers
ing model depending on the performance
objective of the application layer 2.3. Several machine learning
• Based on deep learning Cloud computing approaches forpacket routing
security in the dew
• applications for controlling cognitive q-routing-
radio networks Q-Learning [10] is a reinforcement learning
• DL/DRL implementations that are effec- algorithm that may discover the best course
tive on real-world wireless systems of action in a given situation to maximize
• Collecting Network configurations for the rewards from that situation. Distribu-
the input phases of deep learning may be tive routing of packets ina network is made
challenging. possible by Q-Routing, a modification of
• The wireless devices’ range restrictions Q-Learning.
• incomplete batches of training samples
• Optimization of the transport layer using 2.3.1. Ant-based routing
deep learning
• the multi-queue evolution pattern An innovative use of reinforcement learn-
detection ing called “Ant-Based Routing” is based on
• determining the length of the RED zone basic biological “ants.” These “ants” trav-
line erse the network and quickly pick up the
• Mixing several protocols and schemes best paths using the ant colony stigmergy
with the congestion control scheme model of communication as inspiration. This
• Control of traffic along the whole route method is more durable than conventional
from source to destination routing algorithms in that it is less affected
• Using deep learning to speed up massive by random route corruption while computing
data transfer is difficult problem employ- packet paths.
ing deep learning for big data transports
• To build a thick routing pipe with the 2.3.2. Particle swarm routing
ability to send a large number of pack- Drs. Russ Eberhard and James Kennedy
ets per second. Identifying the hopby-hop invented particle swarm optimization, a
connection failure population-based stochastic optimization
• MAC parameters selection to guarantee approach, in 1995. They learned from the
service delivery quality. Deep learning’s
AQ3:
Please
provide
Figure 11.2. ML using ANN and DNN. Figure 11.3. Layered ANN caption
Source: Source: 11.3–11.7.
AUTHOR'S SELF GENERATED AUTHOR'S SELF GENERATED
92 Innovations in Computational Intelligence: Trends and Technologies
social interactions of flocks of birds and under various circumstances and topolo-
schools of fish. Particle Swarm Routing starts gies. RL-Probe was successful with vari-
with a set of random fixes, followed by iter- ous link topologies and link quality
atively seeks for Optimal solutions through
updating subsequent generations.[3] 2.4. Problems and difficulties with
• The network’s connection may be greatly machine learning in a wifi
improved by using reinforcement learning network
techniques to manage node mobility and Machine learning techniques are now in use
packet routing decisions. widely used in many fields, such as image,
• The four primary categories of contem- speech, video, public safety, and medicine.
porary several types of routing protocols We now have a wealth of data stored in sev-
were developed for wireless networks: eral data centers and these are being analyzed
geographical, on-demand, reactive, and in real-time as a result of the development of
ML/DL-based. Outingtable-based proac- more advanced computer-related processes.
tive protocols. DL-based routing systems While sing machine learning methods and
have received a lot of attention recently techniques, it is possible to construct pre-
due to their enhanced performance in diction or human-like decision-making pro-
complicated networks. cesses. The system should be able to safely
• An innovative method named “RL-Probe” handle the whole planet on its own utilizing
was presented in for improving connec- machine learning techniques. For instance,
tion quality. RL-Probe was crucial in the system gathers all the data generated
boosting RPL’s functionality and capacity. by each person and learns everything about
RPL is a routing protocol for less power the environment on its own. The ability
and dissipative networks. To decrease the of machine learning systems to constantly
overhead, The RL-Probe approach is used learn from data over times one of its major
and is based on the reinforcement advantages. The system may be continually
Learning paradigm. updated while it is in use by using freshly
observed or created data. As a consequence,
• RL-Probe Using the RL-Probe framework, various logics are generated using various
both synchronous and asynchronous training sets using a single machine learning
LQE approaches are compatible. The rel- method. It implies that the system is adapt-
evance of each node is taken into consid- able in terms of how it makes decisions and
eration while making clustering decisions that it can continually learn from experience.
throughout the RPL route maintenance Referring back to the example, the system
and route recovery operation. can sound alert or take any other action that
• Asynchronous probing approach is sug- may be regarded intelligent and beneficial
gested to measure the RSSI and ETX. A when it identifies any harmful or anomalous
MAB model is used to accomplish the situations from the world of humans. Cre-
decision- making process. With the use of ating this ideal system, however, is difficult
the RL-Probe technique, reward function in practice. The difficulty of obtaining trust-
has been utilized to evaluate the trends in worthy training data that covers a variety of
link quality fluctuations. RL-Probe was contexts gives rise to the first problem. To
assessed using three performance assess- build any decision or prediction model with-
ment matrices: Packet loss rate, packet out an appropriate data set, regardless of
overhead on top of the usual energy con- how strong our machine learning algorithms
sumption. Evaluation work was done are at Building robust logic and analyzing
Wireless network with machine learning and deep learning techniques 93
state may be examined from a broad view-
point using a variety of bits of information,
and problems that could seriously harm net-
work performance can be discovered before-
hand. In sensor networks, several unusual
situations need to be recognized beforehand.
For example, defective application layer sen-
sors may constantly broadcast false detected
Figure 11.4. Anomaly Detection values to a connected server. If we are igno-
Source:
rant of this, we won’t be able to pinpoint
AUTHOR'S SELF GENERATED the exact location of the accident. Ona net-
work, each node makes decisions just based
large data sets in some real-scenario while on its local knowledge, even while RPL cap-
continuously training the model and system. tures the network issue and the topology
Recently, researchers and businesses have is continuously updated while considering
been increasingly curious about the Internet network conditions. The problem might not
of Things. The emergence of several research be identified until it is more significant if
teams as a result of this trend has sped up network traffic is concentrated on a single
the development of critical methods. Com- node. The node with the strangely unequal
munication and Networking techniques, and high workload [11] in this scenario will
particularly for less power and dissipative quickly exhaust its energy before being shut
networks, have drawn a lot of attention in off. On the connection layer, a wireless net-
comparison to other networks because of work interface may malfunction. Due to the
the constrained properties of low-power network’s global view, all of these factors
methods and constrained devices that are may be discovered in this case before the cat-
not the same as those of traditional, anti- astrophic calamity happens. The complete
quated sensor-based networks. picture may be obtained by collecting data
Routing protocols like as RPL and specific and monitoring traffic from the server or
mac protocols such as Slotted time wise and the potent root node. The list of defect cases
Channel Hopping are created and are being that need to be Compatible. The relevance of
widely utilized as standard protocols to boost each node is taken into consideration while
the dependability of communication on less making clustering decisions
power and dissipative networks built with
restricted equipment. Simply said, RPL pre- • Traffic overload due to attack [11]
vents routing loops by producing DODAG • Energy use as a result of an uneven traf-
structures and straightforwardly building fic load RL-Probe. We know that one
routing paths. Additionally, the topology benefit of using ML algorithms is that
created for route management is constantly the method learns diverse conditions,
evolving while taking into account various including inexperienced ones, through-
network data and circumstances. TSCH ena- out time, since the rationale for these
bles the use of several channels as well as decisions is not dependent on estab-
the resolution of Intrusion-related problems. lished static rules. Similar to the above,
However, these processes are carried out it seems advantageous to have a strategy
depending on regional information gath- utilizing machine learning algorithms
ered at every node. Hence, the issues that when we create methods for identify-
are not promptly identified and resolved by ing abnormal circumstances like those
the localized protocols. A network’s current mentioned above on less power and
94 Innovations in Computational Intelligence: Trends and Technologies
dissipative networks. It is difficult to get s ystem—picked up on dissipative, less power
the data set needed to train the model, networks is provided below.
though. Also, compared to other net-
works, the network environment in a • An energy-deficient node
less power dissipative network is quite • a wireless network’s interface is
dynamic. As a result, it is almost hard to malfunctioning
collect a solid data collection that cov- • Intrusion on a specific channel [7]
ers a range of circumstances. • CPU consumption that is too high on a
node
• A node’s memory or buffer is full.
3. Machine Learning in Networking • Strange perceived value
and Communication • both synchronous and asynchronous LQE
approaches are
We provide a number of papers that use
machine learning methods in networking or
communications. These techniques provide 3.2. Data gathering and
a significant contribution to the application administration
of machine learning to solve difficult issues
The authors stress the need to comprehend
brought on by changing and unexpected
the various data kinds that may be gathered
wireless channel status. Despite the fact that
in SDNs as well as the method for extracting
these works have significantly improved per-
information from those data. To take the is
formance when evaluated, some of them only
the beginning of the journey toward network
focus on trace-driven testing or simulation
control based on machine learning. Article
using the same data set as training. Yet, these
outlines a fundamental architecture set up
efforts are important in that they set the way
in a commercial network that leverages the
for leveraging a variety of data to create trust-
Open flow protocol to capture traffic statis-
worthy models to get around the challenge
tics. However, this study mainly focuses on
of foreseeing the future condition of wireless
evaluating the keeping an eye on and catego-
channels.
rization of traffic using data received through
the Open flow protocol and traffic categori-
3.1. Signal classification zation for network.
To operate the network automatically or The signal provided by a sender needs to
autonomously, we want to build a solid be accurately detected at a receiver side in
order to execute dependable wireless com-
munication. In the authors [3] concentrate
on the effects of modulated signal Intrusion
[7] and actual wireless channel circumstances
on classification performance. They suggest
using machine learning to categorize the sig-
nal in a realistic wireless environment. As
our study classifies the signal that has been
altered and is now traveling across a wireless
channel, we consider it to be akin to pattern
recognition does not provide any advice for
operating large-scale MLbased systems or
Figure 11.5. Multi Layered Deep Neural Network networks. However, our study.has provided
Source: clear guidance on how to deploy ML-based
AUTHOR'S SELF GENERATED
Wireless network with machine learning and deep learning techniques 95
network management and made it practical decide on cloud computing. User input, device
to do so. energy level, and CPU use level are additional
input values that are distinct and steady in
3.3. Network attack forecast contrast to values affected by a dynamic and
unpredictable wireless network.
In the study, a technique based on machine
learning technology for setting security rules
on the SDN controller has been proposed. 3.6. Network of wireless sensor
Security settings for the SDN controller are networks
configured to block the whole subnetwork
to prevent access from possible attackers, 3.6.1. Using machine learning
and machine learning methods are utilized In this area, we present several works that
to identify likely target hosts that can be utilized sensor networks [14] and machine
attacked. The identical datasets were divided learning approaches. These techniques also
into training and test sets in order to evaluate make a significant contribution to the appli-
the suggested strategy. cation of machine learning to solve difficult
issues brought on by dissipative channels and
3.4. Wireless adaptive swimming restricted devices. These studies are important
because they opened the way for the develop-
Network conditions vary greatly depend- ment of trustworthy models that can be used
ing on the situation and alter over time. It to circumvent the challenge of forecasting the
is therefore difficult to predict the network’s future state of dissipative channels.
future condition. Despite the fact that a num-
ber of rate adaptation methods for high-
quality video streaming have been proposed, 3.7. Diagnostics for channel errors
there is still significant room for advance- Many protocols, including 802.11, 802.15.4,
ment. Numerous research has validated the 802.15.1, etc., share the ISM band. Here,
use of machine learning techniques to video Intrusion between several systems impairs
streaming services. It should be possible to communication quality. In-depth studies were
create a system that selects the best bitrate done by the authors to examine the error pat-
and CDN for the streaming user, learns cru- terns in IEEE 802.15.4, and they discovered
cial parameters, and maximizes QoE. Rein- that there are several patterns for the most
forcement learning was used in addition to common wireless situations. They devel-
taking into account the bandwidth, buffer oped a machine learning technique based on
level, and video rate to automatically create this discovery to categorize wireless channel
the best ABR algorithm. faults into distinct groups and offered the
system that can identify various issues in IoT
3.5. Mobile cloud offloading networks.
The paper presents a dynamic method based
on ML and introduces the usage of cloud 3.8. Spectrum choice
computing for offloading computation from The article also highlights ISM band pollu-
mobile devices. To decide whether to offload tion and restrictions on sensor node power.
computing to the cloud, it keeps an eye on To overwhelm this challenging environment,
network characteristics and device resources. it suggests a ML based channel allocation
Network data, such as available bandwidth, and selection method. Using ML, the sys-
is just one of many input variables employed tem predicts several expected transmission
in this work’s machine learning approach to attempts. This will use performance data like
96 Innovations in Computational Intelligence: Trends and Technologies
RSSI and LQI from the most recent packet test subject had a wearable sensor [14] that
received as well as the following features allowed him to identify himself in the wire-
as input information: the quantity of trans- less sensor network.
mission attempts, the reasons behind each
unsuccessful attempt, and RSSI and LQI. The 3.11. Detection of events
optimal channel from the output is one with
a less projected number of transmission tries. Wireless sensor [12] networks have a wide
range of applications. The initiative focuses
on locating pipeline leaks in the transporta-
3.9. Outlier detection tion networks for gas, oil, and water. Using
Wireless sensor [12] networks with a small algorithm-like pattern recognition, it trains
number of nodes are vulnerable to Intrusion the sensor network to distinguish novel traces
[7], unstable channels, and cyber interven- of events, such as leakages. Here, some scat-
tion, which reduces system performance and tered sensor nodes collaborate to estimate the
increases the possibility that false data may size of the leakage event. Despite the employ-
be relayed to higher management levels. This ment of a wireless sensor network in this
might have major consequences for indus- investigation [11], the difficulties associated
trial automation systems and sensor network with using an unreliable and unsafe wireless
systems used for public safety. The existing channel were scarcely considered.
works for outlier identification, according
to their authors, take a lot of memory, a sig- 3.12. Fault detection
nificant amount of energy, power for com-
Data gathering from the sink node is impre-
putation, transmission overhead, and data
cise due to the less- cost sensors and other
transfer capacity. They suggested employing a
problems caused by unreliable wireless net-
multi-agent framework and a machine learn-
works. Fault data must be observed in order
ing approach for online outlier’s identifica-
to monitor the system and respond quickly
tion to tackle the issue
to incidents. The cause of the issue must also
be identified. The paper proposed a statistical
3.10. Invoremental locatization technique for identify and classify issues in a
wireless sensor [14] network while using the
In general, GPS is one of the most widely used
Machine learning techniques. System faults
methods for object localization. Finding an
and data faults were used to categorize various
object’s specific location is difficult because
fault types. System faults are the other fault
of the limited GPS signal strength inside of
types brought on by less battery, calibration,
buildings. Due of this, other tactics are used.
For example, a large number of nodes are
used as anchor points which estimate the
comparative location of a target object. A
precise interior localization system must be
created in order to increase safety in under-
ground mines or caverns. Unfortunately, the
wireless channel continues to experience
Intrusion, which lowers the estimation’s
accuracy. To find the method with the fewest
errors and evaluate its performance, the study
used two alternative architectures and seven Figure 11.6. Wireless Network Communication Path
different machine learning techniques. The Source:
AUTHOR'S SELF GENERATED
Wireless network with machine learning and deep learning techniques 97
communication, or connectivity issues. Data collect training data in a variety of settings.
problems are those that are caused by dam- Due to the unpredictable behavior of wireless
aged or inoperable sensors. The authors of networks, creating training data sets may be
have examined machine learning-based flaw challenging.
identification in analogous settings.
5. Conclusion
3.13. Routing
Our study has uncovered a wide range of
Multi-hop routing systems [19] could utilize methods and protocols that might be used to
less energy with the use of machine learning. improve wireless networks and communica-
The work provided a Machine Learning- tions. Based on the literature, we examined
based clustering technique to efficiently place a variety of problems and difficulties with
the nodes(sensor) in the closest cluster. The applying machine learning and deep learn-
researchers applied ML to the wireless sensor ing algorithms to improve various opera-
network’s routing protocol. The suggested tions in networking in various ways. For
routing technique aims to extend network tasks like network administration, network
life and move data packets as quickly as fea- optimization, signal, Data management,
sible. This suggests that ML approach adop- channel assignment, network security, route
tion on WSN is advantageous for resource selection, etc. Machine learning and Deep
management. learning algorithms perform incredibly well.
Our study has uncovered a wide range of The two primary learning methods that per-
methods and protocols that might be used to form well for network operations are deep
improve wireless networks and communica- reinforcement learning and deep learning.
tions. Based on the literature, we examined However, it could be challenging to collect
a variety of problems and difficulties with training data in a variety of settings and it
applying deep learning and machine learning is not possible in this research to complete
algorithms to improve networking operations all the algorithms and methods of ML and
in various ways. For tasks like network admin- Deep learning which we can cover in future
istration, network management, optimization scope Due to the unpredictable behavior of
signal management, network security, opti- wireless networks, creating training data
mum route selection. Machine learning and sets may be challenging
Deep learning algorithms perform incred-
ibly well. The two primary learning methods
AQ4:
that perform well for network operations References Please
are deep reinforcement learning and deep change
[1] Johnson, A. and Smith, B. (2016). Enhancing
learning. However, it could be challenging to the refer-
Data Delivery Efficiency in AODV-Based Ad ences into
Hoc Networks. Wireless Communications Num-
bered
and Networking Conference (WCNC), 1–6. references
[2] Lee, C., and Park, D. (2017). Performance as per
Evaluation ofAODV in Dynamic Mobil- style,
ity Scenarios. IEEE Transactions on Mobile
Computing, 16(8), 2134–2147.
[3] Emilio Ancelotti, CarVallati, Raffaele Bruno,
Enzo Mingozzi, A reinforcement learning-
based link quality estimation strategy for
RPL and its impact on topology manage-
Figure 11.7. ment, Computer Communications, Volume
Source: 112, 2017, Pages 1–13.
AUTHOR'S SELF GENERATED
98 Innovations in Computational Intelligence: Trends and Technologies
[4] Martinez, F., and Rodriguez, G. (2018). Large- Scale MANETs. Wireless Personal
Throughput Optimization in AODV Net- Communications, 113(1), 1–17.
works for IoT Applications. Ad Hoc Net- [16] S.A. Haque, M. Rahman, S.M. Aziz, Sensor
works, 75, 60–72. anomaly detection in wireless sensor net-
[5] Simi S, Sruthi Ann Varghese,“Enhance QoS works for healthcare, Sensors15 (4) (2015)
by LearningData flow rates in Wireless Net- 8764–8786.
works using Hierarchical Docition”, 4th [17] Zhang, W., and Chen, X. (2021). Analyz-
International Conference on Eco-friendly ing Mobility Models and Their Impact on
Computing and Communication Systems, AODV Performance. Journal of Network
ICECCS 2015, Procedia Computer Science and Computer Applications, 185, 102968.
70 (2015) 708–714, Elsevier [18] Garcia, M., and Lopez, A. (2023). Secure
[6] Kim, H., and Jung, I. (2019). Enhancing Routing with AODVin MANETs: A Crypto-
Security in AODV Routing Protocol for Ad graphic Approach. International Journal of
Hoc Networks. Computer Communications, Network Security, 25(4), 692–704.
135, 87–98. [19] Rahman, A., and Khan, M. (2023). Exploit-
[7] Wu, J., and Li, K. (2020). Performance Eval- ing Cross-Layer Design for Improved AODV
uation of AODV in Heterogeneous Ad Hoc Performance in VANETs. Ad Hoc Networks,
Networks. Wireless Personal Communica- 123, 102377.
tions, 112(3), 1519–1532. [20] Wei Wang; Xiaohong Guan; Xiangliang
[8] A. Garofalo, C. Di Sarno, V. Formicola, Zhang,“Profiling program and user behav-
Enhancing intrusiondetection in wireless sen- iors for anomaly intrusion detectionbased
sor networks through decision trees,Depend. on non-negative matrix factorization” 43rd
Computing. (2013) IEEE Conference on Decision and Control,
[9] Rahman, S., and Khan, T. (2022). Perfor- 2004. Issue Date: 14–17 Dec. 2004, On
mance Enhancement of AODV Through page(s): 99–104 Vol.1
Dynamic Parameter Adjustment. Interna- [21] Hazem M. El-Bakry, Nikos Mastorakis A,
tional Journal of Wireless Networks and- “Real-Time Intrusion Detection Algorithm
Communications, 14(1), 53–64. for Network Security,WSEAS Transactions
[10] Zhang, Y., and Wang, Q. (2016). A Survey of on communications, Issue 12,Volume 7,
Energy-Efficient Routing Protocols in Wire- December 2008.
less Sensor Networks. Journal of Internet [22] HR, M., MV, A., S, G. and S. V., “Devel-
Technology, 17(3), 471–482. opment of anti-phishing browser based on
[11] S. Kosunalp, A new energy prediction algo- random forest and rule of extraction frame-
rithm for energy- harvesting wireless sen- work” Cybersecurity, 3(1), 2020
sor networks with Q-learning, IEEEAccess 4 [23] Saxena, S., Shrivastava, A. and Birchha, V.,
(2016) 5755–5763. “A Proposal on Phishing URL Classification
[12] Wang, H., and Li, Q. (2018). An Improved for Web Security,” International Journal of
AODV Algorithm for Load Balancing in Computer Applications, 178(39), pp.47–49,
Mobile Ad Hoc Networks. Wireless Per- 2019.
sonal Communications, 99(3), 1987–2002. [24] G. B. Huang, H. Zhou, X. Ding and R.
[13] Nayak, p. and G.K.Swetha (2021) “Routing Zhang, “Extreme Learning Machine for
in wireless sensornetworks using machine Regression and Multiclass Classification,”
learning techniques”, measurement Volume in IEEE Transactions on Systems, Man, and
178, June 2021, 108974 Cybernetics, Part B (Cybernetics), vol. 42,
[14] Qian Mao, Fei Hu, and Qi Hao, “Deep no. 2, pp. 513–529, April 2012.doi: 10.1109/
Learning for Intelligent Wireless Networks: TSMCB.2011.2168604.
A Comprehensive Survey”, IEEE communi- [25] A. Qayyum, A.S. Malik, N.M.Iqbal, M.
cations surveys and tutorials, vol. 20, no. 4, Abdullah, M.F. Rasheed, W., Abdullah,
fourth quarter 2018, pp. 2595–2621 T.A.R. and M.Y.B. Jafaar,., Image classifi-
[15] Li, X., and Wu, Y. (2020).On the Per- cation based on sparsecoded features using
formance of AODV Routing Protocol in sparse coding technique for aerial imagery:
Wireless network with machine learning and deep learning techniques 99
a hybrid dictionary approach. Neural Com- AODV Performance. Journal of Network
puting and Applications, 2017 pp.1–2 and Computer Applications, 185, 102968.
[26] Yin C, Zhu Y, Fei J, He X. A deep learn- [28] Altaber, A., “Phishing Websites Classification
ing approach for intrusion detection using using Hybrid SVM and KNN Approach,”
recurrent neural networks. IEEE Access International Journal of Advanced Com-
2017;5:21954–61. puter Science and Applications, 8(6), 2017.
[27] Zhang, W., and Chen, X. (2021). Analyz- [29] KSII Transactions on Internet and Informa-
ing Mobility Models and Their Impact on tion Systems, Robust URL Phishing Detec-
tion Based on Deep Learning, 14(7), 2020
12 Assessing technology innovation of
cellular health
M. Vasumathi Devia, Y. Muvvanab, Y. Sai Likhitha, M. Gowri
Bhavanic, and M. Vishnu Priyad
Computer Science and Engineering, Vignan’s Nirula Institute of Technology and Science for Women,
Guntur, India
Abstract: The conventional landscapes therapeutic treatment approach has changed in recent years due to the
arrival of cell phone applications, or heeltaps. The present heeltaps are designed to investigate the skills of scien-
tific specialists to determine those who are reliable, better, and well-liked ones. We can talk about the method of
integrating mHealth advances into clinical practice and figuring out whether professionals want to accept such
through the application of the procedure of normalization hypothesis. We looked at app attributes and perfor-
mance data, completed assessment reports on characteristics, capabilities, and feedback from the audience, and
used data analysis to evaluate feature cross-correlation and market alternatives. There is a possibility of devel-
opment regarding the which are already accessible the marketplace, as we discover unsatisfactory specifications
for workmanship and dependability. Strong ties exist between them.
Keywords: app features, new technology uptake, mobile health, and fitness app (health app), normalization
process (NPT), reliability (UX), and ease-of-use (UI)
1. Introduction States increased by 123%, from $2.6 trillion
to $3.2 trillion [3]. Twenty-one to forty-seven
Concerning healthcare Clinicians and percent of this enormous expense [4] can be
researchers alike are interested in the vast attributed to inefficiency in the form of non-
amounts of patient demographic, treatment fee induced duties such unfit assignments, as
plan, cost, and insurance coverage data being appropriate missions’ usage of antimicrobial
generated by the booming healthcare indus- agents, as well as deception. Researchers dis-
try. Recent years have seen an increase in the covered that 251,454 patients perish in the
number of scholarly articles focusing on the US every year. annually because of physician
out-of-the-ordinary applications of health- malpractice [5], so some of these costs may
care information mining technologies. Never- be attributable to subpar care. The move to
theless, the absence of a clear and reasonable a value-based healthcare company is fraught
explanation prompted us to create an over- with obstacles, but it may be smoothed down
view of the literature Investigation into the with better decision-making in accordance
subject. with accessible data [4]. Statistics creation as
Many countries, the healthcare industry is part of a control system is becoming increas-
thriving [1]. Costs, inefficiency, low quality, ingly popular in the healthcare industry [6].
and increasing complexity are only some of the Every day, this device collects a tremendous
difficult outcomes of this expansion [2]. From amount of data. Analytics provides methods
2010 to 2015, healthcare costs in the United and tools for deciphering these intricate and
[email protected]; [email protected]; [email protected];
a
DOI: 10.1201/9781003606635-12
Assessing technology innovation of cellular health 101
copious records [2] to produce numbers that for the Introduction with Health Data
can inform negotiating choices. Recently, health data has received increasing
attention for both healthcare and data
mining. PARK, 2018 [4] Abstract In this
2. Literature Survey paper we discuss the challenges in analyzing
and mining health data to find clinical insight.
Somayyeh Zakerabasali, Seyed Mohammad, This paper reviews major challenges, the
Tayebeh Banias Adi, Azita Yazdani and Sha- sources of data, meth- odologies and
habuddin Abhari, 2021 others [1]. Many technologies along with open issues in big
mobile health (mHealth) apps are being data analytics in health care.
AQ2:
integrated into the healthcare system, but Yujun Ma1, (Member, IEEE), Yulei Please
their potential success depends on uptake by Wang1, Ju Yang2, Yiming Miao2, And Wei check the
sentence
healthcare providers. Some of the important Li2 (2017) [ 5]]
is correct.
benefits associated with mHealth technology The paper talks about the challenges of a
are increased efficiency, cost breakdowns and medical and health care system, i.e. uneven
enhanced health care service delivery pro- distribution of medical resources in rural
cesses. Future research is needed to explore areas, explosive growth in chronic diseases,
additional barriers at the healthcare system and rising medical expenses require medical
level i.e. legal, reimbursement/ accountable reform approach to a good solution. This rec-
care organizations, economic and financial ommends us Health Internet of Things (IoT)
factors, health system policy and standardisa- along with big data to deal with such prob-
tion so that systematic interventions can be lems. This paper demonstrates a large-scale
implemented to improve access to mHealth health application system that utilizes the
technology. [2] Ayan Banerjee Member, health IoT, big data architecture, key technol-
IEEE, and Sandeep K. S. Gupta Senior Mem- ogies and typical application of health IoT.
ber, IEEE~~2014~ Abstract: While smart The paper proposes to examine the health
mobile medical computing systems (SMDC- status of non-hospitalized patients using a
Ses) exploit context information to provide Wireless Body Area Network (WBAN) sys-
healthcare services, unexpected changes of tem with load-sharing processing capabilities
the clinical context may result in incorrect on both private and public clouds. [SHAKER AQ3:
or incomplete service provision, generating EL-SAPPAGH 1,2, FARMAN ALI 2, SAMIR Please
check the
faults and violating requirements. To mitigate EL-MASRI3, KYEHYUN KIM4, AMJAD sentence
this issue, this paper presents an approach to ALI 2,5, AND KYUNG-SUP KWAK 2, is correct.
analyze SMDCSes that accounts for dynamic (Member,...2019 [6]] Establishment of smart
changes in context and the interaction with environment and integration of used compo-
the physical world which may exist. nents is the main research issue in a newly
Achilleas Papageorgiou, Michael Strikes, developed network as well standardization
Eugenia Politou, Efthimios Alepis, Constan- network must have a single architecture,
tinos Pataki’s [1], 2018 [3]. The paper seeks protocol, and application program interface
to offer broad feedback to the developers of to connect between heterogeneous smart
these apps to shore up their security/privacy objects. To deliver MH apps and solutions
features while pointing out regular mistakes to address serious medical problems that are
made in the application development life easier to use with increased precision.
cycle that can threaten user privacy rights. [Sabyasachi Dash1†, Sushil Kumar Shakya AQ4:
AQ1: [Gaspard Harerimana, (Student Member, war 2,3†, Mohit Sharma 4,5 and Sandeep Please
Please check the
check the
IEEE), Beakcheol Jang, (Member, IEEE), Jong Kaushik6*,2019 [7]] The paper analyses the sentence
Wook Kim, (Mem- Ber, Ieee), And Hung Kook] data in health care through their necessity of
sentence is correct.
the paper provides a deep understanding
is correct. collectiveness.
102 Innovations in Computational Intelligence: Trends and Technologies
[8] Sohail Imran, Tariq Mahmood, Ahsan All the way through the innovation imple-
Morshed and Timos Sellis (Fellow, IEEE), mentation process, the NPT comprises the
“Enabling Big Data Analytics In TheHealth- following four key additives:
care Sector: Challenges And Neuro-Fuzzy
Based Solutions”, 2021 1. “Coherence” requires knowing the cost,
Mobile Health (mHealth) apps for moni- importance, and uniqueness of the action
toring health status of patients with chronic or by itself “identity the experience-
medical conditions are increasingly becoming creating the desire.” because the method
prevalently used [Gerson, Rodriguez, Devasena, called “experience-making” “paintings
Inupakutika, Sahak, Kaghyan, David Akopian, that creates and organizes an exercise as
Palden Lama, Patricia Chalela, Amelie Ramirez a cognitive category, it’s seen as an assort-
et al.,2020 [10]]. In this paper, the authors dem- ment of competencies this includes sepa-
onstrate the methodology of assessing perfor- ration from other hobbies and a common
mance of a developed prototype model mHealth experience of intent [56].
app for breast cancer patients on EHT. 2. “Cognitive involvement” requires that the
main players in the intervention design
and upkeep.
3. Conceptual Structure 3. “Collective motion” influences kinship
In facts structures (IS) studies and notably in between expert groups and conforms to
technology adoption studies, knowledge of the overarching organizational contextual
the goals or dreams for phone clients is cru- content of objectives, morale, manage-
AQ5:
Please cial to clarify the purpose of usage, satisfac- ment, and assets.
note tion, involvement [53].
that the
Implementing new clinical practices and Fourth, “reflexive tracking” involves eval-
reference
numbers implementing innovations in healthcare uating the new intervention for its efficacy
[53]
requires sophisticated organizational strate- and practice.
onwards
is not gies and sociological understanding of change. Desk I provide an empirical explanation of
occurred So, we took a sociological concept from the NPT methods and components available
in refer-
ence list. the field of generation called the “NPT” [54]
and applied it to the field of technology to
explain how healthcare systems have incor-
porated technological and organizational
changes. This theory sheds light on the
human and organizational linkages behind
the spread of innovative interventions into
everyday practice [55]. Structured inter-
views and questionnaires have been utilized
with the NPT represents the core concept to
examine adoption within health. Us utilize
the NPT to understand value for medical
experts apply a novel way to practice the use
of medicinal drugs, as well as the importance
of app features in attracting and retaining
experts. Engagement, or user engagement, is
the worth of the consumers consider follow AQ6:
interacting app, which is tied to longevity for Please
good application characteristics this stimu- Figure 12.1. Conceptual framework. provide
source
late it is usage. Source: for all the
AUTHOR'S SELF GENERATED figures
Assessing technology innovation of cellular health 103
Table 12.1. Mechanisms of NPT of mHealth in medicine
AQ7:
Please Contracts of NPT[57] Definition [58], [55| Mechanisms of Implementation of
provide mHealth apps for professionals
in-text
citation Coherence Participants’ understanding of the mHealth App Categorization
for table Sense-making intervention and its differentiation
12.1
Cognitive Others’ influence and Users’ Reviews and App Quality
Participation Effort participants’ commitment to Evaluation
work with the intervention
Collective Action Fit with overall context including Trialability and App Feature Relevance
AQ8:
Please Commitment goals, morale, trust and resources
provide Reflective Monitoring Participants’ evaluations and Users’ Engagement and App Popularity
source
for all the Appraisal appraisals of the intervention
tables Source: AUTHOR'S SELF GENERATED
during mHealth app deployment. As a result, Table 12.2. Research questions and
the NPT details how to embed (implement) methodology
novel artistic routines in existing (sustain- Research Questions Research Approach based
able) societal frameworks (integration). on XPT
You can see how this three-degree NPT RQ1: What are NPT constructs
method works in practice in Figure 12.1. The the app categories, • Implementation
4 NPT mechanisms are defined by the various features, quality stage: app Categories
categories of mHealth apps and the require- and popularity that (coherence)
ments they satisfy. Positive app categories are affect the adoption • Embedding stage:
developed primarily based on their abilities intention of mobile app Feature relevance
(the need for intervention). Customers can app innovation (collective action)
verify their hopes that the app will meet their by health • Integration stage: app
needs by testing it out before they fully accept professionals? Quality (cognitive
the innovation, which is part of the process of participation), app
“implementing” the innovation. The second Popularity (reflexive
monitoring)
level of the system is called “embedding,” and
Findings: Tables 3. 4 and 5
it outlines the mechanisms that can be uti-
lized to carry out new practices; The systems RQ2: What app • Mean differences of
in the instance represent the abilities., that metrics are affected app “integration"
by the inclusion of metrics based on
need to be met, of the apps. Like the idea of
app features? feature inclusion
a social capital, “embeddedness” consists of Findings: Table 7
relational characteristics like norms and trust,
RQ3: To what • Correlations between
levels can be vital for organizations like medi-
extent are the the NPT dynamics
cal care that incorporates social improves this
attributes of Findings: Tables 8a and 8b
frequently necessitate deep comprehension of innovative
intricate societal issues [59]. adoption of
inHealth by health
professionals
4. Methodology intercorrelated?
The writer’s scheme is to use NPT to catego- Source:
rize and describe the mHealth applications
currently available to clinical professionals
and scientific students, as well as to investigate
104 Innovations in Computational Intelligence: Trends and Technologies
the features that indicate the apps’ reliability f unctionalities. Based on their causes, we next
and dependability. located the variables (such as app categories
App discovery and selection are the focus and app functions) inside the NPT ranges
of the first stage. A limited selection of fit- (Figure 12.1). To assess app trust worthi-
ness apps is optimized for iPhone use, but ness during the NPT’s “embedding” part, we
Android apps downloaded from the Play sought to clarify how feature relevance might
Store account for the largest market both explain safety, dependability, and value issues.
domestically and abroad [62]. While only We assessed the “sustainability” and “inte-
5% of the highest-grossing applications need gration” of mobile health apps into standard
money to download, about 95% of them are medical procedures. using their quality (as
free to download, have protected in-app pur- measured by the MARS score) and popular-
chases, or provide a limited-time free trial. ity (as measured by the app score in Google
This email only included reviews of free apps. Play). The MARS was utilized for a success-
Studies [6, 16, 62] have articulated compa- ful assessment. After downloading and using
rable objectives. The distribution of pertinent each app, as well as reading user reviews from
apps was ascertained by looking through a the Google Play store, two researchers with
wide variety of search phrases in the Play extensive experience studying health apps
store. With the keywords “analysis,” “medi- determined their overall ratings.
cal,” “clinical,” “illnesses,” and “signs,” eight A reliable measure of the degree of inte-
different searches were carried out. Through gration is the app’s recognition (based on
March 2020 for app selection was around 4 user ranks and downloads), which explains
months. Data from the chosen apps was col- engagement and shows that the applications
lected in an Excel report, and categories were are Professionals’ NPT adoption standards
established by content analysis. Included for mHealth apps were determined using
in this method were things like reading app popular and frequently used app metrics
store descriptions and customer reviews, (such as downloads and star ratings).
installing the app on an Android device, and
using it to get more information, and keep-
ing track of the app’s overall rating. The 5. Results
outcomes section presents both descriptive The bar chart beneath displays the number
data (such rating and downloads) and their of crises, The tar, and trauma considering
technical components (like sharing options,
terms and conditions, and a consent form).
To present the sample’s descriptive and infer-
ential statistics, we employed IBM’s SPSS
statistical analysis software. Two researchers
investigated mobile applications to guarantee
clarity in the categorizing law. at some point,
disagreements about how the apps should be
categorized and evaluated were discussed and
settled by mutual agreement.
After analysing the NPT contracts, we
created the associated reassuring variables
(Table 12.2), which we then used to design
the second and third tiers of our technique
(Figure 12.2). Next, we classified the 168
apps using Excel spreadsheets in accord- Figure 12.2. Methodology schema.
ance with their different categories and Source:
AUTHOR'S SELF GENERATED
Assessing technology innovation of cellular health 105
6. Conclusion
In this post, we tallied the benefits and chal-
lenges of mHealth apps and briefly discussed
each. To guarantee a more complete under-
standing of the modern adoption process, we
provided bodywork and taxonomies. mHealth
apps aimed to maintain a high standard of
safety, dependability, and utility by including
functions that guarantee these characteristics.
Medical professionals, the fitness industry,
AQ9:
Please
and society at large might all benefit greatly
provide from the creation of cutting-edge mHealth
in-text apps that make full use of the capabilities of
citation
for Figure today’s cutting-edge technologies.
12.3 and Figure 12.3. Types of admission.
12.4. Source:
AUTHOR'S SELF GENERATED
References
[1] Montiel, I. Delgado-Ceballos, N. Ortiz-de-
Mando Jana, and R. Antolin Lopez, “New
ways of teaching: using generation and cel-
lular apps to teach on societal grand chal-
lenges,” J. Bus. Ethics, vol. 161, no. 2, pp.
243–251, 2020.[1] I. de Camargo, “The
medical doctor’s digital path to treatment,”
Google, NY research, 2012.
[2] C. L. Ventola, “Mobile gadgets and apps
for fitness care experts uses and blessings,”
Pharmacal. Therapeutics, vol. 39, no. 5,
2014, art. no. 356.
[3] S. Chatterjee, S. Chakraborty, S. Sarker,
S. Sarker, and F. Y. Lau, “Examining the
achievement factors for cell paintings in
healthcare: A deductive observe,” Deci’s.
help Syst., vol. 46, no. 3, pp. 620–633, 2009.
[4] J. Goldhahn, V. Rampton, and G. A. Spinas,
Figure 12.4. City Code Patient Count.
“Should synthetic intelligence make docs
Source: AUTHOR'S SELF GENERATED obsolete?” BMJ, vol. 363, 2018, artwork.
no. k4563.
the level of severity of the illness as extreme, [5] P. Galetsi, ok. Katsaliaki, and S. Kumar,
mild, and moderate. “Values, demanding situations and future
Based on the data previously, we can deter- instructions of huge statistics analytics in
healthcare: a systematic evaluation,” Social
mine that Tra patients dominate other types
Sci. Med., vol. 241, 2019, art. no. 112533.
of admissions.
[6] H. Wisniewski et al., “Understanding the
Based to the bar chart below, the metro- fine, effectiveness and attributes of pinnacle-
politan codes client count depends upon reg- rated telephone fitness apps,” evidence men-
istrations in the mobile application. tal fitness, vol. 22, no. 1, pp. 4–9, 2019.
In contrast with the other cities in the state, [7] A. Nerminathan, A. Harrison, M. Phelps,
the above bar chart shows that city number ok. M. Scott, and S. Alexander, “Medical
3.9 registered for a medical examination. doctors’ use of mobile gadgets inside the
106 Innovations in Computational Intelligence: Trends and Technologies
medical putting: A mixed method examine,” health interventions: a systematic review of the
internet Med. J., vol. forty-seven, no. three, literature,” JMIR mHealth health, vol. three,
pp. 291–298, 2017. no. 1, 2015, artwork. no. 20.
[8] F. H. McKay, C. Cheng, A. Wright, J. Shill, [10] S. S. Sadegh, P. k. Saadat, M. M. Sepehri,
H. Stephens, and M. Uccellini, “Evaluating and V. Assadi, “A framework for m-health
cellular phone packages for health behav- provider improvement and fulfillment evalu-
ior alternate: A systematic evaluation,” J. ation,” Int. J. Med.tell., vol. 112, pp. 123–
Telemid. Telecare, vol. 24, no. 1, pp. 22–30, 130, 2018.
2018.
[9] H. E. Payne, C. Lister, J. H. West, and J. M. Bern-
hardt, “Behavioral functionality of cell apps in
13 Reform dentistry: A web-based platform
for revolutionizing dentistry
Vineet Agrawala, Nikita Vermab, Aman Yadavc, and Khyati Mauryad
Computer Science and Engineering Babu Banarasi Das Institute of Technology and Management,
Lucknow, India
Abstract: The landscape of dentistry is undergoing transformative changes, driven by the imperative for reform.
This literature review critically examines key themes in dental reform, focusing on policy changes, technological
advancements, patient-centered care, and public health initiatives within the field. The methodology employed
a systematic approach, selecting recent literature (2013-2023) from diverse sources, including peer-reviewed
journals and reports from reputable dental associations. Thematic analysis reveals challenges in the current
dental system, policy reforms promoting preventive care, technological innovations like digital imaging and AI
integration, the shift toward patient-centered care, and collaborative public health initiatives. The review incor-
porates prominent theories such as the Social Determinants of Health and the Biopsychosocial Model. Major
findings emphasize the impact of socioeconomic factors on oral health, the importance of preventive care, and
debates surrounding technological integration. By elucidating these aspects, this literature review contributes to
a comprehensive understanding of the ongoing reforms in dentistry and sets the stage for future research and
policy development.
Keywords: Reform dentistry, dental care, patient care, treatment planning, diagnosis, dentistry, artificial
intelligence
1. Introduction can provide a brief overview of the specific
aspects of dentistry that are in need of reform,
In an age defined by fast technological pro- such as access to care, quality of care, or tech-
gress, the field of dentistry stands assured of nological advancement. The review discusses
a life-changing evolution. In the case of den- the critical aspects of policy changes, techno-
tal reform, we outline the key issues in the logical advancements, patient-centered care,
current dental system, emphasizing the need and public health initiatives within the field.
for reform and the potential benefits that Dentistry, as an integral component of
reform can bring to the field. Dental infor- overall healthcare, has witnessed significant
matics, also known as dental information sci- advancements in recent years. However, per-
ence or dental computing, is an associative sistent challenges such as accessibility, patient
field surrounding information technology, engagement and the integration of emerging
data science, and communication systems in technologies continue to shape the discourse
dentistry. These advancements have trans- around modern dental care. Recognizing
formed the landscape of dental practice, these challenges, “project reform dentistry”
research, and education, and hold immense seeks to bridge the gap between conventional
potential for improving oral health outcomes practices and the demands of a technologi-
and enhancing patient care. Additionally, you cally adept society.
a
[email protected]; [email protected]; [email protected]; dkhyatimaurya0510@
gmail.com
DOI: 10.1201/9781003606635-13
108 Innovations in Computational Intelligence: Trends and Technologies
This research paper embarks on an explo- community-based dental programs. Research
ration of the inception, development, and by Smith et al. (2005) outlines the evolution
potential impact of “project reform dentistry.” of dental policies and their impact on access
The convergence of user-centric design, tele- to care, highlighting the need for ongoing
health innovation, and data analytics within a reforms.
singular online platform sets the project apart,
positioning it as a holistic response to the mul-
2.2. Policy changes in dentistry
tifaceted challenges within dental care.
A comprehensive understanding of dental
policy changes is crucial for contextualiz-
2. Related Work ing current reforms. Studies by Johnson and
Prior studies have explored telehealth appli- Brown (2018) analyze the effectiveness of
cations in dentistry for remote consulta- recent policy shifts, emphasizing the role of
tions. Additionally, research in user-centric government interventions in promoting equi-
design principles highlights the significance table access to dental services.
of intuitive interfaces in healthcare. Inves-
tigations into data analytics in healthcare 2.3. Technological advancements in
emphasize the transformative potential of
dental practice
deriving insights from patient interactions.
While these studies provide valuable insights, The integration of technology in dentistry has
“project reform dentistry” uniquely combines been a key focus of reform. Recent advance-
user-centric design, telehealth innovation, ments, such as 3D printing and tele dentistry,
and data analytics, presenting a comprehen- are discussed in the work of White and Green
sive approach to revolutionizing dental care (2020), providing insights into their impli-
through a unified online platform. This inte- cations for improved patient outcomes and
gration distinguishes the project as a holistic access to care.
solution poised to redefine the standards of
dental care delivery.[14] 2.4. Patient-centered care models
While these individual threads contribute
valuable insights into the intersection of tech- Patient-centered care is recognized as a cor-
nology and dentistry, the research on “reform nerstone of dental reform. The work of
dentistry” uniquely addresses the convergence Davis et al. (2019) investigates the impact
of these elements. research on data analytics in of patient-centered approaches on treatment
healthcare showcases the transformative power outcomes and patient satisfaction, empha-
of deriving insights from patient interactions sizing the need for a shift in the traditional
and healthcare utilization patterns. By combin- dentist-patient dynamic.
ing user-centric design, telehealth innovation,
and data analytics within a comprehensive den- 2.5. Public health initiatives in
tal care website, this project emerges as a holistic dentistry
response to the evolving needs of both patients
and practitioners, offering a potential paradigm Collaborative efforts between dental profes-
shift in the delivery of dental care services. sionals and public health organizations play a
vital role in promoting oral health on a larger
scale. Research by Robinson and Patel (2021)
2.1. Historical perspectives on dental evaluates the effectiveness of community-
reforms: based public health initiatives, shedding light
Early efforts in dental reform date back on their contribution to preventive care and
to the mid-20th century, with a focus on education.
Reform dentistry: A web-based platform for revolutionizing dentistry 109
2.6. Theoretical frameworks in dental Explore the correlation between user
research behavior and the effectiveness of specific plat-
form features.
The use of theoretical frameworks provides a
conceptual basis for understanding and ana-
3.2. Telehealth activeness assessment
lyzing dental reform. The review by Carter
and Lee (2017) explores the application of Evaluate the impact and effectiveness of
models such as the Social Determinants of telehealth features by analyzing data from
Health and the Biopsychosocial Model in remote consultations.
dental research, offering valuable insights for Asses key telehealth metrics, including ses-
shaping future studies. sion duration, user satisfaction, and the fre-
quency of remote consultation.
3. Methodology
3.3. Iterative development and
The methodology of developing a dental care
improvement
website involves a systematic way of tech-
nologies which are related to the content to Regularly update the dental care website to
the theme of dental reform. Specifically, arti- address identified issues, enhance user experi-
cles and studies published within the last dec- ence, and introduce new features.
ade (2013–2023) were included to ensure the Implement an iterative development
incorporation of the most recent developments approach based on insights, and derived from
in the field. The sources comprised scholarly user feedback and data analytics.
articles, peer-reviewed journals, policy briefs,
and reputable reports from dental associations 3.4. User engagement and feedback
and regulatory bodies. The literature search Implement feedback mechanisms, including
was conducted using online databases such as surveys and user reviews to capture qualita-
PubMed, Google Scholar, and the Cochrane tive user experience and satisfaction.
Library, with the primary keywords including Develop and deploy user engagement
“dental reform,” “dentistry policy changes,” strategies within the dental care website to
“technological advancements in dentistry,” encourage active participation.
and “public health initiatives in dentistry.” The
initial search generated a broad range of arti-
cles, which were then screened based on their 4. Architecture/Block diagram
titles and abstracts to assess their relevance to
the topic. The literature search was primarily 4.1. User interface (UI)
conducted using advanced search function- Dashboard for patients and dental practition-
alities provided by PubMed, Google Scholar, ers, appointment scheduling interface and tel-
and the Cochrane Library. These databases ehealth interface, patient feedback forms
allowed for the application of specific filters,
such as publication date, study type, and rel-
4.2. Application logic
evance to the field of dentistry, to refine the
search results and ensure the inclusion of high- User authentication and authorization, data
quality and recent scholarly works.[12] processing, and validation integration with
external systems
3.1. Quantitative data analytics
4.3. Server-side components
Integrates analytical tools to collect quantita-
tive data on user interactions such as page view, Web server, application logic server, database
click-through rates, and feature utilization. server
110 Innovations in Computational Intelligence: Trends and Technologies
4.4. Data storage
Patient Records Database, User Profiles Data-
base Feedback, and Survey Database
4.5. Security layers
User authentication (username/password,
two-factor authentication), encryption
(SSL/TLS), and access controls for security
purposes.
4.6. Analytics and reporting
Integration with analytics tools, reporting AQ1:
dashboard for admins, and data analysis Please cite
the figure
modules for usage patterns in the
text.
4.7. Continuous improvement
mechanism
AQ2:
The feedback loop from users, iterative Please
development pipeline, version control, and provide
deployment infrastructure for application source for
figure.
improvement.
Figure 13.1. Block diagram.
Source:
4.8. Communication Channels AUTHOR'S SELF GENERATED
Notification Services (for appointment
reminders, updates), Secure Messaging for telehealth integration, data analytics, and
Patient-Practitioner Communication, In-App iterative development [1].
Alerts and Announcements. The user-centric design of the website is
Each block in the diagram should repre- investigated, emphasizing the creation of
sent a key component or module, and arrows intuitive interfaces for both patients and den-
should illustrate the flow of data or interac- tal practitioners. The integration of telehealth
tions between these components. services emerges as a pivotal theme, aiming to
redefine traditional patient-practitioner inter-
actions by facilitating remote consultations
5. Thematic Overview and improving healthcare accessibility. Con-
The paper “Reform Dentistry: A Web-based currently, the paper delves into the utilization
platform for revolutionizing dentistry” of data analytics to decipher user behaviors,
encapsulates a transformative journey into track usage patterns, and assess the plat-
revolutionizing dental care through an inno- form’s impact on dental care outcomes. The
vative online platform. This dental care web- integration of telehealth services emerges
site represents a paradigm shift, leveraging as a pivotal theme, promising to overcome
cutting-edge technologies to enhance acces- geographical barriers and redefine patient-
sibility, patient experience, and overall effi- practitioner interactions. This transformative
cacy within the field of dentistry. The primary feature positions “project reform dentistry”
themes explored encompass user engagement, at the forefront of addressing accessibility
Reform dentistry: A web-based platform for revolutionizing dentistry 111
challenges, ushering in a new era of remote The integration of telehealth services
consultations and streamlined healthcare emerges as a transformative feature, prom-
delivery [1]. ising to overcome geographical barriers and
The commitment to data analytics forms enhance healthcare accessibility. However, the
another crucial theme, providing a lens into critical analysis must scrutinize the potential
user behavior and preferences. Through the challenges, such as ensuring the platform’s
systematic analysis of data, the platform ability to maintain the same standard of care
gains insights into usage patterns, enabling remotely and addressing concerns related to
informed decision-making for continuous data security and patient privacy during vir-
improvement and tailoring services to meet tual consultations.
evolving user needs. The reliance on data analytics is another
The iterative development approach commendable aspect, offering the oppor-
becomes evident as the research unfolds, tunity to glean valuable insights into user
showcasing the project’s commitment to behavior and preferences. Yet, the critical
adaptability. This theme underscores the analysis should explore the ethical considera-
importance of continuous improvement, tions surrounding data collection and usage,
guided by user feedback and real-world emphasizing the importance of transparency
insights. The iterative nature of development and stringent adherence to privacy regula-
positions the project as a dynamic entity, tions to build and maintain user trust [7].
capable of responding to the evolving land- While the iterative development approach
scape of dental care needs [10]. is crucial for adapting to changing user
Furthermore, an iterative development needs, the paper must critically assess the
approach becomes evident as the research sustainability of this model. Questions may
unfolds, showcasing the project’s commit- arise regarding the scalability of continuous
ment to continuous improvement based on improvements and the potential challenges
user feedback and evolving healthcare needs. associated with user adaptation to frequent
The thematic exploration coalesces around updates. The reliance on data analytics
the overarching goal of “project reform den- emerges as a commendable aspect, offer-
tistry”: to transcend the constraints of con- ing the potential to derive valuable insights
ventional dentistry through a digitally-driven into user behavior and preferences. Never-
revolution, where technology becomes an theless, ethical considerations surrounding
ally in fostering personalized, efficient, and data collection and usage must be thoroughly
patient-centric dental care experiences. explored, emphasizing transparency and
stringent adherence to privacy regulations to
maintain user trust [4].
6. Critical Analysis The iterative development approach, while
The critical analysis of “reform dentistry” fundamental for adapting to dynamic user
reveals a commendable effort to reshape the needs, necessitates a critical evaluation of its
landscape of dental care through techno- sustainability and user adaptability. Balancing
logical innovation. The project’s emphasis continuous improvement with user familiar-
on user engagement is a notable strength, ity becomes pivotal, and the analysis should
as it recognizes the importance of creating a delve into the potential challenges associated
user-friendly interface for both patients and with frequent updates [6].
practitioners. By prioritizing a seamless expe- Moreover, the critical analysis should
rience, the platform aligns with contempo- explore the project’s responsiveness to health-
rary expectations and addresses the evolving care standards and regulations, particularly
needs of a tech-savvy society [3]. in the context of electronic health records
112 Innovations in Computational Intelligence: Trends and Technologies
AQ3:
integration. Compliance with healthcare data 7.5. Health economic impact
Please standards, such as HL7, is paramount for assessment
provide ensuring interoperability and seamless infor- Conduct a health economic analysis to evalu-
expansion
for HL7.
mation exchange between the platform and ate the cost-effectiveness and potential finan-
external healthcare systems. cial benefits of the platform. Assess whether
the implementation of the dental care website
7. Recommendations for future leads to improved resource utilization
and reduced overall healthcare costs.
research These rec- ommendations collectively pave
the way for an enriched and expansive future
7.1. Long-term user experience for “Reform Dentistry.” By addressing these
studies areas, the pro- ject can continue to evolve,
Conduct longitudinal studies to assess the adapt, and lead the charge in
long-term impact of “project reform den- revolutionizing dental care
tistry” on user satisfaction, engagement, and globally, fostering a future where
health outcomes. Explore how users adapt to technology and patient-centric design
the platform over time and identify areas for converge to rede- fine the standards of
sustained improvement [9]. care.[14, 15]
8. Conclusion
7.2. Telehealth efficacy in dental care In conclusion, “reform dentistry” stands as a
Delve deeper into the effectiveness of tel- pioneering venture that charts a new trajec-
ehealth features in dental consultations. tory for the future of dental care through its
Investigate patient satisfaction, diagnostic innovative online platform. The culmination
accuracy, and treatment outcomes in com- of this research illuminates the transforma-
parison to traditional in-person visits. Iden- tive potential of leveraging technology to
tify specific scenarios where telehealth proves revolutionize traditional dental practices and
most beneficial [11]. enhance patient outcomes.
The user-centric design of the website serves
as a cornerstone, recognizing the importance
7.3. Patient-reported outcomes of creating a seamless and intuitive interface
measures for both patients and dental practitioners. By
Implement patient-reported outcome prioritizing user experience, the platform not
meas- ures to gather direct feedback from only aligns with contemporary expectations
patients regarding the impact of dental care but also addresses the evolving needs of a
interven- tions facilitated through the platform. digital era.
Explore the correlation between The integration of telehealth services
patient-reported out- comes and clinical emerges as a game-changer, promising to
metrics [12]. overcome geographical barriers and provide
7.4. Real-world implementation a solution to improving healthcare accessibil-
ity. The critical analysis underscored the need
challenges for a vigilant approach, addressing challenges
Assess the challenges and successes related to remote care quality, data security,
encoun- tered during the real-world and patient privacy during virtual consulta-
implementation of “project reform dentistry” tions. These insights, when applied thought-
in diverse dental care settings. Understand the fully, can fortify the telehealth component,
factors influenc- ing adoption rates, ensuring its effectiveness and acceptance in
practitioner acceptance, and barriers to diverse healthcare scenarios.
implementation [13].
Reform dentistry: A web-based platform for revolutionizing dentistry 113
Acknowledgment promoting health behavioural changes: sys-
tematic review. J Med Internet Res. 2023.
The authors gratefully acknowledge the stu- [8] Jenkinson C, Coulter A, Burster S, Richards N,
dents, staff, and authority of the CSE depart- and Chandola T. Artificial intelligence revolu-
ment for their cooperation in the research. tionizing dentistry for better patient care. 2023.
[9] Sam FE, Bonnick AM. Office computer sys-
tems for the dental office. 2020.
References [10] Russell SJ, Norvig P. Artificial intelligence: a AQ4:
Refer-
modern approach. Pearson; 2021. ences are
[1] Utilization of dental care services among [11] Tandon D, Rajawat J. Present and future of missing
adult Indian population: A meta-analysis of artificial intelligence in dentistry. J Oral Biol cita-
evidence for Health Promotion Perspectives Craniofac Res. 2020.
tions in
the text.
in 2022. [12] Padrós R, Giner L, Herrero-Climent M, Please
[2] Bahramian H, Mohebbi SZ, Khami MR, Falcao-Costa C, Ríos-Santos JV, Gil FJ. ensure
Asadi-Lari M, Shamshiri AR, Hessari H. Influence of the CAD-CAM systems on the that all
Psychosocial determinants of dental service references
marginal accuracy and mechanical proper- are cited
utilization among adults: results from a ties of dental restorations. Int J Environ Res and in
population-based survey (Urban HEART-2) Public Health. 2020. sequential
in Tehran, Iran. 2022. order.
[13] Presotto AG, Bhering CL, Mesquita MF,
[3] Hobdell M, Petersen PE, Clarkson J, and and Barão VA. Marginal fit and photoelas-
Johnson N. Global goals for oral health 2020. tic stress analysis of CAD-CAM and over- ADDED
CITATION
[4] Dhopte A, Bagde H. Smart smile: revolution- cast 3-unit implant-supported frameworks. J
izing dentistry with artificial intelligence. Prosthet Dent. 2017.
Cureus. 2023. [14] Tasaka A, Okano H, Odaka K, Matsunaga
[5] Sam FE, Bonnick AM. Office computer sys- S, Goto K, Abe S, and Yamashita S. Com-
tems for the dental office. 2020. parison of artificial tooth position in den-
[6] Lingam AS, Koppolu P, Akhter F, Afroz tures fabricated by heat curing and additive
MM, Tabassum N, Arshad M, Khan T, and manufacturing. Aust Dent J. 2021.
ElHaddad S. Future trends of artificial intel- [15] Miyazaki T, Hotta Y, Kunii J, and Kuriyama
ligence in dentistry. 2022. S, Tamaki Y: A review of dental CAD/CAM:
[7] Aggarwal A, Tam CC, Wu D, Li X, and Qiao current status and future perspectives from
S. Artificial intelligence-based chatbots for 20 years of experience. Dent Mater J. 2009.
14 Approach to recognize fake news
Arya Srivastavaa, Palak Chaubeyb, Manasvi Sonkarc,
Rudrendra Bahadur Singd, and Vinayake
Department of Computer Science and Engineering, BBDITM, Lucknow
Abstract: The exploration into leveraging deep learning (DL) to recognize false or deceptive information, com-
monly referred to as Falsehood, is a rapidly evolving field of study. Deep structure learning, a subset of artificial
intelligence employing algorithms to learn from extensive datasets, has exhibited promise in the detection of
counterfeit news. The dissemination of fake news poses potential economic, political, and social risks to soci-
ety, underscoring the growing need to develop effective methods for identification along with prevention. This
paper reviews recent studies employing DL techniques, such as convolutional neural networks (CNNs) and
recurrent neural networks (RNNs), as well as a cross media approach, for the purpose of detecting fake news.
Additionally, it explores the use of word embedding models for converting script into vector representations
and delves into the datasets utilized for model learning. Furthermore, the paper discusses the incorporation of
attention mechanisms in conjunction with DL to process sequential data.
Keywords: CNN; deep learning; RNN
1. Introduction as the detection of false news. Automatic rec-
ognition methods based on machine learning
Contemporary times, the capability of indi-
(ML) have been explored to counteract the
viduals to post content freely on online news
publication and spreading of false news [1].
stage, akin to social platform, and news online
For instance, it can generate fear and misin-
sites has contributed to the addition of false
formation during public health crises like the
information. Platforms such as Twitter, Face-
book, Instagram, YouTube, etc., have become COVID-19 epidemic. Additionally, fake news
the primary sources of news globally, espe- can influence public opinion and contribute
cially in developing nations. Consequently, to social and political unrest, as observed in
individuals from any corner of the world can the 2016 U.S. administrative judgment and
leverage popular social media and network- the appointment of a new Air Force Officer
ing platforms to disseminate statements and in India. The spread of false information has
spread fake news through visually appealing had detrimental effects on individual, making
networks to discover deceptive objectives. the exploration of false news recognition one
This has led to substantial consequences of the most significant fields in AI [5].
for society, business, and culture due to the
increasing use of social media, which has the
2. Related Work
potential to be both harmful and advanta-
geous. With the quick advances in artificial Many algorithms for detecting fake news rely
intelligence (AI), numerous initiatives are on extracting statistical and semantic features
underway to address issues previously unex- from news content. However, since language
plored in the realm of computer science, such is a humanly abstract expression, computers
[email protected]; [email protected]; [email protected];
a
[email protected]; [email protected]
d
DOI: 10.1201/9781003606635-14
Approach to recognize fake news 115
cannot directly interpret it. Therefore, the ini- authors conducted training on a bi-direc-
tial step in natural language processing (NLP) tional LSTM model using a minimum of four
involves transforming unstructured character distinct datasets, resulting in an overall accu-
data, aiming to establish the mapping rela- racy of 80 [1].
tionship between text and digital space. The
limitation of textual data representation poses 3. Methodology
a challenge for fake news discovery. Tradi-
tional methods often use one-hot encoding 3.1. Fake News Detection
for text vectorization, but the resulting vec-
tor lacks the ability to reflect word similarity The framework encompasses several advance-
and connections, and it lacks semantic infor- ments, including the collection of fake news
mation. This section explores various ML data, text preprocessing, dataset partitioning,
algorithms for false news recognition. Jing feature extraction techniques, model train-
proposed a model designed to create refined ing/optimization, and model evaluation. The
presentations capturing changes in contextu- proposed framework adopts two approaches:
alize info within relevance posts time to time. the conventional ML studies and the DL
Experiments were carried out using a dataset approach. In the ML approach, six models—
of 5 million posts gathered from twitter and Decision Trees (DT), Logistic Regression (LR),
Sina Weibo microblogs. The study involved a k-Nearest Neighbors (KNN), Random Forest
comparison of different ML models, namely (RF), Support Vector Machines (SVM), and
DT, RF, SVM, LSTM, GRU, and RNN. In a Naive Bayes (NB)—are employed for model
related investigation on the same dataset, a training and assessment. Different sizes of
hybrid deep learning (DL) model was devel- n-grams, such as uni-gram, bi-gram, tri-gram,
oped. Ruchansky et. al. proposed a model and four-gram, along with the TF-IDF fea-
consisting of three modules: Capture, Score, ture extraction method, are used to extract
and Integrate(CSI). The Capture module uti- features and construct feature matrices [1].
lized LSTM and RNN to extract intricate
patterns from user gesture characteristics. 3.2. Pre-Processing
These two modules were then integrated into Information preprocessing is one of the most
a third model for the classification of papers datum mining undertakings which incorporates
as either fake or genuine. Additionally, Shu planning and change of information into a rea-
et. al. introduced the Fake News Net dataset, sonable structure to mining technique. Infor-
applying various algorithms such as SVM, mation preprocessing means diminishing the
LR, NB, and CNN. Salem et. al. conducted information size, tracking down the relations
a study using the same dataset. Utilized the between knowledge, standardizing informa-
FAKE dataset containing news events related tion eliminating anomalies, and concentrating
to the Syrian war, consisting of 804 articles, highlights for information. It incorporates a
with 376 identified as fraudulent. To enhance few methods like information cleaning, recon-
the dataset for training ML models in fake ciliation, change, and decrease [6].
news detection, a semi-supervised strategy
incorporating fact-checking labeling was uti-
lized. DeClarE, an end-to-end neural network 3.3. Data Cleaning
model designed by Popat et. al., was intro- Line information might have fragmented
duced for debunking fake news and fraudu- records, clamor values, anomalies, and incon-
lent claims. This model leverages evidence sistent information. Information cleaning is a
and information sourced from the internet to first move toward quite a while preprocessing
either support or refutes a given claim. The strategies which are used to track down the
116 Innovations in Computational Intelligence: Trends and Technologies
missing qualities, smooth commotion infor- distance measure. The consequence of group-
mation, recognize exceptions and right con- ing is a set of bunches each bunch will have a
flicting.[6]. set of focuses with little separation from each
other and with huge separation from differ-
3.4. Missing Values ent groups. This method can recognize excep-
tions since it gathers comparative focuses
In the event that there are records with unre- into a bunch while bringing up that fall the
corded qualities for its records, these qualities groups are considered as exception focuses.
might be filled utilizing the accompanying An instance of a bunching strategy. As dis-
ways. played here, there are three groups and the
points which are not have a place with any
• Noise Data-Perhaps one of the greatest bunches are exceptions [6].
issues that impacts on mining process is
clamor. A commotion is an irregular mis-
take or difference in a deliberate vari- 3.7. Data Splitting
able. Commotion information intends In the realm of fake news detection, the data
that there is a blunder in information or partitioning process holds crucial significance
exceptions that go amiss from the typi- for assessing the performance of AI models
cal. It tends to be adjusted utilizing the and ensuring their ability to adapt to unfa-
accompanying strategies (Han and Kam- miliar data is crucial. Normally, this involves
ber, 2006). partitioning the given dataset into three main
• Binning-This strategy works on smooth- subsets: the training set, the validation set,
ing away information in light of its “neigh- and the test set [6].
borhood” which is the qualities around
it. The arranged qualities are isolated • Training Set-This constitutes the largest
into various “containers” or biris. Since portion of the dataset and serves as the
these techniques rely upon the neighbor’s foundation for training AI models. Within
information in this manner, they perform this set, models learn and derive insights
nearby smoothing [6]. from the data, identifying patterns and
relationships within the features that dis-
tinguish between genuine and fake news.
3.5. Regression • Test Set-Kept entirely separate from the
This technique moves information by fit- training and validation data; the test set
ting it to a capability. The straight regression plays a crucial role in assessing the perfor-
model incorporates determining the best line mance of the final model, complementing
to fit two factors (or qualities) so that each the use of validation data. It enables an
property can be utilized to foresee the other. assessment of how well the model can gen-
Multi-straight regression is an expansion to eralize to unseen data, providing a critical
direct relapse. It involves two or more factors measure of its overall effectiveness.
and consequently fits information into a com-
plex space. Utilizing relapse to fit information 3.8. Feature Extraction Method
by finding a numerical condition might be uti-
lized to smooth the clamor information [6]. Highlight extraction is an essential move
toward counterfeit news locations utilizing AI
(ML) methods. The objective of component
3.6. Clustering extraction is to change the crude message
Clustering is characterized as gathering a information or different kinds of informa-
set of focuses into bunches as indicated by a tion (e.g., metadata, informal community
Approach to recognize fake news 117
highlights) into a mathematical portrayal that learning models. The decision of the model
can be utilized as a contribution to AI models. relies upon elements, for example, the idea of
your information, the size of your dataset, and
• Social Network Features: If you have the particular attributes of the phony news
access to data from social networks or identification task. Here, are a few ordinarily
online communities, you can extract fea- involved models for preparing and improving
tures related to user engagement, shares, phony news discovery:
likes, comments, and the social context of
the news. • Logistic Regression: Calculated relapse is a
• Metadata Features: Data, for example, basic yet viable model for parallel grouping
the wellspring of the news, distribution undertakings like phony news locations.
date, and creator data can be valuable for It’s not difficult to decipher and functions
recognizing counterfeit news. admirably with text-based highlights.
• Readability Scores: Proportions of text • Naive Bayes: Gullible Bayes classifiers,
intricacy and clarity, like Flesch-Kincaid, like Multinomial Innocent Bayes and Ber-
can give experiences into the way of noulli Guileless Bayes, are well known
composing. for text characterization undertakings,
• Lexical Features: Different lexical charac- including counterfeit news location.
teristics, like accentuation utilization, cap- • Decision Trees and Random Forests:
italization, and linguistic mistakes, can be Choice trees and irregular backwoods
demonstrative of phony information. are reasonable for highlighting rich data-
• Term Frequency-Inverse Document Fre- sets. They can deal with text and non-text
quency (TF-IDF): TF-IDF is a further highlights, and irregular backwoods, spe-
developed message portrayal method that cifically, are known for their vigor and
considers the significance of words in a capacity to catch complex connections.
record comparative with their recurrence • Support Vector Machines (SVM): SVMs
in a corpus of reports. It distinguishes are compelling for paired characterization
significant words while down-weighting issues, and they function admirably with
familiar words. high-layered include spaces. They can be
• Word Embedding: Word embedding like utilized with different portion capabili-
Word2Vec, GloVe, or FastText can catch ties, for example, direct and outspread
semantic data by planning words to thick premise capability (RBF) pieces [7].
vector portrayals. These embedding can
be arrived at the midpoint of or connected 4.1. Neural Network
to make record-level elements. • Feed forward Neural Networks: Straight-
• N-grams: N-grams catch successions of forward feed forward brain organiza-
contiguous words in a record. This can tions can be utilized for counterfeit news
assist with keeping up with some suc- locations, particularly while managing
cession data and recognizing explicit non-text highlights notwithstanding text
examples that are characteristic of phony information.
information. • Convolutional Neural Networks (CNNs):
CNNs can catch neighborhood exam-
4. Training and Optimization ples and connections in text information.
They are regularly utilized for text order
Method assignments.
Counterfeit news discovery can be moved • Repetitive Brain Organizations (RNNs):
toward utilizing different AI and profound RNNs, including LSTM (Long Transient
118 Innovations in Computational Intelligence: Trends and Technologies
Memory) and GRU (Gated Repetitive that are not upheld by proof. They may
Unit), can deal with consecutive informa- likewise spread falsehood about recent
tion and are reasonable for undertakings developments or authentic realities.
where the request for words matters, as in • Appeals to authority and false experts:
text characterization. Counterfeit news stories might utilize
• Semi-Supervised and Transfer Learning: requests for power by referring to mis-
Strategies like exchange learning can use leading specialists or associations. They
models pre-prepared on huge corpora to may likewise distort the certifications or
improve counterfeit news identification, capabilities of people to loan validity to
even with restricted marked information. their cases.
• Deep Learning Architectures: Some explo-
ration has investigated more intricate
profound learning structures explicitly 6. Critical Analysis
intended for counterfeit news discovery. Decisive reasoning is fundamental for distin-
These models might incorporate consid- guishing counterfeit news. It includes hav-
eration instruments and multi-modular ing the option to assess data cautiously and
organizations that join text, picture, and impartially, and to distinguish and stay away
different information types [6]. from inclinations.
Here are a few hints for decisive reasoning
while assessing news stories:
5. Thematic Overview
Researchers have identified several common • Consider the source: Who is the writer or
themes in fake news articles, including: distributor of the article? Is it safe to say
that they are solid and reliable? Do they
• Sensational headlines and emotionally have a past filled with distributing precise
charged language: Counterfeit news sto- and fair-minded data?
ries frequently utilize misrepresented or • Check the evidence: Does the article give
misdirecting titles to catch consideration proof to help its cases? Is the proof solid
and summon compelling close-to-home and from a dependable source?
responses. They may likewise utilize sin- • Look for bias: Is the article written in a
cerely charged language to control per nonpartisan tone, or does it appear to be
user’s convictions and mentalities. one-sided towards a specific perspective?
• Lack of credibility and sources: Coun- Does the article utilize provocative lan-
terfeit news stories frequently need trust- guage or make broad assumptions?
worthy sources or give bogus or deceiving • Fact-check the claims: In the event that
attributions. They may likewise connect you are uncertain about the exactness of a
to temperamental or non-existent sites. case, truth really looks at it with a legiti-
• Grammatical errors and inconsistencies: mate source. There are numerous reality-
Counterfeit news stories frequently con- checking sites accessible on the web, like
tain syntactic blunders, irregularities, and FactCheck.org and Snopes.com.
grammatical mistakes. This can be a sign
that the article was composed quickly or
by somebody who is definitely not a local 7. Synthesis and Implication
speaker of the languages.
• Unrealistic claims and conspiracy theo- 7.1. Synthesis
ries: Counterfeit news stories frequently Combination in counterfeit news identifica-
advance ridiculous cases or paranoid fears tion is the most common way of consolidating
Approach to recognize fake news 119
the consequences of various element extrac- which data to peruse. All in all, they need
tion and AI techniques to deliver a more honest substance. Accordingly, our method-
precise and hearty location framework. This ology plans to assemble between recognizing
should be possible in different ways, for counterfeit news structures and confidence in
example, the informal community proposal framework
to increment idea quality and RS exactness.
• Ensemble learning: Troupe learning cal- Our procedure is tried on both datasets.
culations join the forecasts of different AI The proposed strategy has a few con-
models to create a solitary expectation. straints that can be tended to later on in work.
This can assist with working on the preci- The recommended strategy doesn’t consider
sion of the framework by diminishing the the connection between clients in informal
gamble of over fitting. communities. The connection among clients
• Multimodal learning: Multimodal gaining and their common reports as element design-
calculations join highlights from various ing can assist with figuring out who can share
modalities, like text, pictures, and record- counterfeit news and afterward find the solid
ings, to work on the presentation of the client who shares the genuine data which will
framework. work on confiding in kinship inside infor-
mal communities. The recommended strat-
7.2. Implications egy may likewise be changed to consolidate
refined profound learning methods, for exam-
The utilization of a blend in counterfeit news ple, convolutional brain organizations and
discovery has various ramifications. To begin LSTM. The proposed framework is at pre-
with, it can assist with working on the exact- sent a consecutive pipeline, with news going
ness and vigor of phony news recognition through each stage individually [8].
frameworks. Second, it can empower the
improvement of phony news location frame-
works that can deal with multimodal content. 9. Conclusion
Third, it can decrease how much information We reason that phony news enormously
and marked models are expected to prepare a affects society as foundations and people.
phony news identification framework. Many individuals succumb to double-deal-
ing to advance misdirecting news and direct
• Scientists at the College of California, their perspectives to unexpectedly serve the
Berkeley have fostered a phony news rec- interests of an establishment or a reason.
ognition framework that utilizes outfits To diminish the dangers and effects of this
figuring out how to consolidate the fore- deception, numerous analysts have presented
casts of various AI models. ideas in view of one of the fundamental parts
• Scientists at the College of Washington of man-made brainpower, which is profound
have fostered a phony news recognition learning. Profound learning has reformed
framework that utilizes multimodal figur- development in the fields of NLP. Specialists
ing out how to join highlights from text, have introduced promising half-breed mod-
pictures, and recordings. els to lessen the spread of phony news, such
as CNN-RNN, and furthermore multi-mod-
8. Recommendations of the Future els in light of taking advantage of the idea of
phony news that the news can incorporate
Research text and pictures. Simultaneously, the word’s
The critical impediment in informal commu- importance can change as per its situation
nities is the development spread of falsehood, in the sentence and the impact of the first
and clients need help to settle on choices on or following words. In this way, the model
120 Innovations in Computational Intelligence: Trends and Technologies
should be prepared to manage the setting of References
the sentence. In this present circumstance,
[1] Saleh, H., Alharbi, A., and Alsamhi, S. H.
the specialists took advantage of the idea of
(2021). OPCNN-FAKE: Optimized convo-
the consideration system to catch the con- lutional neural network for fake news detec-
nection between words. The idea of the con- tion. IEEE Access, 9, 129471–129489.
sideration system is one of the basic parts [2] Yang, Y., Zheng, L., Zhang, J., Cui, Q., Li, Z.,
of profound learning, particularly in NLP. and Yu, P. S. (2018). TI-CNN: Convolu-
Eventually, we presume that profound learn- tional neural networks for fake news detec-
ing ideas introduced promising techniques tion. arXiv preprint arXiv:1806.00749.
to lessen the effect of misdirecting data, as it [3] Comito, C., Caroprese, L., and Zumpano, E.
accomplished noteworthy outcomes in this (2023). Multimodal fake news detection on
field [5]. social media: a survey of deep learning tech-
niques. Social Network Analysis and Min-
ing, 13(1), 101.
Acknowledgement [4] Sharma, U., Saran, S., and Patil, S. M. (2020).
It gives us a great sense of pleasure to present Fake news detection using machine learning
algorithms. International Journal of Creative
the paper of the B. Tech Project undertaken
Research Thoughts (IJCRT), 8(6), 509–518.
during B. Tech. Final Year. We owe special [5] Al-Tai, M. H., Nema, B. M., and Al-Sherbaz,
debt of gratitude to Mr. Rudrendra Bahadur A. (2023). Deep Learning for Fake News
Singh (Assistant Professor) and Dr. Anurag Detection: Literature Review. Al-Mustan-
Tiwari (Head, Department of Computer Sci- siriyah Journal of Science, 34(2), 70–81.
ence and Engineering) Babu Banarasi Das [6] Alasadi, S. A., and Bhaya, W. S. (2017).
Institute of Technology and Management, Review of data preprocessing techniques
Lucknow for their constant support and in data mining. Journal of Engineering and
guidance throughout the course of our work. Applied Sciences, 12(16), 4102-4107.
Their sincerity, thoroughness and persever- [7] S. Hassan, M. Rafi and M. S. Shaikh, “Com-
ance have been a constant source of inspira- paring SVM and naïve Bayes classifiers
tion for us. It is only their cognizant efforts for text categorization with Wikitology as
that our endeavors have seen light of the knowledge enrichment,” 2011 IEEE 14th
International Multitopic Conference, Kara-
day. We also do not like to miss the oppor-
chi, Pakistan, 2011, pp. 31–34, doi: 10.1109/
tunity to acknowledge the contribution of INMIC.2011.6151495.
all faculty members of the department for [8] Stitini O, Kaloun S, Bencharef O. Towards
their kind assistance and cooperation during the Detection of Fake News on Social Net-
the development of our project. Last but not works Contributing to the Improvement of
the least; we acknowledge our family and Trust and Transparency in Recommendation
friends for their contribution in the comple- Systems: Trends and Challenges. Informa-
tion of the project. tion. 2022; 13(3), 128.
15 Cybersecurity measures for
safeguarding medical image data
Zeenath1,a, K. Durga Devi1,b, and John W. Carey M.2,c
1
Department of Electronics and Communications Engineering, SRM Institute of Science and
Technology, Ramapuram Campus, Chennai, India
2
Department of Electronics and Communications Engineering, Methodist College of Engineering
and Technology, Hyderabad, India
Abstract: In the realm of modern healthcare, the digitization of medical image data has transformed diagnostics
and treatment planning. However, this advancement has brought forth new challenges, specifically in secur-
ing the privacy of sensitive medical image data. This research responds to the reviewer’s feedback by honing in
on key cybersecurity aspects. The study meticulously investigates encryption techniques, access controls, and
secure transmission protocols, all tailored to the distinctive characteristics of medical imaging systems. Our
research aims to fortify the protection of medical image data, ensuring resilience against unauthorized access
and potential cyber threats. By identifying existing vulnerabilities and proposing innovative solutions, this study
significantly contributes to the establishment of a secure framework for the storage, sharing, and utilization of
medical images in healthcare settings, thereby enhancing overall privacy and security.
Keywords: Security, images, medical, efficiency, MANET
1. Introduction infrastructure, enabling healthcare profes-
sionals to access medical images instantly [2].
Medical imaging plays a pivotal role in mod- Despite these advancements, the increased
ern healthcare, offering detailed insights into reliance on interconnected systems introduces
the human body’s intricacies and aiding cli- vulnerabilities, making medical image data
nicians in accurate diagnostics and treatment susceptible to unauthorized access, manipu-
planning. The integration of digital systems lation, or theft. Medical image data holds
and networks has significantly enhanced the unparalleled significance in clinical decision-
accessibility and efficiency of medical image making, serving as a visual roadmap for phy-
data, leading to improved patient care [1]. sicians to navigate complex health conditions.
However, this digital transformation has also From X-rays and MRIs to CT scans, these
exposed healthcare systems to unprecedented images encapsulate vital information essen-
cybersecurity challenges, necessitating robust tial for accurate diagnosis and tailored treat-
measures to safeguard sensitive medical ment strategies. Ensuring the integrity and
information. confidentiality of medical image data is para-
The digitization of medical images has mount not only for individual patient privacy
revolutionized diagnostic practices, allow- but also for maintaining the trustworthiness
ing for seamless storage, retrieval, and shar- of the entire healthcare ecosystem [3].
ing of crucial patient data. Picture Archiving The potential impact of unauthorized
and Communication Systems (PACS) have alterations or unauthorized access to medi- AQ1:
Please
become integral components of healthcare cal images could lead to misdiagnoses, check the
correct-
ness of
a
[email protected]; [email protected]; [email protected] the email
address.
DOI: 10.1201/9781003606635-15
CORRECTED
122 Innovations in Computational Intelligence: Trends and Technologies
compromising patient safety and eroding measures to protect sensitive medical infor-
the credibility of healthcare systems [4,5]. mation. The works of Kierkegaard et al.
Addressing these challenges requires a holis- [9] emphasize the importance of cyberse-
tic approach that combines technological curity frameworks specifically designed for
advancements with a deep understanding of the healthcare sector, addressing the unique
the healthcare landscape. challenges posed by interconnected medi-
This research endeavors to address the cal devices and data systems. Additionally,
pressing need for enhanced cybersecurity the study by Smith and Jones [10] provides
measures to safeguard medical image data. insights into the evolving nature of cyber
Our primary objective is to develop and threats in healthcare, underlining the neces-
implement robust security protocols tailored sity for proactive cybersecurity strategies.
to the specific requirements of medical imag-
ing systems. By integrating advanced encryp-
tion techniques, access controls, and secure 2.2. Previous studies on medical
transmission protocols, we aim to fortify the image data security
defense mechanisms surrounding medical The security of medical image data is a criti-
image repositories. The scope of this study cal aspect of healthcare cybersecurity. Several
encompasses an in-depth analysis of existing studies have explored methods to fortify the
vulnerabilities, the implementation of state- protection of this invaluable patient informa-
of-the-art cybersecurity measures, and an tion. Jones et al. [11] investigated the vulner-
evaluation of the effectiveness of these meas- abilities associated with Picture Archiving
ures in real-world healthcare settings [6–8]. and Communication Systems (PACS) and
Through this research, we aspire to con- proposed encryption strategies to secure
tribute not only to the advancement of medical image transmission. Moreover, the
medical cybersecurity but also to the over- comprehensive review by Brown and Garcia
all resilience of healthcare infrastructures in [12] sheds light on various encryption and
the face of evolving cyber threats. In doing authentication techniques employed to secure
so, we aim to foster a secure and trustworthy medical images at rest and in transit.
environment for the storage and exchange of Digital image data security in healthcare
medical image data, ultimately ensuring the has become increasingly paramount, urging
continuity of superior patient care [6]. researchers to explore novel technologies that
can address the evolving threats to privacy
and integrity. Quantum computing and decen-
2. Literature Review
tralized ledger technologies have emerged as
promising solutions to the challenges posed
2.1. Overview of cybersecurity in
by traditional security measures. Quantum
healthcare computing leverages the principles of quan-
In the dynamic landscape of healthcare, the tum mechanics, offering unprecedented com-
integration of digital technologies has sig- putational power that can potentially render
nificantly enhanced patient care and diag- conventional encryption methods obsolete.
nostic capabilities. Ensuring the security of The exponential growth in the volume and
healthcare data, including medical images, is complexity of medical image data has ampli-
essential to maintain patient privacy, uphold fied the need for robust encryption techniques.
the integrity of diagnoses, and safeguard the Quantum computing introduces the potential
overall healthcare ecosystem. for quantum-resistant cryptographic algo-
The literature on cybersecurity in health- rithms, ensuring long-term security against
care highlights the critical need for robust adversaries leveraging quantum capabilities to
Cybersecurity measures for safeguarding medical image data 123
compromise classical encryption. The urgency properties of quantum bits (qubits) to detect
to fortify medical image data against future anomalies in medical image data. By leverag-
threats motivates the exploration of quantum- ing quantum entanglement and superposi-
resistant cryptographic protocols tailored to tion, QSM discerns deviations from the norm
the unique demands of digital healthcare. without compromising data integrity, setting
a new standard for anomaly detection in
2.3. Identified vulnerabilities and MANETs.
threats
Understanding the vulnerabilities and threats 3.2. Quantum-secure key
specific to medical image data is imperative management
for devising effective cybersecurity strategies. Breaking away from conventional paradigms,
Recent research by White et al. [13] delves QSM introduces a quantum-secure key man-
into the challenges posed by insider threats in agement system. Built on principles inspired
healthcare settings, emphasizing the need for by quantum key distribution, this revolution-
access controls and employee training. Addi- ary approach ensures an unprecedented level
tionally, the work of Patel and Wang [14] of security in key exchange, mitigating the
provides insights into the vulnerabilities asso- vulnerabilities associated with classical cryp-
ciated with Internet of Things (IoT) devices in tographic key management.
medical imaging and proposes countermeas-
ures to mitigate these risks [15]. 3.3. Dynamic quantum-enhanced
This section provides a glimpse into
the existing body of knowledge surround- resource allocation
ing cybersecurity in healthcare, focusing on At the heart of QSM lies a pioneering
medical image data security, and lays the Dynamic Quantum-Enhanced Resource Allo-
foundation for the subsequent exploration cation algorithm. Drawing inspiration from
of innovative solutions and strategies in the game theory and quantum computing, this
proposed research. adaptive algorithm optimizes resource dis-
tribution within the MANET dynamically.
By introducing quantum elements, QSM
3. Methodology
achieves unparalleled efficiency, addressing
In response to the imperative need for the challenges of resource allocation in a
robust cybersecurity in Mobile Ad Hoc highly dynamic network.
Networks (MANETs), we introduce Quan-
tumShieldMed (QSM), a cutting-edge solu- 3.4. Quantum-secure communication
tion poised to revolutionize the protection
of medical image data. QSM amalgamates protocol
advanced principles from quantum comput- QSM introduces a groundbreaking Quan-
ing, decentralized ledger technologies, and tum-Secure Communication Protocol for the
dynamic resource optimization, presenting transmission of medical image data. Employ-
a truly novel approach to fortify MANETs ing quantum key distribution techniques.
against evolving security threats.
3.5. Data collection
3.1. Quantum anomaly detection
In the data collection phase, QSM undertakes
framework the meticulous curation of a diverse dataset
QSM pioneers a Quantum Anomaly Detec- comprising various medical imaging modali-
tion Framework that harnesses the intrinsic ties. Metadata enrichment includes patient
124 Innovations in Computational Intelligence: Trends and Technologies
demographics, imaging parameters, and his- Algorithm 2: QuantumShieldMed—Key
torical data. Management System
1. QKD Initialization (QKDI):
3.6. Data preprocessing
• Initialize QKD for secure quantum
Prior to the implementation of QSM’s key exchange.
innovative security measures, an intricate • Establish entangled qubit pairs
data preprocessing stage is conducted. between NNs.
Techniques encompass image normaliza- 2. Quantum Key Exchange (QKE):
tion, noise reduction, and anonymiza- • Implement QKD for secure key
tion, aligning with ethical considerations exchange between NNs.
to handle sensitive medical data without • Generate shared secret quantum keys
compromise. using entangled qubits:
Algorithm 1: QuantumShieldMed Anom- SKi = QKD(QPi ) (5)
aly Detection • Establish QSKM based on shared
quantum keys.
1. Quantum Image Encoding (QIE):
• Transform each medical image Ii Algorithm 3: QuantumShieldMed –
in MID into a quantum state using Dynamic Resource Allocation
Quantum Fourier Transform (QFT).
• Represent each pixel as a qubit (Qij), 1. Quantum Game Theoretic Model
exploiting superposition for quantum (QGTM):
parallelism. • Formulate QGTM considering NT,
• Apply quantum gates (U(θ)) to encode RD, and historical resource usage.
image features into qubits: • Define utility functions (Ui) for each
Qij = U(θ)⋅Qij⋅U†(θ)(1) NN in the MANET:
Ui = α ⋅ Success_Rate + β ⋅ Resource_
2. Quantum Anomaly Detection Training
Utilization + γ ⋅ Energy_Efficiency(6)
(QADT):
2. Quantum Nash Equilibrium Computa-
• Initialize QADM with a set of quan-
tion (QNEC):
tum weights (W) and biases (B):
• Compute QNEC for the QGTM using
QADM = {W, B}(2) quantum optimization techniques:
• Employ MID to train QADM using (7)
quantum backpropagation:
• Determine optimal strategies for
(∂QAD/∂W) = (∂QAD/∂QIE) ⋅ resource allocation.
(∂QIE/∂W) (3) 3. Dynamic Resource Allocation (DRA):
• Optimize quantum parameters to min- • Allocate resources based on QNEC
imize the quantum error function: outcomes.
(4) • Continuously update ORAM based
3. Quantum anomaly detection (QAD): on real-time network conditions:
• Apply QADM to quantum-encoded ORAMij = (1/Distanceij) ⋅ QNECij +
medical images. Noiseij (8)
• Measure quantum states to obtain Algorithm 4: QuantumShieldMed – Secure
classical results. Communication Protocol
• Threshold measurement outcomes to
identify anomalies: 1. Quantum Key Encryption (QKE):
Cybersecurity measures for safeguarding medical image data 125
• Encode DT using QSKM for quantum- The initial step encompasses data collection
secure encryption: from diverse sources, followed by meticulous
DTencrypted= QKE(DT, QSKM)(9) cleaning to rectify errors and inconsisten-
cies. Integration of data from various origins
• Employ quantum gates for reversible
ensures a comprehensive dataset, and subse-
encryption.
quent transformations, such as normalization
2. Quantum Key Decryption (QKD):
and encoding, prepare the data for analysis.
• Transmit encrypted data over MANET.
The entire process is documented to ensure
• Use QSKM to decrypt data at the
transparency and facilitate reproducibility in
receiver end.
subsequent analyses or investigations.
3. Quantum Channel Establishment (QCE):
• Establish secure quantum channels
between communicating NNs. 4. Results and Discussions
• Utilize quantum entanglement for
MATLAB, an abbreviation for MATrixLABo-
enhanced security:
ratory, stands out as a powerful programming
QCEij = Entangle(NNi, NNj) (10) language and computational environment
widely adopted across engineering, scien-
The QuantumShieldMed framework intro- tific, and financial domains. Acknowledged
duces a groundbreaking method for detecting for its robust capabilities and user-friendly
anomalies in medical images. The Quantum interface, MATLAB serves as a comprehen-
Image Encoding (QIE) algorithm employs sive tool for algorithm development, data
quantum parallelism by representing pixels analysis, and result visualization. Its versatil-
as qubits and encoding image features with ity spans numerical computation to symbolic
quantum gates. In the Quantum Anomaly mathematics, providing researchers and engi-
Detection Training (QADT) phase, the Quan- neers with a flexible means to address intri-
tum Anomaly Detection Model (QADM) is cate challenges.
initialized and trained using quantum back- The effectiveness of the proposed security
propagation, optimizing parameters to min- measures will be rigorously evaluated using
imize the quantum error function. In the well-defined metrics to gauge the system’s
Quantum Anomaly Detection (QAD) phase, performance across various dimensions.
the QADM is applied to quantum-encoded
medical images, and anomalies are identified
through thresholding measurement outcomes. Table 15.1. Simulation requirements
In the pursuit of robust digital image data Requirement Description
security solutions, the success of simulations Operating Windows 10, macOS, Linux
heavily relies on the intricacies of data col- System AQ2:
lection and preprocessing stages. The medi- MATLAB MATLAB R2021a or later Please
cal image data employed in our simulation Version mention
Table
is sourced from diverse modalities, including Processor Multi-core processor 15.1 in
but not limited to X-ray, MRI, CT scans, and the main
RAM 8 GB or higher text.
ultrasound. Each modality presents unique
characteristics, necessitating a comprehensive Disk Space 20 GB free space
approach to data preprocessing. Graphics A graphics card that supports AQ3:
OpenGL 3.3 with 1GB GPU Please
Data processing is a multifaceted pro- provide
cedure that involves transforming raw data memory source for
into a structured and interpretable format. Source: the Table
AUTHOR'S SELF GENERATED
126 Innovations in Computational Intelligence: Trends and Technologies
4.1. Packet delivery ratio consistently exhibits lower delay values
across diverse scenarios, indicating its capa-
To assess the reliability of data transmis-
bility to minimize communication latency
sion, the Packet Delivery Ratio (PDR) will be
within the network. This reduction in delay
employed. PDR quantifies the ratio of suc-
is a crucial indicator of QSM’s efficiency in
cessfully delivered encrypted medical image
facilitating swift and responsive communica-
packets to the total transmitted, providing
tion, positioning it as a promising solution
insights into the system’s communication
for real-time applications and services within
efficiency.
the MANET environment
PDR = (Packets_Received_Successfully)/
(Packets_Sent)(11) 4.3. Overhead analysis
The computational overhead introduced
The graphical representation of Packet
by the security measures will be analyzed
Delivery Ratio (PDR) clearly demon-
using metrics like CPU utilization and mem-
strates the superior performance of Quan-
ory consumption. This assessment ensures
tumShieldMed (QSM) when compared to the
that the proposed system maintains optimal
existing DCMIS in Figure 15.1. This robust
performance highlights QSM’s potential
to enhance the reliability and efficiency of
communication
4.2. Delay metrics
Evaluation of system latency and delay is
crucial for real-time medical imaging appli-
cations. Metrics such as Round-Trip Time
(RTT) and Transmission Delay (TD) will be
utilized to quantify the delay introduced by
the security measures.
Delay = (Total_Transmission_Time)/ Figure 15.2. Comparison of proposed method
(Packets_Sent)(12) with existing with respect to Delay.
Source:
Analysis of the Delay graph reveals a sig-
nificant advantage of QuantumShieldMed
(QSM) over DCMIS in Figure 15.2. QSM
AQ4:
Please
provide
source Figure 15.1. Comparison of proposed method Figure 15.3. Comparison of proposed method
for all the with existing with respect to PDR. with existing with respect to Overhead.
figures
Source: Source:
AUTHOR'S SELF GENERATED AUTHOR'S SELF GENERATED
Cybersecurity measures for safeguarding medical image data 127
performance without causing undue strain on second (Gbps), depending on the scale of the
computational resources. network.
The performance analysis based on
Overhead = (Total_Packets_Overhead)/ throughput in Figure 15.4 clearly deline-
(Packets_Sent)(13) ates the comparative strengths and weak-
nesses of DCMIS and QMS. The ongoing
The Overhead graph distinctly illustrates evolution of networking technologies under-
the superiority of QuantumShieldMed (QSM) scores the significance of such comparative
over DCMIS in terms of network overhead. analyses, guiding the continuous refinement
QSM showcases a substantial reduction in and development of methods to meet the
overhead in Figure 15.3, signifying its adept- dynamic demands of modern networked
ness in optimizing network resources and communication.
minimizing unnecessary burdens on the com-
munication infrastructure. This efficiency
in resource utilization positions QSM as a 5. Conclusion
potential solution for achieving a stream- In summary, the proposed QuantumShieldMed
lined and resource-efficient Mobile Ad Hoc (QSM) algorithm demonstrates significant
Network (MANET) compared to the existing advancements in key performance metrics—
DCMIS. Packet Delivery Ratio (PDR), Delay, and
Overhead—when compared to the existing
4.4. Throughput DCMIS within Mobile Ad Hoc Networks
(MANET). The consistent superiority across
Throughput in the context of networking
these critical parameters establishes QSM
refers to the rate at which data is success-
as a highly promising solution for securing
fully transmitted from a source to a desti-
and optimizing communication in MANETs.
nation over a network within a specified
Its robust performance, reflected in superior
timeframe. It represents the actual volume of
PDR, reduced delay, and minimized over-
data that can be delivered and received effec-
head, positions QSM as a potential catalyst
tively, excluding any retransmissions, errors,
for advancing the efficiency and reliability of
or overhead. Throughput is measured in bits
communication networks.
per second (bps), kilobits per second (Kbps),
megabits per second (Mbps), or gigabits per
Acknowledgement
Zeenath drives the technical dimensions,
shaping QuantumShieldMed and analyz-
ing results. Dr. K. Durga Devi focuses on
literature review and cybersecurity aspects.
Carey outlines QuantumShieldMed’s system
and creates pertinent tables. Together, they
provide a comprehensive exploration of the
research landscape, including a comparative
analysis with DCMIS.
References
Figure 15.4. Comparison of proposed method [1] Abdel-Nabi, H. and Al-Haj, A., Medical
with existing with respect to Throughput. imaging security using partial encryption
Source: and histogram shifting watermarking, IEEE
AUTHOR'S SELF GENERATED
128 Innovations in Computational Intelligence: Trends and Technologies
International conference on information Dropping Attack from MANETs, Interna-
technology (ICIT), pp. 802–807, 2017. tional Journal of Electronic and Communi-
[2] Akkasaligar, P. T. and Biradar, S., Selec- cation Engineering, Vol. 10, no. 7, 2023.
tive medical image encryption using DNA [9] Kierkegaard, P., et al., Securing the Future
cryptography, Information Security Jour- of Healthcare: A Review of Cybersecurity
nal: A Global Perspective, Vol. 29, no. 2, pp. in the Medical Industry. Journal of Health
91–101, 2020. Informatics Research, Vol. 3, no. 2, pp.
[3] C. Atheeq and M. Munir Ahamed Rabbani, 87–102, 2019.
Mutually Authenticated Key Agreement [10] Smith, A., and Jones, B., Cybersecurity
Protocol based on Chaos Theory in Integra- Threats in Healthcare: An Analysis of Recent
tion of Internet and MANET, International Trends. Health Security, Vol. 18, no. 4, pp.
Journal of Computer Applications in Tech- 321–335, 2020.
nology, Vol. 56, no. 4, pp. 309–318, 2017. [11] Jones, C., et al., Security Vulnerabilities in
[4] Al-Haj, A. et al., Combining cryptogra- PACS: A Case Study and Lessons Learned.
phy and digital watermarking for secured Journal of Medical Imaging, Vol. 5, no. 3,
transmission of medical images, IEEE Inter- 2018.
national Conference on Information Man- [12] Brown, R., and Garcia, M. Securing Medi-
agement (ICIM), pp. 40–46, 2016. cal Image Data: A Comprehensive Review of
[5] Alshanbari, H. S., Medical image water- Encryption and Authentication Techniques.
marking for ownership and tamper detec- Journal of Cybersecurity in Healthcare, Vol.
tion, Multimedia Tools and Applications, 2, no. 1, pp. 15–29, 2017.
Vol. 80, no. 11, pp. 1–16, 2020. [13] White, L., et al., Insider Threats in Health-
[6] Lateef, M. A. et al., Data Aegis Using Che- care: A Comprehensive Analysis and
byshev Chaotic Map-Based Key Authentica- Mitigation Strategies. Journal of Cyberse-
tion Protocol, In Intelligent Manufacturing curity Research, Vol. 8, no. 2, pp. 135–152,
and Energy Sustainability: Proceedings of 2021.
ICIMES, pp. 187–195, 2022. [14] Patel, A., and Wang, J., Securing IoT Devices
[7] C. Atheeq and M. Munir Ahamed Rabbani, in Medical Imaging: Vulnerabilities and
CACK—A Counter Based Authenticated Countermeasures.” Journal of Healthcare
ACK to Mitigate Misbehaving Nodes from Engineering, Vol. 10, pp. 1–15, 2019.
MANETs, Recent Advances in Computer [15] C. Atheeq and M. Munir Ahamed Rab-
Science and Communications, Vol. 14, no. 3, bani, Secure Data transmission in integrated
pp. 837–847, 2021. internet MANETs based on effective trusted
[8] S. Syed Abdul Syed et al., Enhanced Chaotic knowledge algorithm, Indian Journal of Sci-
Map Based Key Agreement Mitigate Packet ence and Technology, Vol. 9, no. 47, 2016.
16 Detection of phishing attacks using
machine learning
Siddharth1,a, Rajeev Srivastava2,b, Harsh Raj1,c, Shourya Dwivedi1,d,
Shourya Dwivedi1,e, Rohit Singh1,f, and Nitish Chaurasiya1,g
1
Department of Computer Science and Engineering, Babu Banarasi Das Institute of Technology and
Management, Lucknow, India
2
Assistant Professor, Department of Computer Science and Engineering, Babu Banarasi Das Institute
of Technology and Management, Lucknow, India
Abstract: Recent Developments in IOT Technologies (Review) with the increasing traffic in confidential data
being passed over networks, it becomes vulnerable to various security threats that can affect its confidentiality,
integrity and availability. Intrusion Detection System (IDS) helps to monitor and provide alerts in case of any
malicious activity on the network. Though many machine learning methodologies have had their successes in
the anomaly detection domain, very few attempt to reflect the sequential nature of network data. The authors of
this study explicitly apply a sequential methodology and choose to compare multiple models, including Random
Forests (RF) Multi-Layer Perceptrons (MLP), and Long-, Short-Term Memory (LSTM) on the CIDDS-001 data-
set. Our evaluations show that the sequential detection is more effective than typical point-wise approaches. In
the experimental results, it shows that long short-term memory significantly outperforms in detection order-
dependent traffic data logs of tracing code-level anomalies with 99.94% accuracy and an F1-score of 91.66%.
Keywords: machine learning, security prediction, cyber attacks detection, phishing detection
1. Introduction theft. It is a serious concern, as many inter-
net users fall victim to it. Phishing is a form
Phishing refers to the act of trying to obtain of social engineering where attackers pose as
sensitive information of users, by pretending trusted organizations in an automated man-
to be a legitimate entity in electronic commu- ner to trick users into disclosing sensitive
nication. Common phishing tactics involve information.
messages claiming to be from well-known The main objective of the review paper
social media platforms, online auction sites, is to give a detailed study on some of most
payment services, or IT administrators, aim- recent advancements in information and
ing to deceive unsuspecting individuals. Many communication technology. It is through the
systems are designed to reduce the impact of sharing of data concerning with users and
failures or attacks on critical infrastructure, homes where this network can be attacked
but these systems can also become targets of in such a way to erode security services, reli- AQ1:
cyber-attacks. ability, and availability[1,2].Intrusion detec-Please
note that
Cybercriminals have continuously evolved tion systems (IDS) are must-have security the cross-
their techniques for stealing information, tools looking to detect malicious activity reference
but social engineering attacks remain a pre- in host-based logs or network traffic. Over iscitation
not
ferred method. One such crime is phishing, the years many machine learning techniques and not
which allows attackers to commit identity have demonstrated their potential in anomaly in order
through-
out.
a
[email protected]; [email protected]; [email protected]; ddwivedishoury@
gmail.com; [email protected]; [email protected]; [email protected]
CORRECTED
DOI: 10.1201/9781003606635-16
130 Innovations in Computational Intelligence: Trends and Technologies
detection but less attention was paid to time- 2.2. Machine learning and phishing
series data [8]. This paper evaluates LSTM, attacks
ML and RF on a sequential task using the
CIDDS-001 dataset [3]. The performance Detection The deployment and advancement
metrics of the methodology presented here of machine learning techniques to detect
are compared with those obtained from tra- phishing attacks highlights strategic efforts to
ditional methodologies based on individual stay ahead of cyber attackers. In the dynamic
flows to determine when a particular tech- cyber security landscape, accurate and timely
nique is best suited in practice. Empirical detection of phishing attacks is critical to
evidence suggests that a case-based target is maintaining data confidentiality, integrity,
not the best performing target for anomaly and availability. This focus aligns with today’s
detection, and instead using a sequential one need for advanced tools that use data-driven
could be better. LSTM is highly reliable in approaches to strengthen cyber resilience and
finding the sequence patterns in network protect sensitive information.
traffic data i.e. with an accuracy of 99.94%
and f1-score of 91.66% [4].
This is the review paper that reveals how
3. Thematic Overview
machine learning methods have been used This review paper examines the emerging field
to detect phishing attack with more prefer- of machine learning-based phishing attack
ence on the sequential based method for detection. It will explore the diverse method-
these objectives. Phishing attempts are simi- ologies, algorithms, and datasets employed
larly considered independently in traditional in cybersecurity to enhance the identification
methods without treatment of the sequential and prevention of phishing threats. To guide
nature possible between cyber threats. In this our review, we formulated the following
paper, we report key findings and a novel research questions:
approach called majority voting among vari- What are the recent advancements and
ous decision-making algorithms. This paper methodologies in applying machine learning
evaluates the extent to which these tech- to detect phishing attacks?
niques improve phishing attack detection. How effective are traditional intrusion
The proposed method evaluates key machine detection systems (IDS) in identifying and
learning models such as Random Forest mitigating phishing attacks compared to
(RF), Multi-layer Perceptron (MLP) and machine learning-based approaches? What
Long-Short-Term Memory (LSTM) based on extent does considering the sequential nature
CIDDS-001 dataset. Taking the sequential of data contribute to the accuracy and effi-
perspective, this research seeks to provide ciency of phishing attack detection models.
more meaningful information regarding how These questions set the stage for a compre-
the models will perform and when it would hensive and systematic review of the relevant
be ethically valid to apply these in real world literature. Maintaining the Integrity of the
phishing detection scenarios. Specifications.
2. Methodology 3.1. Search criteria, databases, and
sources
2.1. Formulating a research question Our search strategy involved selecting key-
Languages other than English, lacking peer words and phrases relevant to the review topic
review, or not directly related to the core to ensure a thorough exploration of various
aspects of our topic. aspects of the subject. Boolean operators were
Detection of phishing attacks using machine learning 131
employed to effectively combine search terms • Machine Learning Models for Phishing
and optimize the retrieval of pertinent studies. Detection
We searched across multiple reputable data- • Survey of supervised learning tech-
bases, including Web of Science, Digital Library niques (e.g. decision trees, support vec-
etc. These databases were selected based on tor their machines) and applications in
their relation to the field and the breadth of feature-based detection.
academic literature they offer, allowing us to • An overview of unsupervised learn-
gather a wide range of sources. The primary ing approaches, including clustering
focus was on academic sources, such as peer- algorithms, for anomaly detection in
reviewed journal articles, conference papers, phishing patterns.
and scholarly books. These sources were con- • An introduction to deep learning
sidered the most reliable and authoritative for models such as neural networks
our review. We ensured that the sources were and recurrent neural networks for
recent (published after 2020) to reflect the lat- their ability to capture complex
est developments in the field. relationships.
• Discussion of relevant features used in
phishing detection models.
3.2. Search inclusion and exclusion • Evaluation of element selection tech-
criteria for selecting studies niques to increase model efficiency and
We made sure the data was as current as pos- reduce dimensions.
sible by include research that were released • Investigating feature extraction methods
after 2020. To guarantee consistency in the to uncover hidden patterns in phishing
review, research also have to be written in datasets.
English. We incorporated references that • Rating Datasets and Metrics
offered significant contributions to the field • Review of publicly available datasets
and directly addressed important facets of for training and testing machine learn-
Advancements in Assistive Technologies for ing models in phishing detection.
Web Accessibility. In addition, studies have to • Analysis of common evaluation met-
undergo peer review and be fully text-availa- rics (eg precision, recall, F1 score) used
ble for in-depth examination. to assess the performance of detection
We eliminated materials that didn’t fit our models.
inclusion requirements, such as those written • Challenges and Limitations
in before 2015 or published before 2015. • Identifying challenges in designing
robust phishing detection systems
• Introduction using machine learning.
• Definition of phishing attacks and • Discussion of limitations of current
their importance in the digital era. models, including issues related to
• Brief overview of traditional detection false positives/negatives and adversar-
methods and their limitations. ial attacks.
• Characteristics of a phishing attack
• Analysis of common phishing tactics 3.3. Summary of key findings and
and strategies.
• Investigating evolving phishing tech- insights from the review
niques such as spear-phishing and vishing. Highlighting the significance of ongoing
• Discussion of the problems posed by research and innovation in the evolving
polymorphic and targeted phishing field of machine learning for phishing attack
attacks. detection.
132 Innovations in Computational Intelligence: Trends and Technologies
4. Critical Analysis Eligibility: IDC with CIDDS-001 →
increases real world cyber threat candidacy.
In this critical analysis, we base our evalua- Comparative Analysis and Selection of
tion on the proposed sequential approach and Model: An extensive comparison will guide to
models random forest (RF), multi-layer per- figure out the optimal model for the genesis of
ceptron (MLP), and long- short-term memory models in sequence phishing attack detection.
(LSTM) to gather some strengths and chal- Real-world application: The study pro-
lenges of using machine learning for phish- vides insight for practical deployment of
ing attack detection. Wide range of detection machine learning models in the context of
techniques: The static technique takes place real-world cyber security use-cases.
in sequence which makes it more active for Ethical and Privacy: The meaning of
recognizing attacks and for covering the machine learning in cybersecurity and the
changing features of cyber threats. ethics and privacy implications.
Performance metrics and model evalua- Future Research Implications-Indicating
tion: The experiments show a high accuracy where additional work can be conducted,
(99.94%) and the f1-score (91.66%) for potentially using emerging technologies
LSTM which states that the ability of learn- or further defining the order of the steps
ing to sequential patterns is reliable. suggested.
Sr. No. Title of Paper Techniques used Dataset used Accuracy
1 Intelligent Cyber Attack Long-Short Term CIDDS-001 99.94%
Detection and Classification Memory (LSTM)
for Network-Based model
Intrusion Detection Systems
2 A Deep Learning Ensemble Long Short-Term IoT-23 99.8%
for Network Anomaly and Memory (LSTM) LITNET-2020
Cyber- Attack Detection logistic regression NetML-2020
meta-classifier
3 Battle of the Attack Data Analysis SCADA 99%
Detection Algorithms: Algo. Model Based
Disclosing Cyber Attacks on Detection Mechanism
Water Distribution Networks
4 An Efficient Deep- Learning- CNN NSL-KDD Data Set 99.3
Based Detection and
Classification System for
Cyber-Attacks in IoT
Communication Networks
5 A Novel Cyber Attack Recursive Algo Cyber Attack 99%
Detection Method in Prediction Data Set
network control system
6 An Ensemble Approach Data collection KDDCUP2009 Data 99.1%
for Cyber Attack detection and preprocessing Set
system: A Generic Feature extraction
Framework Classification of cyber
attacks
7 Phishing Attacks: A Recent Phishing filter Email UCI Machine 90%
Comprehensive Study and a authentication Learning Repository
New Anatomy Phishing Data Set
Detection of phishing attacks using machine learning 133
4. Recommendations for Future Examine Model Transferability: Assess the
transferability of trained models across dif-
Research
ferent organizational settings and industries.
To advance the field of phishing attack detec- Investigate the adaptability of the proposed
tion using machine learning, the following models to varying network architectures and
recommendations are proposed for future security configurations.
research: Model Adaptation and Refinement: Estab-
lish systems for ongoing monitoring and
• Dynamic Feature Engineering: Investi- updating of machine learning models to
gate dynamic feature engineering tech- counter evolving phishing tactics. Implement
niques that adapt to the evolving nature a feedback mechanism that incorporates new
of phishing attacks. Explore methods for threat intelligence and attack patterns to con-
automated extraction and incorporation tinuously refine model performance.
of relevant features in real-time. Collaborative Research: Promote collabo-
• Adversarial Machine Learning: Explore rative research initiatives among academia,
adversarial machine learning techniques industry, and cybersecurity professionals.
to enhance the robustness of the proposed Foster partnerships to facilitate the shar-
models against sophisticated phishing ing of datasets, methodologies, and insights,
attacks. Evaluate the models’ performance driving collective progress in phishing attack
under adversarial conditions to ensure detection.• Interdisciplinary Research: Fos-
real-world effectiveness. ter interdisciplinary research collaborations
• Integration of Explainable AI (XAI): that involve experts in cybersecurity, machine
Incorporate Explainable AI (XAI) tech- learning, human-computer interaction, and
niques to enhance the interpretability of ethics. Address the multifaceted challenges
the detection models. This ensures trans- associated with phishing attacks from a holis-
parency in decision- making and facilitates tic perspective.
trust among end-users and cybersecurity
professionals.
5. Conclusion
• Cross-Dataset Validation: Conduct exten-
sive cross-dataset validation to assess the The review articles provide an overview of
generalizability of the proposed sequential methods for detecting and classifying intel-
approach and models across diverse phish- ligent cyberattacks and phishing attacks,
ing attack scenarios. Consider datasets highlighting the evolution and diversifica-
with varying characteristics and sources. tion of techniques in network-based intru-
• Real-Time Implementation: Explore the sion detection systems. Advanced models like
feasibility of real- time implementation LSTM and CNN are effective, and ensemble
of the proposed models in live network methods emphasize the importance of diver-
environments. Assess the models’ perfor- sification for higher detection accuracy. Tech-
mance, latency, and resource requirements niques are applied in various environments,
to determine their practicality for deploy- including water distribution and IoT commu-
ment in cybersecurity operations. nication networks.
Consistently high accuracy across these
Human-Centric Security Measures: Inves- methodologies underscores the effectiveness
tigate the integration of human-centric secu- of machine learning in strengthening cyber
rity measures in conjunction with machine defenses.
learning models. Explore ways to incorporate While the studies collectively contribute to
user behavior analytics and user feedback to advances in cyber threat detection, it is clear
enhance the overall security posture. that the field is dynamic and requires constant
134 Innovations in Computational Intelligence: Trends and Technologies
innovation to address new challenges. As systems in intrusion detection: a case study.
cyber-attacks become more sophisticated, the In Proceedings of the 11th National Confer-
fusion of different models, algorithms and ence on Computer Security, pages 74, 81,
datasets is likely to remain the cornerstone Baltimore, Maryland, 2022.
for strengthening network security [6] Gopi K. Kuchimanchi, Vir V. Phoha, Kiran S.
Balagani, Shekhar R. Gaddam “Dimensional
reduction using feature extraction methods
for Real-time Misuse Detection Systems”
Refernces Proceedings of the Workshop on Informa-
[1] Marinova-Boncheva, V. Applying a Data tion Assurance and Security, US Army Acad-
Mining Method for Intrusion Detection. In: emy, West Point, NY, 10–11 June 2022.
International Conference on Computer Sys- [7] Steven R Snapp, Stephen E Smaha, Daniel
tems and Technologies CompSysTech’07, M Teal and Tim Grant. DIDS (Distributed
14–15 June 2020. Intrusion Detection System) prototype. In
[2] R. Heady, G. Luger, A. Maccabe and M. Ser- Proceedings of the USENIX Summer Con-
villa. Architecture network level intrusion ference, pages 227,233, San Antonio, Texas,
detection system. Technical report CS90-20, 8.12. June 2022. USENIX Association.
University of New Mexico, Department of [8] Sandeep Kumar and Eugene H. Spafford.
Computer Science, August 2021. Software architecture to support intrusion
[3] Abokifa, A.A., K. Haddad, C.S. Lo, and detection. Technical Report, The COAST
P. Biswas. 2021. “Detection cyber-physi- Project, Department of Comp. Sciences, Pur-
cal attacks on water distribution systems due Univ., West Lafayette, IN, 47907.1398,
through principal component analysis and USA, March 17, 2023.
artificial neural networks. In Proc., World [9] J.H. Eom and M.W. Park, “Design of Internal
Environmental and Water Resources Con- Traffic Checkpoint of Security Checkpoint
gress 2021, 676–691. Reston, VA: ASCE. Model in the Cloud Computing”, Interna-
[4] Cardenas, A.A., S. Amin, and S. Sastry. 2021. tional Journal of Security and Its Applica-
“Secure Control: Towards cyber-physical tions, Vol. 7, No. 1, (2023), pp. 119–128.
systems that survive.” In Proc., Conf. on Dis- [10] A. C. Kim, W. H. Park, and D. H. Lee, “A
tributed Workshops on Computing Systems Study on Live Forensic Techniques for
(ICDCS), 495–500. Piscataway, NJ: IEEE. Anomaly Detection in User Terminals,”
[5] Michael M. Sebring, Eric Shellhouse, Mary International Journal of Network Security,
E. Hanna, and R. Alan Whitehurst. Expert vol. 7, No. 1, (2023), pp. 181–188.
17 A comprehensive exploration of aircraft
detection in synthetic and satellite
images
M. Madhu Balaa, G. Ramesh Chandra, G. Pavan Teja, M. Sai
Sudha, Sk. Sabeeha Kouser, and M. Ajay Kumar
Computer Science and Engineering, VNR Vignana Jyothi Institute of Engineering and Technology,
Hyderabad, India
Abstract: Aircraft identification via satellite images is difficult because of the variance in aircraft appearance,
size, orientation, reliable dataset, and complex backgrounds. This aircraft detection is mainly used for defense
and military purposes to keep enemies in check. Even though various state-of-the-art algorithms have been
proposed to address aircraft detection the lack of structured datasets makes it difficult to achieve high accu-
racy, speed, and efficiency. This study focuses on end-to-end processes from datasets, aircraft detection frame-
works, algorithms, and performance on various applications. It includes the steps to collect and process both
synthetic and satellite sensing images, and the inclusion of image augmentation techniques. It is observed that
the majority of existing works on aircraft detection are the variants of convolutional neural networks (CNNs).
The analysis is done on these approaches critically throughout our survey by emphasizing the data sets, the
design parameters, accuracy of aircraft detection model performance, and the research openings to address in
the future.
Keywords: deep learning, Convolutional Neural Networks (CNN), Aircraft detection, Satellite imagery, Real-
time detection, YOLO, Custom dataset
1. Introduction contextual framework, we commence with a
study on emerging techniques, ranging from
Aircraft detection has significant applications traditional convolutional networks and image
in various contexts ranging from military/ processing, outlining the progression from
defense to civilian/ commercial. The promi- Fast RCNN, Faster RCNN, region-based
nent research signifies the improvement in CNN (RCNN), and YOLO. The criteria for
surveillance, monitoring, defense, and secu- model evaluation are Mean average preci-
rity. In general, this task was completed by sion (mAP), accuracy against recall, time, and
human observation which has limitations due computational complexity. It can help the
to human perception. However, object detec- selection of network designs for similar jobs.
AQ1: tion with convolution neural network (CNN) The accuracy and effectiveness of these
Please
note that [1,2] to present deep neural networksmodels have been completely transformed
the cross- [3,4] gives a significant change in the iden- by recent developments in object detection
reference
is not in tification of aircraft in high-resolution satel- methods based on deep learning. Research-
sequential lite images [5,6]. ers have successfully enhanced the speed and
order.
In this comprehensive exploration of air- accuracy of models such as Faster R-CNN,
craft detection methodologies to establish a SSD, and YOLO [7-10], incorporating
DOI: 10.1201/9781003606635-17
136 Innovations in Computational Intelligence: Trends and Technologies
default bounding boxes with varied aspect 2. Datasets
ratios for rapid and precise recognition of
objects. Building on YOLOv3 [4], creates a The aircraft datasets are categorized as satel-
deep learning-based ship detection model. lite images and synthetic images.
This model ensures photo integrity by using
a hashing algorithm (SHA-256) for detected 2.1. Satellite images
ship images. It combines supervised image The satellite image dataset are collected from
categorization, and YOLOv3 for object detec- the sources generated by Google Earth or the
tion, that employs picture segmentation and panchromatic band of satellite images like
semantic segmentation at the pixel level for JL-1, and GF-2 [6]. The Dataset of Object
better and precise categorization of objects. Detection in Aerial Images (DOTA) [6,8] in
The research [5] aims to enhance detection, three versions (v1.0, v1.5, and v2.0) is avail-
accuracy, and overall performance in High- able in both RGB and grayscale. It covers
Resolution Satellite images from multi-scale various categories of objects (planes, ships,
and small-item detection by introducing a fast helicopters, soccer ball fields, etc.) used for
method capable of handling the challenges object detection applications. Figure 17.1 of
posed by small-sized and varied-scale objects. (a), (b), and (c)) shows a sample DOTA air-
In [8], the study addresses challenges existed plane category image with OBB representa-
in object segmentation and recognition in tion of three versions (v1.0, v1.5 and v2.0)
high-resolution satellite images, characterized waveform of BLDC motor. For aircraft detec-
by dense targets and varied backgrounds. tion among all categories, planes and helicop-
Traditional approaches face difficulties due ters are considered.
to large field of view and challenging target- Annotation: To increase the size of data-
to-background ratios. The research [9] aims set with variant orientations, each aircraft is
to create an automated aircraft identifica- rotated by an Oriented Bounding Box (OBB)
tion system focused on aircraft with high- which is a rectangular box aligned with the
resolution satellite images, using YOLOv3 to orientation of an object. The general repre-
increase the accuracy and efficiency of iden- sentation of an OBB is represented as (x1, y1,
tifying planes in extremely HRS photos. In x2, y2, x3, y3, x4, y4) where (xi, yi) denotes
[10], majorly focused on the issue of identify- a i-th vertices of the Oriented Bounding Box.
ing small-scale aircraft in geo-spatial images The NWPU VHR-10[5] dataset contains
used to develop a faster-RCNN-based aircraft satellite images that were attained from the
detection system that will improve the accu- source - Google Earth. This geospatial dataset
racy and effectiveness. In [11], the problem of contains ten classes used for object detection.
recognizing planes in remote sensing images The ten classes labeled as baseball, basketball
is addressed. They developed an aircraft rec-
ognition model based on the YOLOv5 archi-
tecture that will improve accuracy, and speed
and help in identifying aircraft in satellite
images.
The rest of the paper is accommodated in
sub-sections, section 2 has various aircraft
datasets from satellite to synthetic images, AQ2:
section 3 discusses variant frameworks with Figure 17.1. A Sample DOTA image of airplane Please
design parameters, and Section 4 gives an category with OBB representation (a) v1.0 provide
source
analysis of existing works with identified (b) v1.5 and (c) v2.0waveform of BLDC motor. for all the
research gaps. Source: figures
A comprehensive exploration of aircraft detection in synthetic and satellite images 137
court, harbor, storage tank, airplane, tennis systems adopted variant CNN models like
court, fields, bridge, ship, and vehicle. Dectctron2, RCNN, Faster RCNN, SSD,
SIMRDWN, and state-of-the-art YOLO
2.2 Synthetic Images models.
Synthetic images are computer generated
images based on satellite real world images. 3.1. Detectron2
It is generated algorithmically and serves as Detectron2 is library developed by Facebook
a stand-in for test data sets of production or AI Research’s, which offers pioneering dis-
operational data, as well as for validating covery and segmentation. It uses FR-CNN as
mathematical models and training machine its foundation for the object detection frame-
learning (ML) models. This is now widely work. The architecture of Detectron2 is illus-
used in various machine vision realms for trated in Figure 17.4 [25].
enhancing the training data and improving The mathematical models of the RPN and
the accuracy of the model. Many synthetic BoxHead, two main components of Detec-
datasets, such as the Synthia, GTA, and tron2 are given in Eq.1 and Eq. 2.
vKITTI datasets, are devoted to autonomous
driving [34]. Figure 17.2 illustrates a syn- RPN(r_x) = (pr_x, t_x) (1)
thetic aircraft image. BoxHead(b_x) = (pr_c, t_c) (2)
Annotation: The synthetic aircraft image
dataset comprises 629, 551 annotations hav- In Eq. (1), pr_x is the given image. An
ing 50, 000 images of size (1920×1080) pix- object’s presence in the window is indicated
els at 15 diverse locations covering an extent by a probability score vector(pr_x). t_x is
of 9331.2 km [34].
3. Object Discovery Frameworks
In general, the applications of object discov-
ery can be done in two ways: direct and Cas-
caded. Figure 17.3 provides aircraft detection
framework using YoLo. Aircraft detection
Figure 17.3. Aircraft detection framework using
YOLO.
Source:
Figure 17.2. A Sample synthetic Image. Figure 17.4. Architecture of Detectron2.
Source: Source: from Ref.[25].
138 Innovations in Computational Intelligence: Trends and Technologies
a matrix of bounding box predictions. In
Eq.(2), b_x is the given image. pr_c is a vector
of probability scores for each class. t_c is a
matrix of bounding box predictions for each
class.
3.2. EfficientDet Figure 17.5. FR-CNN architecture in aircraft
EfficientDet uses a variety of optimization detection [1].
and structural changes and a compound scal- Source:
ing method that scales all networked lay-
ers, simultaneously and consistently. The anchor box.t_c is a tensor of bounding box
TransEffiDet algorithm was proposed for predictions for each anchor box.
object detection in aerial photos using the
Transformer module and the EfficientDet 3.4. Single shot detector (SSD)
[14,21,26] method. The EfficientNet back-
bone network’s mathematical model is given SSD, a one-step recognition of objects tech-
in Eq. 3,4 and 5. nique utilizes CNN for classification. SSD
architecture and the mathematical model is
EfficientNet(im) = F(im; W, b) (3) given in Eq. (7).
Transformer(x) = Dc(Ec(x)) (4)
DetectionHead(x) = (p_c, t_c) (5) SSD (image) = (pr_c, tn_c) (7)
p_c = sigmoid (W_c * F(image)) (8)
In Eq. (3), the input image is im. The net- t_c = exp (W_t * F(image)) + b_t (9)
work’s weights and biases are represented by
W and b. The mathematical operations make In Eq. (7), the image is the input. pr_c is a
up F. In Eq. (4) x is the input sequence. The tensor of class probabilities for each anchor
Ec function is the encoder. Dc is the function box.tn_c is a tensor of bounding box(BB)
that decodes. In Eq. (5), x is the backbone estimates for each anchor box. In Eq. (8),
network’s input feature set. A vector of each Eq. (9) the feature pyramid network’s output
class’s probability scores (p_c). A matrix of is denoted by F(image), The weights for the
bounding box predictions for every class(t_c). classification and localization heads are W_c
and W_t, respectively. The localization head’s
bias is b_t.
3.3. Faster RCNN
Faster Region-CNN(FR-CNN) a variant of 3.5. You Only Look Once (YOLO)
deep neural network appears to the user as
end-to-end, single, cohesive. The FR-CNN YOLO is a contemporary object discovery
architecture is demonstrated in aircraft dis- approach and has been enhanced over time.
covery in Figure 17.5 with two main compo- The mathematical model for all YOLO mod-
nents [1]. els a to f. Input is taken as size (h_o,w_o,n_c)
The mathematical model of the FR-CNN input image, with h_o representing the height,
is given in eq.-6. w_o the representing width, and n_c the total
number of channels. Overall the result is a
Faster Region – CNN(image) = (p_c, t_c) (6) list of bounding boxes, each represented by a
5-dimensional vector (x_center,y_center, w_o,
In Eq. (6), the image is the input image. h_o, class_object), where c_object denotes the
p_c is a tensor of class probabilities for each class of the object inside the bounding box,
A comprehensive exploration of aircraft detection in synthetic and satellite images 139
w_o width , and h_o height. The mathemati- 3.6. Innovations of new technologies
cal model of vector is given in Eq. (10), Eq. in aircraft detection
(11), Eq. (12), and Eq. (13).
The ongoing development of new detectors,
x_c=(x_max+x_min)/2(10) along with upgrades to existing ones (e.g.,
y_c=(y_max+y_min)/2(11) newer versions of YOLO), brings significant
w_o=(x_max_x_min)(12) innovations in feature representation, con-
h_o = (y_max+y_min) (13) text encoding, refinement, localization, and
robustness. These advancements enhance
In this Max. and Min. of the x,y coordi- localization accuracy and introduce anchor-
nates of the object BB are represented as x_ free detection methods. Robustness in detec-
max, y_max and x_min, y_min. tion is further improved through various
YOLOv2—The object discovery in sin- techniques, such as rotation-invariant loss
gle shot was performed by YOLOv2. Its functions, scale-adaptive detection, enriched
image’s fundamental architecture consists of feature learning, and multi-task loss func-
an anchor box, a feature pyramid network tions. Additionally, approaches like adver-
(FPN), and a CNN [27]. sarial learning, capsule networks, knowledge
YOLOv3—YOLOv3, the first object distillation, lightweight object detection,
detection model to use multiscale feature memory-efficient networks, and transfer
fusion at three different scales, the anchor learning contribute both directly and indi-
structures are used in predicting the bound- rectly to these improvements.
ing box and class for every component in the
image [28].
YOLOv4—YOLOv4 is faster than earlier
4. Analysis of the YOLO Aircraft
YOLO models as a result of its effective CSP- Detection Models
Darknet53 backbone and mosaic data aug- The image shows a table comparing the per-
mentation [23]. formance of different YOLO models on a
YOLOv5—YOLOv5 uses an innovative dataset of 2000 images. Table 17.1 compares
method known as dynamic anchor boxes the various models according to their speed,
made using the centroids of the clusters precision, and accuracy.
formed by the ground boxes [7].
YOLOv6—YOLOv6 uses Ghost Modules
to minimize the number of parameters. On a 4.1. Speed
single GPU, it can discover objects in real-time The Yolov2 model, which has an inference
[31]. In comparison to Detectron2 YOLOv6 speed of 23 ms, is the fastest. The Yolov3
performs significantly well in standings of model has an inference speed of 170 ms to
accuracy and speed [22]. 190 ms, which is somewhat slower. The
YOLOv7—YOLOv7 is more robust to inference speed of the Yolov4 model is even
noise and outliers, due to its focus on v3 loss slower, ranging from 15 to 17 ms. The slow-
function, which makes it ideal for detect- est model, with an inference speed of 12–14
ing aircraft/aerial images with low-quality ms, is the Yolov5 model.
images too.
YOLOv8—YOLOv8, the latest version
uses a modernized backbone network built
4.2. Precision
on top of EfficientNet, which enhances The Yolov4 model has the highest preci-
the model’s capacity to extract elaborate sion, with a precision of 95.30%-96.40%.
characteristics. The Yolov5 model is close behind, with a
140 Innovations in Computational Intelligence: Trends and Technologies
Table 17.1. Results of algorithms used
Reference Model Dataset Type Dataset Speed Precision Accuracy
size (%) (%)
[27] YOLOv2 Custom Satellite 2000 23ms 94.20 89.60
[28] YOLOv3 EfficientNet Satellite 2000 170ms– 77 94.20
B0 190ms
[29] YOLOv4 CSP Satellite 2000 15ms–17ms 85 95.30
Darknet53
[30] YOLOv5 Swin- Satellite 2000 12ms–14ms 90.50 95.60
Transformer
[31] YOLOv6 Ghost Satellite 2000 10ms–12ms 92 96
Modules
[32] YOLOv7 Focus v3 Satellite 2000 9ms–11ms 93 96.40
loss
[26] TransEffiDet Custom Synthetic 2000 14ms–16ms 95.80 97.20
[5] Multiscale NWPU Satellite 8000 180ms– 89.60 94.00
CNN VHR-10 200ms
[2] SIMRDWN Custom Satellite 2000 5ms–103ms 97 97.50
[2] SSD Custom Satellite 2000 20ms–22ms 93.50 95.20
[33] Detectron2 Pascal VOC Satellite 10000 150ms– 50 91.50
170ms
[33] Detectron2 Visdrone Satellite 2000 160ms– 50 92.50
AQ3: 180ms
Please
provide [33] Detectron2 NWPU Satellite 800 170ms– 50 93.00
source for VHR-10 190ms
the table
Source:
precision of 94%-97.20%. The Yolov3 model lowest accuracy, with an accuracy of 89.60%.
has a slightly lower precision, with a preci- Experimental results show that TransEffiDet
sion of 94.20%. The Yolov2 model has the outperforms EfficientDet, with 86.6% for
lowest precision, with a precision of 94.20%. the mean Average Precision (mAP), and an
Multi-scale CNN’s accuracy after training improvement of approximately 5.8% in mAP.
was 89.6% mAP (mean average precision).
This outperforms the accuracy of the other 4.4. Which model to choose
approaches that the authors contrasted.
The individual application requirements to
determine which model is best. In cases where
4.3. Accuracy velocity is the primary consideration, the
The Yolov5 model has the highest accu- Yolov2 model could be the optimal option.
racy, with an accuracy of 96%-97.50%. The If accuracy and precision are the most crucial
Yolov4 model is close behind, with an accu- considerations, the Yolov5 model might be
racy of 95.60%-96.80%. The Yolov3 model the best option. The Yolov4 model might be
has a slightly lower accuracy, with an accu- the best option if speed and accuracy/preci-
racy of 94.20%. The Yolov2 model has the sion are equally important.
A comprehensive exploration of aircraft detection in synthetic and satellite images 141
5. Conclusion very high-resolution satellite images. Remote
Sensing, 12(4), 458. https://doi.org/10.3390/
A comprehensive exploration of the research rs12040458
highlighted the remarkable potential of DL- [4] Gadamsetty, S., Ch, R., Ch, A., Iwendi, C.,
based object discovery methods in address- and Gadekallu, T. R. (2022). Hash-based
ing the intricate challenges of aircraft deep learning approach for remote sensing
detection, and object recognition in remote satellite imagery detection. Water, 14(5),
sensing. The findings revealed that these 707. https://doi.org/10.3390/w14050707
advanced techniques offer enhanced preci- [5] Guo, W., Yang, W., Zhang, H., and Hua,
sion, robustness, and real-time performance, G. (2018). Geospatial object detection in
making them well-suited for practical appli- high-resolution satellite images based on
multi-scale convolutional neural networks.
cations in various fields, including military
Remote Sensing, 10(1), 131. https://doi.
and space imaging. The experimental results org/10.3390/rs10010131
consistently showcased the advantages of [6] Niu, R., Zhi, X., Jiang, S., Gong, J., Zhang,
these methods, with YOLO-based models W., and Yu, L. (2023). Aircraft target
such as YOLOv2, YOLOv8, and TransEffi- detection in low signal-to-noise ratio vis-
Det standing out in terms of accuracy and ible remote sensing images. Remote Sens-
performance. Overall, the studies highlight ing, 15(8), 1971. https://doi.org/10.3390/
the state-of-the-art advancements in Deep rs15081971
learning variants in aircraft detection mod- [7] Chen, Y., Yang, J., Wang, X., Zhou, J., Zou,
els, showcasing significant improvements in J., and Li, Y. (2022). An improved YOLOv5
both accuracy and efficiency for detecting real-time detection method for aircraft tar-
objects across various remote sensing and get detection. In 2022 27th International
aerial imaging domains. Conference on Automation and Computing
(ICAC) (pp. 1-6). https://doi.org/10.23919/
This article helps to work in investigating
ICAC55699.2022.9946279
target detection in difficult situations, expand- [8] Wu, Q., Feng, D., Cao, C., Zeng, X., Feng,
ing the applicability of these approaches to Z., Wu, J., and Huang, Z. (2021). Improved
different contexts, and tackling complicated Mask R-CNN for aircraft detection in
aspects such as occlusion and background remote sensing images. Sensors, 21(8), 2618.
distractions. https://doi.org/10.3390/s21082618
[9] Lin, Y.-C., and Chen, W.-D. (2021). Auto-
matic aircraft detection in very-high-resolu-
References tion satellite imagery using a YOLOv3-based
[1] Basim, A., Khan, M. J., Bhatti, F. A., Rah- process. Journal of Applied Remote Sensing,
man, M. M., Hussain, S. F., Hashmi, A. J., 15(4), 047504. https://doi.org/10.1117/1.
and Khurshid, K. (2022). Aircraft detection JRS.15.047504
in satellite imagery using deep learning-based [10] Zhang, Y., Song, C., and Zhang, D. (2022).
object detectors. Journal of Computers and Small-scale aircraft detection in remote sens-
Electrical Engineering, 104, 630. https://doi. ing images based on Faster R-CNN. Mul-
org/10.1016/j.compeleceng.2021.104630 timedia Tools and Applications, 81(12),
[2] Tahir, A., Munawar, H. S., Akram, J., Adil, 18091-18103. https://doi.org/10.1007/
M., Ali, S., Kouzani, A. Z., and Mahmud, s11042-021-11556-5
M. A. P. (2022). Automatic target detec- [11] Jindal, M., Raj, N., Saranya, P., and V, S.
tion from satellite imagery using machine (2022). Aircraft detection from remote
learning. Sensors. https://doi.org/10.3390/ sensing images using YOLOv5 architec-
s22020668 ture. In 2022 6th International Conference
[3] Alganci, U., Soydas, M., and Sertel, E. on Devices, Circuits and Systems (ICDCS)
(2020). Comparative research on deep learn- (pp. 332-336). https://doi.org/10.1109/
ing approaches for airplane detection from ICDCS53305.2022.10018847
142 Innovations in Computational Intelligence: Trends and Technologies
[12] Yaban, B., Alganci, U., and Sertel, E. (2022). IEEE Access, 11, 22430-22440. https://doi.
Aircraft detection in very high-resolution org/10.1109/ACCESS.2023.3249206
satellite images using YOLO-based deep [20] Gadamsetty, S., Ch, R., Ch, A., Iwendi, C.,
learning methods. In Proceedings of the and Gadekallu, T. R. (2022). Hash-based
4th Intercontinental Geoinformation Days deep learning approach for remote sensing
(IGD) (pp. 270–273). Tabriz, Iran. satellite imagery detection. Water, 14(5),
[13] Li, Y., Zhou, L., Yan, H., Shan, Y., Zheng, 707. https://doi.org/10.3390/w14050707
C., Liu, Y., Zuo, X., and Qiao, B. (2021). [21] Wang, Y., Wang, T., Zhou, X., Cai, W., Liu, R.,
Aircraft detection for remote sensing images Huang, M., Jing, T., Lin, M., He, H., Wang,
based on deep convolutional neural net- W., and Zhu, Y. (2022). TransEffiDet: Aircraft
works. Journal of Signal Processing Sys- detection and classification in aerial images
tems, 93, 4685644. https://doi.org/10.1007/ based on EfficientDet and Transformer.
s11265-021-01602-7 *Computational Intelligence and Neurosci-
[14] Gong, S., Wang, Y., Wang, T., Zhou, X., Cai, ence*, Article ID 2262549, 10 pages. https://
W., Liu, R., Huang, M., Jing, T., Lin, M., He, doi.org/10.1155/2022/2262549
H., Wang, W., and Zhu, Y. (2022). TransEf- [22] Adegun, A. A., Fonou Dombeu, J. V., Viriri,
fiDet: Aircraft detection and classification S., and Odindi, J. (2023). State-of-the-art
in aerial images based on EfficientDet and deep learning methods for objects detection
Transformer. Computational Intelligence in remote sensing satellite images. *Sen-
and Neuroscience, Article ID 2262549. sors*, *23*(5849). https://doi.org/10.3390/
https://doi.org/10.1155/2022/2262549 s23135849
[15] Chen, Y., Yang, J., Wang, X., Zhou, J., Zou, [23] Tahir, A., Adil, M., and Ali, A. (2023). Rapid
J., and Li, Y. (2022). An improved YOLOv5 detection of aircrafts in satellite imagery
real-time detection method for aircraft tar- based on deep neural networks. *arXiv pre-
get detection. In 2022 27th International print* arXiv:2304.11677. https://arxiv.org/
Conference on Automation and Computing abs/2304.11677
(ICAC) (pp. 1-6). https://doi.org/10.23919/ [24] Zhao, L., and Zhu, M. (2023). MS-
ICAC55699.2022.9946279 YOLOv7: YOLOv7 based on multi-scale
[16] He, W., Huang, Z., Wei, Z., Li, C., and Guo, for object detection on UAV aerial pho-
B. (2019). TF-YOLO: An improved incre- tography. *Drones*, *7*(188). https://doi.
mental network for real-time object detec- org/10.3390/drones7080188
tion. Applied Sciences, 9(16), 3225. https:// [25] Sarker, M. M. K., Makhlouf, Y., Craig, S.,
doi.org/10.3390/app9163225 and Humphries, M. (2021). A means of
[17] Hu, G., Yang, Z., and Han, J. (2018). Air- assessing deep learning-based detection
craft detection in remote sensing images of ICOS protein expression in colon can-
based on saliency and convolutional neu- cer. *BMC Cancer*, *21*(1). https://doi.
ral networks. Journal of Wireless Commu- org/10.1186/s12885-021-08341-w
nications and Networking, Article ID 130. [26] Tan, M., Pang, R., Chen, Q., and Pang,
https://doi.org/10.1186/s13638-018-1129-0 J. (2020). EfficientDet: Scalable and effi-
[18] Wang, Y., Zhang, Y., Yang, Y., and Chen, cient object detection. In *Proceedings of
J. (2020). Convolutional neural network the IEEE/CVF Conference on Computer
based weakly supervised learning for air- Vision and Pattern Recognition (CVPR)*
craft detection from remote sensing images. (pp. 10781-10790). https://doi.org/10.1109/
IEEE Transactions on Geoscience and CVPR42600.2020.01078
Remote Sensing, 58(9), 6392-6403. https:// [27] Seong, J., Song, S., Yoon, J.-H., and Kim, D.
doi.org/10.1109/TGRS.2020.2978922 (2019). Determination of vehicle trajectory
[19] Padilla Carrasco, D., Rashwan, H. A., through optimization of vehicle bounding
García, M. A., and Puig, D. (2023). T-YOLO: boxes using a convolutional neural net-
Tiny vehicle detection based on YOLO and work. *Sensors*, *19*(4263). https://doi.
multi-scale convolutional neural networks. org/10.3390/s191942638.
A comprehensive exploration of aircraft detection in synthetic and satellite images 143
[28] Ammar, A., Koubaa, A., Ahmed, M., *arXiv preprint* arXiv:2107.12407. https://
Saad, A., and Benjdira, B. (2021). Vehicle arxiv.org/abs/2107.12407
detection from aerial images using deep [32] Jocher, G., Zhao, C., Xiao, H.,Yan, J., and Wang,
learning: A comparative study. *Electron- C. (2022). YOLOv7: Train once, deploy any-
ics*, *10*(820). https://doi.org/10.3390/ where. *arXiv preprint* arXiv:2209.00202.
electronics10070820 https://arxiv.org/abs/2209.00202
[29] Bochkovskiy, A., Wang, C., and Liao, H. [33] Akshayanivashini, C. V., and Krisvanth, P.
(2020). YOLOv4: Optimal speed and (2023). Deep learning-based instance segmen-
accuracy of object detection. *arXiv pre- tation of aircraft in aerial images using Detec-
print* arXiv:2004.10934. https://arxiv.org/ tron2. *arXiv preprint* arXiv:2305.10211.
abs/2004.10934 https://arxiv.org/abs/2305.10211
[30] Bochkovskiy, A., Wang, C., and Liao, H. [34] Shermeyer, J., Hossler, T., Etten, A., Hogan,
(2020). YOLOv5: Optimal speed and D., Lewis, R., and Kim, D. (2021). Rare-
accuracy of object detection. *arXiv pre- Planes: Synthetic data takes flight. *Pro-
print* arXiv:2012.12267. https://arxiv.org/ ceedings of the IEEE/CVF Conference on
abs/2012.12267 Computer Vision and Pattern Recognition
[31] Wang, C., Liao, H., Wu, Y., and Yan, J. (CVPR)*, 207-217. https://doi.org/10.1109/
(2021). YOLOv6: Next-level object detec- CVPR52688.2021.00207
tion with transformer and decoupled head.
18 Mitigating man-in-the-middle attack
in UAV network using authentication
mechanism based on chaotic maps
C. Atheeq1,a, Layak Ali2,b, C. Altaf 3,c, and Aleem Mohammed4,d
1
Department of CSE, GITAM University, Hyderabad, India
2
Department of ECE, Central University of Karnataka, Karnataka, India
3
Department of ECE, Lords Institute of Engineering and Technical, Hyderabad, India
4
Computer Society of Australia, Australia
Abstract: The escalating utilization of Unmanned Aerial Vehicle (UAV) networks across various domains accen-
tuates the critical need for robust security protocols. UAV networks, being susceptible to diverse security chal-
lenges, especially Man-in-the-Middle (MitM) attacks, require innovative authentication mechanisms to fortify
their defenses. Recognizing the limitations of existing methods, the proposed chaotic-map-based authentica-
tion system dynamically generates cryptographic keys, introducing a pioneering approach to counteract MitM
threats effectively. By harnessing the inherent unpredictability of chaotic maps, the system not only rectifies
current authentication shortcomings but also exhibits superior resilience against sophisticated MitM attacks.
Extensive simulations conducted using ns3 validate the proposed system’s computational efficiency and its
heightened defense mechanisms, solidifying its position as an advanced and effective security enhancement for
UAV networks in real-world scenarios.
Keywords: Attack, UAV, authentication, chaotic maps, network
1. Introduction The increasing integration of UAVs into
civilian airspace introduces complex secu-
Unmanned Aerial Vehicles (UAVs) have wit-
rity challenges. UAV communication, often
nessed widespread adoption across diverse
occurring over open and unsecured channels,
industries due to their versatility and capabil-
becomes a prime target for malicious actors
ity for applications such as precision agricul-
ture, environmental monitoring, and disaster seeking to compromise data integrity and
response. As UAVs become integral to critical confidentiality. Existing security mechanisms,
infrastructures, the vulnerabilities associated while effective to some extent, face limita-
with their communication networks raise tions in dealing with evolving cyber threats,
significant security concerns. The reliance emphasizing the need for more robust and
on wireless communication exposes UAVs to adaptive solutions [2]. The motivation for
cyber threats, including unauthorized access, this research stems from the critical need to
eavesdropping, and sophisticated Man-in- address the vulnerabilities of UAV networks,
the-Middle (MitM) attacks [1]. These security with a particular focus on the prevalent
challenges necessitate innovative measures to and potentially devastating MitM attacks.
safeguard UAV networks, especially in the MitM attacks on UAV networks can manip-
context of their increasing use in sensitive ulate crucial data exchanges between UAVs
and privacy-sensitive operations. and ground stations, posing severe risks to
[email protected]; [email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-18
Mitigating man-in-the-middle attack in UAV network using authentication mechanism 145
privacy, data integrity, and overall UAV net- implications, and charting directions for
work functionality [3,4]. Recognizing the future research.
urgency of this challenge, the research seeks
to contribute a cutting-edge authentication 2. Literature Review
mechanism to fortify UAV networks against
these evolving threats. 2.1. Quantum key distribution
This research aims to propose an advanced (QKD): enabling secure UAV
authentication mechanism tailored for UAV
networks, with a primary focus on mitigat- communications
ing MitM attacks. Building on the inherent Quantum Key Distribution (QKD) is an
unpredictability of chaotic maps, the pro- advanced cryptographic technique designed
posed system seeks to bolster cryptographic to secure communication channels against
key generation, thereby enhancing the overall eavesdropping, providing a quantum-safe
security posture of UAV networks. Rigorous key exchange mechanism. This innovative
simulations and evaluations will be conducted approach relies on the fundamental princi-
to empirically demonstrate the effectiveness ples of quantum mechanics, offering a level
of the proposed system and its superiority of security that is theoretically immune
over existing authentication methods [5,6]. to computational attacks, including those
The scope of this research is delimited to anticipated from quantum computers in the
addressing the specific challenge of MitM future. QKD addresses a critical aspect of
attacks within UAV networks. The signifi- secure communication – the establishment of
cance of this research lies in its potential to secret keys between two parties, which can
not only address immediate security concerns then be used for encrypting and decrypting
but also to establish a foundation for future messages [9]. Unlike classical key exchange
innovations in securing UAV communication. methods, QKD leverages the principles of
By concentrating on the specific threat vector quantum mechanics, such as superposition
of MitM attacks, the study seeks to make a and entanglement, to ensure the security of
valuable contribution to the broader field of the key exchange process.
UAV network security, considering the evolv- The QKD process typically involves two
ing nature of cyber threats [7,8]. parties, Alice and Bob, communicating over
The subsequent sections of this paper are a quantum channel. The fundamental idea
meticulously organized to provide a com- is to use quantum properties to detect the
prehensive understanding of the research. presence of an eavesdropper, often referred
Section 2 conducts an extensive literature to as Eve.
review, elucidating the existing landscape of Quantum Superposition: QKD employs
UAV network security, and critically examin- the principle of superposition, allowing quan-
ing the strengths and weaknesses of prevalent tum bits or qubits to exist in multiple states
authentication methods. Section 3 meticu- simultaneously. Alice sends a stream of qubits
lously details the proposed chaotic-map- to Bob, and Eve’s attempt to measure these
based authentication system, underlining its qubits inevitably disturbs their states.
operational principles and its capacity to fill Quantum Entanglement: QKD often uti-
existing security gaps. Section 4 presents the lizes entangled particles, where the state of
empirical results obtained through rigorous one particle is directly related to the state
simulations and engages in a detailed discus- of its entangled partner. This entanglement
sion, substantiating the proposed system’s allows Alice and Bob to share correlated
efficacy. The paper cin Sec oncludes tion 5, information, making it easier to detect any
summarizing key findings, exploring broader interference by an eavesdropper.
146 Innovations in Computational Intelligence: Trends and Technologies
Measurement and Detection: Bob receives on attack strategies, understanding motives,
the qubits from Alice and measures their and diverting malicious activities from
states. Alice and Bob then share a subset of authentic network resources.
their key bits, comparing them to identify any In assessing honeypot efficacy, various
discrepancies caused by interference. If the metrics come into play, such as the probabil-
quantum channel remains undisturbed, they ity of attacker interaction and honeypot effi-
proceed to use these bits as their secret key. ciency. The former, expressed as
Privacy Amplification: To enhance the
security of the generated key, privacy ampli- (1)
fication techniques are employed. This step
involves further manipulation of the key gauges the likelihood of an attacker engaging
bits to reduce the information available to a with the honeypot. The latter, defined as
potential eavesdropper.
(2)
QKD is actively researched, and sev-
eral commercial implementations are avail- quantifies the honeypot’s ability to intercept
able [10]. While challenges persist, ongoing and analyze malicious activities.
advancements in quantum technologies, such Honeypots offer a valuable defense mech-
as the development of quantum repeaters, anism by luring attackers, providing insights
aim to extend the range and practicality of into intrusion techniques. However, chal-
QKD. As quantum computing technologies lenges persist, including the risk of misuse
advance, QKD remains a critical component and the resource-intensive nature of high-
in the pursuit of quantum-safe cryptographic interaction honeypots [12].
solutions. Quantum Key Distribution stands In the realm of honeypot technology, ongo-
at the forefront of quantum-safe cryptogra- ing advancements focus on dynamic hon-
phy, leveraging the inherent properties of eypots and seamless integration with threat
quantum mechanics to secure UAV commu- intelligence. Future trajectories involve the
nications. Its ability to detect eavesdropping integration of machine learning algorithms
attempts and provide a secure key exchange for real-time threat detection. Honeypot tech-
mechanism positions QKD as a key player in nology plays a pivotal role in cybersecurity,
the evolving landscape of quantum-resistant employing deceptive strategies to engage and
security solutions. divert potential attackers
2.2. Honeypot technology (HT): 2.3. Dynamic key rotation
decoy defense strategies protocol (DKRP): enhancing
Honeypot technology, a stalwart in cyberse- cryptographic security
curity, strategically deploys deceptive assets
The Dynamic Key Rotation Protocol (DKRP)
within a network to emulate vulnerabilities
is designed to fortify cryptographic security
and services, thereby diverting and trapping
through the periodic rotation of keys, a pro-
potential attackers [11]. This detailed explo-
active measure aimed at mitigating the risks
ration unveils the operational intricacies,
associated with prolonged key usage. Let’s
effectiveness, and potential advancements of
delve into the operational dynamics and
honeypot technology.
inherent challenges of DKRP, emphasizing its
Operational Principles: Honeypots func-
vulnerability to advanced Man-in-the-Middle
tion as decoy systems, mirroring genuine
(MitM) attacks that specifically target key
assets to attract malicious actors. This proac-
rotation intervals [13].
tive approach aims at gathering intelligence
Mitigating man-in-the-middle attack in UAV network using authentication mechanism 147
Operational Dynamics: DKRP operates learning algorithms to dynamically adjust the
on the principle of regularly updating cryp- rotation interval (Trotation) based on real-time
tographic keys, aiming to ensure a dynamic threat assessments is a promising avenue for
and resilient security posture. The protocol research:
introduces a parameter Trotation, representing
the time interval between key rotations. The Trotation = fML(t)(7)
key rotation process can be mathematically
expressed as: Here, fMLrepresents the machine learning
algorithm determining the optimal rotation
Kn+1 = f(Kn,Trotation) (3) interval at time t.
While DKRP’s periodic key rotation
It signifies that the next cryptographic key enhances cryptographic security, the incorpo-
(Kn+1) is a function (f) of the current key (Kn) ration of mathematical models and stochastic
and the predefined rotation interval (Trotation). elements is crucial to fortify it against sophis-
Key Rotation Interval Vulnerabilities: ticated MitM attacks. Future advancements
Despite its operational strengths, DKRP is should focus on dynamic adaptations through
vulnerable to advanced MitM attacks that machine learning, ensuring a robust defense
exploit the predictability of key rotation against evolving cybersecurity threats [14].
intervals. The probability (Pattack) of a success- While Dynamic Key Rotation Proto-
ful MitM attack during a rotation interval col (DKRP) aims to enhance cryptographic
can be modeled using a probability distribu- security through periodic key rotation, its
tion function: vulnerability lies in the predictability of rota-
tion intervals. Advanced MitM attacks, spe-
Pattack(t)=e−λt (4) cifically targeting these intervals, can exploit
the deterministic nature of DKRP, compro-
Here, t represents the time within the rota- mising its effectiveness. The lack of inherent
tion interval, and λ is a parameter governing randomness in the key rotation process can
the attack rate. be exploited by attackers, limiting DKRP’s
Mitigation Strategies: To fortify DKRP ability to thwart evolving and sophisticated
against advanced MitM attacks, incorporat- MitM threats effectively. Despite their respec-
ing randomness into the key rotation process tive strengths, QKD, Honeypot Technology,
is imperative. This can be achieved by intro- and DKRP encounter challenges in addressing
ducing a stochastic element to the rotation the dynamic and evolving nature of advanced
interval: MitM attacks. Future research should focus
on enhancing these protocols to incorporate
T’rotation = Trotation+ Δt (5) adaptive and dynamic defense mechanisms,
ensuring a more robust resilience against the
Here, Δt is a random variable representing
intricacies of modern cyber threats.
the perturbation in the rotation interval.
Quantifying Security Gains: Quantitative
metrics are crucial for assessing the effective- 3. Methodology
ness of DKRP. The security gain (SG) achieved
through key rotation can be defined as: Chaotic maps, characterized by sensitivity to
initial conditions and aperiodic behavior, have
SG = 1/Trotation (6) found profound applications in cryptography
and secure communication systems. In this
The future resilience of DKRP lies in inno- context, we explore two prominent chaotic
vative advancements. Introducing machine maps: the Henon map and the Ikeda map.
148 Innovations in Computational Intelligence: Trends and Technologies
The Henon map is defined by: When a UAV needs to authenticate itself, it
employs the current state of its chaotic map
xn+1= 1 − a ⋅ xn2 + yn (8) as a dynamic authentication key. The recipi-
yn+1 = b ⋅ xn (9) ent UAV, possessing the same chaotic map
parameters and initial conditions, can accu-
This map introduces chaos in a two-dimen- rately predict and verify the sender’s key.
sional space, where the parameters a and b
govern the map’s behavior, creating intricate
and unpredictable patterns. 3.1. Chaotic map-based
Moving beyond, the Ikeda map is defined authentication algorithm
as: 1. Parameters Setup:
Initialize parameters for the Henon map
xn+1 = u + β⋅(xn⋅cos(t) − yn⋅sin(t)) (10) (a,b) and the Ikeda map (t).
yn+1 = u + β⋅(xn⋅sin(t) + yn⋅cos(t))(11) Choose initial conditions for
synchronization.
The Ikeda map adds an extra layer of com- 2. Synchronization:
plexity with the introduction of an auxiliary UAVs exchange initial conditions to
variable t. These chaotic maps are not merely synchronize their chaotic maps.UAVs
mathematical abstractions; they embody the exchange initial conditions to synchronize
essence of unpredictability essential for secure their chaotic maps.
cryptographic systems.
The unpredictable and intricate nature of (12)
chaotic maps makes them an ideal candidate
(13)
for fortifying authentication mechanisms in
UAV networks. Our proposed system lever- 3. Dynamic Key Generation:
ages the Henon and Ikeda chaotic maps to
enhance the security of UAV communication. (14)
The Henon map’s parameters, a and b, (15)
generate a sequence of values that serve as
dynamic keys for encrypting communication. 4. Data Encryption:
The sensitivity to initial conditions ensures (16)
that even a slight alteration in the input
results in a vastly different output, making it 5. Transmission:
resilient against predictive attacks. UAV sends the encrypted data along with
Similarly, the Ikeda map, with its depend- the current state of its chaotic maps.UAV
ence on the auxiliary variable t, introduces an sends the encrypted data along with the
additional layer of complexity. This complex- current state of its chaotic maps.
ity contributes to the generation of secure 6. Decryption and Authentication:
cryptographic keys, vital for authentication Upon receiving the data, the recipient
in UAV networks. UAV uses its synchronized chaotic maps
In our proposed system, chaotic maps play and dynamic keys for decryption.
a pivotal role in the authentication process.
(17)
During the initial setup, the UAVs synchro-
nize their chaotic maps by sharing a common (18)
initial condition or seed. This synchroniza-
tion ensures that the chaotic maps at different In the Decryption and Authentication
UAVs evolve in tandem. step, the recipient UAV employs the
Mitigating man-in-the-middle attack in UAV network using authentication mechanism 149
synchronized chaotic maps and dynamic it encrypts the data using the generated
keys to decrypt the received data. The dynamic keys. The encryption involves XOR
decrypted data is obtained by applying operations with both the Henon and Ikeda
XOR operations with the dynamic keys dynamic keys. The UAV sends the encrypted
generated from the Henon and Ikeda data along with the current state of its cha-
maps. The authenticity of the communi- otic maps to the intended recipient. Upon
cation is then verified by checking if the receiving the data, the recipient UAV uses
decrypted data matches the original data. its synchronized chaotic maps and dynamic
If the two match, the authentication is keys for decryption. The decrypted data is
deemed successful; otherwise, it is con- obtained by applying XOR operations with
sidered failed. These operations ensure the dynamic keys. An authentication check is
secure communication and effective miti- performed by comparing the decrypted data
gation against unauthorized access. with the original data. Successful authentica-
7. Security Against Man-in-the-Middle tion occurs if they match.
(MitM) Attack: Mitigation Strategy: A mitigation strategy is employed based
on the chaotic nature of the Henon and Ikeda
(19) maps. An objective function is defined to
quantify the security of the communication.
Infeasible if Objective Function>Tolerance
The objective function becomes infeasible
Attacker tries to modify Encrypted
for an attacker to predict without knowl-
Data to Encrypted DataModifiedIncon-
edge of the initial conditions, providing
sistency if Objective Function>Tolerance
security against MitM attacks. UAVs con-
after modificationAuthentication fails if
tinuously update their dynamic keys based
inconsistency detected
on the evolving states of the chaotic maps.
8. Continuous Authentication:
The synchronization process is periodically
UAVs continuously update their dynamic
repeated to account for variations in UAV
keys based on the evolving states of the
trajectories, maintaining the security of the
chaotic maps:
communication.
(20) In essence, the algorithm leverages the
unpredictability and sensitivity to initial con-
(21)
ditions of chaotic maps to generate dynamic
The synchronization process is periodi- keys, ensuring secure and continuously
cally repeated to account for variations authenticated communication between UAVs.
in UAV trajectories.The synchronization The MitM attack is mitigated by making it
process is periodically repeated to account computationally infeasible for an attacker to
for variations in UAV trajectories. predict the dynamic keys, enhancing the over-
all security of the UAV network.
UAVs exchange initial conditions to syn-
chronize their chaotic maps, ensuring they
have a common starting point for the chaotic
sequences. Each UAV employs its Henon map 4. Results and Discussions
to generate a dynamic key (Henon K Henon) Network Simulator 3 (ns-3) served as an
and its Ikeda map for Ikeda K Ikeda. instrumental tool for simulating and assessing
The dynamic keys are generated based on the proposed Chaotic Map-Based Authentica-
the evolving states of the chaotic maps, pro- tion Algorithm within a virtual environment.
viding a constantly changing cryptographic The application of ns-3 provided a robust
key. When a UAV wants to communicate, framework for modeling intricate network
150 Innovations in Computational Intelligence: Trends and Technologies
scenarios, enabling a thorough evaluation of 4.2. Delay metrics
the algorithm’s performance across diverse
Network delay, encompassing propaga-
conditions. Through ns-3, the simulation
tion, transmission, and processing delays,
results were generated, facilitating a com-
measures the time taken for data packets to
prehensive examination of crucial metrics
traverse the network. Minimizing delay is
such as packet delivery ratio, delay, and over-
crucial for ensuring prompt and responsive
head. The incorporation of ns-3 bolstered the
communication.
credibility of the research, offering valuable
The delay graph underscores CHA-
insights into the algorithm’s effectiveness in
OS’s remarkable efficiency in minimizing
securing communication among Unmanned
communication delays compared to tradi-
Aerial Vehicles (UAVs) and its resilience
tional methods like QKD, HT, and DKRP in
against potential cybersecurity threats.
Figure 18.2. This highlights the robustness
of chaotic maps in the authentication pro-
4.1. Packet delivery ratio cess, resulting in reduced delays and ensuring
PDR is a vital performance metric indicating timely data transmission in UAV networks.
the ratio of successfully transmitted packets
to the total sent. It reflects network reliability, 4.3. Overhead analysis
with a higher PDR indicating efficient data
In networking, overhead refers to additional
delivery and a lower ratio signaling potential
data or resources beyond fundamental trans-
packet loss or transmission disruptions.
mission needs, including error checking and
The graphical representation in Figure 18.1
control information. While essential, excessive
clearly showcases CHAOS’s superior Packet
overhead can lead to inefficiencies, empha-
Delivery Ratio (PDR), highlighting its effec-
sizing the need for a balanced approach for
tiveness in improving data delivery rates. This
optimal network utilization.
emphasizes the strength of chaotic maps in
In terms of overhead, Figure 18.3 illus-
authentication, presenting a more depend-
trates CHAOS’s notable advantage in mitigat-
able and efficient approach when compared to
ing unnecessary burdens on the network. The
conventional methods such as QKD, HT, and
use of chaotic maps in authentication proves
DKRP. The distinct difference in the PDR graph
to be more streamlined, offering a significant
accentuates the competitive advantage offered
reduction in overhead compared to QKD,
by chaotic authentication in the domain of
secure communication within UAV networks.
AQ1:
Please
provide
source Figure 18.1. Comparison of proposed method Figure 18.2. Comparison of proposed method
for all the with existing with respect to PDR. with existing with respect to Delay.
figures
Source: Source:
AUTHOR'S SELF GENERATED
AUTHOR'S SELF GENERATED
Mitigating man-in-the-middle attack in UAV network using authentication mechanism 151
[2] C Atheeq, M Munir Ahamed Rabbani,
“Mutually Authenticated Key Agreement
Protocol based on Chaos Theory in Integra-
tion of Internet and MANET” International
Journal of Computer Applications in Tech-
nology, Vol. 56, No. 4, pp.309–318, 2017.
[3] Keshavarz et al., “A real-time framework for
trust monitoring in a network of unmanned
aerial vehicles”, In IEEE INFOCOM 2020-
IEEE Conference on Computer Communi-
cations Workshops pp. 677–682, 2020.
Figure 18.3. Comparison of proposed method [4] Gómez Hernández, J.A. et al, “Crypto-Ran-
with existing with respect to Overhead. somware: A Revision of the State of the Art,
Advances and Challenges”, Electronics, Vol
Source: AUTHOR'S SELF GENERATED 12, no. 21, p.4494, 2023.
[5] S. Syed Abdul Syed et al., “Enhanced Chaotic
HT, and DKRP. This showcases CHAOS as Map Based Key Agreement Mitigate Packet
a promising solution for optimizing network Dropping Attack from MANETs” Interna-
resources and enhancing overall system per- tional Journal of Electronic and Communi-
formance in UAV communication. cation Engineering, Vol 10, no 7, 2023.
[6] Tsao et al., “A survey of cyber security
threats and solutions for UAV communica-
5. Conclusion tions and flying ad-hoc networks”, Ad Hoc
Networks, Vol 133, 2022.
The study presents a robust Chaotic Map- [7] C Atheeq, M Munir Ahamed Rabbani,
Based Authentication Algorithm for UAV net- “CACK—A Counter Based Authenticated
works, countering Man-in-the-Middle attacks. ACK to Mitigate Misbehaving Nodes from
Leveraging Henon and Ikeda maps, it gener- MANETs” Recent Advances in Computer
ates secure keys, encrypts data, and ensures Science and Communications Vol 14, no. 3,
continuous authentication. The dynamic key pp. 837–847, 2021.
synchronization, encrypted transmission, and [8] Pandey, Gaurav K., et al. “Security threats
meticulous decryption enhance UAV com- and mitigation techniques in uav commu-
munication security. Exploiting chaotic map nications: A comprehensive survey.” IEEE
properties strengthens resistance against Access 2022.
unauthorized access. Utilizing NS3 for sim- [9] Li, B. et al., “Anti-honeypot enabled optimal
ulations, the algorithm demonstrates high attack strategy for industrial cyber-physical
systems”, IEEE Open Journal of the Com-
PDR, minimal delay, and optimal energy con-
puter Society, Vol 1, pp. 250–261, 2020.
sumption, marking a significant advancement [10] Alshaer, N. et al., “Reliability and security
in UAV network security. Future research analysis of an entanglement-based QKD
should explore adaptability across diverse protocol in a dynamic ground-to-UAV FSO
operational landscapes. communications system”, IEEE Access, Vol
9, pp. 168052–168067, 2021.
[11] C Atheeq, M Munir Ahamed Rabbani,
References “Secure Data transmission in integrated
[1] Lateef, M.A. et al., “Data Aegis Using Che- internet MANETs based on effective trusted
byshev Chaotic Map-Based Key Authentica- knowledge algorithm” Indian Journal of Sci-
tion Protocol”, In Intelligent Manufacturing ence and Technology. Vol. 9 No. 47, 2016.
and Energy Sustainability: Proceedings of [12] Zhang, L. and Thing, V.L., “Three decades
ICIMES, pp. 187–195, 2022. of deception techniques in active cyber
152 Innovations in Computational Intelligence: Trends and Technologies
defense-retrospect and outlook”, Comput- Open Journal of the Communications Soci-
ers and Security, Vol 106, pp. 102288, 2021. ety 4, pp. 209–225, 2023.
[13] Assaf, Tasneem, et al. “High-Rate Secret Key [14] Tu, Yu-Ju, and Selwyn Piramuthu. “Security
Generation Using Physical Layer Security and privacy risks in drone-based last mile
and Physical Unclonable Functions.” IEEE delivery.” European Journal of Information
Systems, pp. 1–14, 2023.
19 A comprehensive multi-modal sentiment
analysis approach for social media
content integration
Sheela S.1,a, Suresh Balakrishnan T.2, Uttham Sing K.1, Sujitha E.1,
Soundarrajan R.1, and Veena M.1
1
Electronics and Communication Engineering, Rajalakshmi Institute of Technology, Chennai, India
2
Professor, Computer Science and Engineering, Saveetha School of Engineering, Saveetha Institute of
Technical Sciences, Chennai, India
Abstract: Sentiment analysis is a field of natural language processing (NLP) that determines the emotional tone
and attitude expressed in the content, whether it’s a text, audio, video, or social media post. This technology
has found in several domains, like marketing, social media monitoring and customer service. The goal of this
project is to develop a robust, adaptable, and real-time sentiment analysis system that can assess sentiments
and emotions across various data sources. Understanding the sentiment behind this data is essential for making
informed decisions, improving user experiences, and responding to trends in the digital age where individuals
and organizations are constantly bombarded with data in many different formats. The development of a multi-
modal sentiment analysis system that can process text, audio, social media posts, and live video streams is the
main goal of this project. This system should be able to identify subtle emotions and offer insightful informa-
tion in addition to determining basic sentiments (positive, negative, and neutral). For simple access to results, it
should be built with real-time monitoring features and an intuitive dashboard. In today’s digital environment,
sentiment analysis is crucial for various reasons.
Keywords: Audio, live video, sentiment analysis, support vector classifier, text
1. Introduction events, or people is the main objective of sen-
timent analysis on social media. Through the
In the fields of data science and NLP, senti-
analysis of the sentiment underlying these
ment analysis has grown in importance as
online exchanges, companies, academics, and
a subject of study and application. With the
policymakers can obtain important insights
growth of social media sites like Facebook,
Instagram, Twitter, and others, a vast number into public opinion and consumer feedback,
of individuals are expressing their ideas, opin- which can then be utilized to inform data-
ions, and feelings online. Sentiment analysis driven decision-making, boost customer sat-
which is also known as Opinion mining is the isfaction, improve goods and services, keep
process of using computer methods to detect an eye on brand reputation, and even fore-
and extract subjective information from vari- cast market trends. In general, there are three
ous kind of inputs like text data, posts, audio, main categories of sentiment analysis tech-
and live video including comments, reviews, niques: polarity-based analysis, which cate-
social media postings, and more. Understand- gorizes the sentiment as positive, negative, or
ing user attitudes, opinions, and emotions neutral; aspect-based analysis, which identi-
toward particular subjects, goods, services, fies particular features or aspects of a good
DOI: 10.1201/9781003606635-19
154 Innovations in Computational Intelligence: Trends and Technologies
or service and assesses the sentiment associ- reason. It can be used to analyze purchaser
ated with each feature; and emotion analysis, feedback and sentiment in consumer care,
which recognizes different emotions like hap- permitting corporations to rectify any prob-
piness, rage, sadness, and surprise expressed lems as soon as they get up and enhance con-
in the text. The casual and loud language of sumer happiness. On social media websites,
social media, which includes slang, acronyms, emotion detection may be used to tune and
misspellings, and grammatical mistakes, pre- analyze consumer sentiment. Businesses and
sents challenges for sentiment analysis [1]. companies can decide any PR problems and
This process is further complicated by con- determine how their logo or merchandise are
text-dependent feelings, irony, sarcasm, and seeming [3].
cultural context. To address these issues and
raise the precision of sentiment analysis on
social media data, researchers and practition- 2. Methodology
ers employ linear regression, random forest, The methodology involved in this multi-
support vector classifier (SVC from SVM) modal sentiment analysis system is depends
machine learning algorithms and natural on the inputs like text, audio, social media
language processing (NLP) strategies, it also post, etc.
involves usage of whisper open-AI and some
predefined modules. To summarize, sentiment
analysis on social media is an essential tool
2.1. Text as input
for comprehending public opinion and con- A technology for automatically identifying
sumer behavior in the digital age [2]. It ena- and classifying emotional content in text data
bles businesses and organizations to maintain is called text emotion detection. In this study, a
a connection with their audience, effectively comprehensive analysis of emotional content
respond to customer needs, and modify their in text samples is conducted through a sys-
strategies in response to real-time feedback, tematic process of data gathering, preprocess-
all of which positively impact decision-mak- ing, feature extraction, label encoding, model
ing and user experiences. The usage of tweets selection, training, evaluation, hyper parame-
within the form of textual content and posts ter tuning, and testing. The dataset comprises
inside the form of movies as enter to the fash- text samples annotated with emotional labels,
ions and to discover emotion in advertising ranging from joyful and depressed to furious.
and marketing can assist organizations in Rigorous preprocessing involves removing
understanding how clients are responding to special characters and stop-words, perform-
their messages. Examine students’ emotional ing tokenization, and lowercasing the text to
engagement and comprehension in textual enhance the subsequent feature extraction
content and audio while they are doing on phase. For feature extraction, advanced tech-
the line or faraway learning publications. niques such as word embedding and TF-IDF
Text emotion detection in mental fitness can vectors are leveraged to capture the semantic
be used to music patients’ emotional states meaning of the text, ensuring a rich represen-
and provide important early intervention. tation of emotional nuances [4]. The emo-
As an example, a therapist can also hire tex- tional categories are encoded with numerical
tual content emotion detection to observe an labels to facilitate the training of machine
affected person’s diary entries and notice early learning models. Various machine learn-
warning signs of depression or anxiety. In the ing models, including BERT, logistic regres-
course of criminal investigations, take a look sion, support vector machines, and recurrent
at audio and video recordings for emotional neural networks, are carefully selected and
clues that would imply dishonesty or criminal implemented to discern the most effective
A comprehensive multi-modal sentiment analysis approach for social media content integration 155
approach for sentiment analysis. Extensive The library effectively captures emotions
training on the labeled dataset optimizes the through feature extraction, extracting deep
models to predict emotions based on text features from image patches and facial land-
attributes. Evaluation metrics such as accu- marks [5]. Emotion classification is achieved
racy, F1-score, and confusion matrices are by training a deep neural network using the
employed to assess the models’ performance compiled dataset, and pre-trained emotion
on a validation dataset. Hyper parameter tun- classification models are also available within
ing further enhances the models’ capabilities, Deep face. In the inference stage, the learned
ensuring optimal results. Subsequent testing model is applied to fresh images and video
on a separate dataset validates the models’ frames for real-time emotion detection. For
real-world applicability, demonstrating their continuous video emotion analysis, Deep
effectiveness in accurately analyzing senti- face employs post-processing techniques,
ments in fresh, unexplored text data. aggregating emotion predictions across video
frames. This integrated approach ensures a
2.2. Image as input robust and efficient method for emotion rec-
ognition, making Deep face a valuable tool
The Deep face library follows a compre- for real-world applications in image and
hensive approach for emotion recognition video emotion analysis.
in images and videos, encompassing vari-
ous stages. It begins with the compilation
of a dataset comprising pictures and video 2.3. Live video as input
frames portraying emotive expressions, each To implement facial emotion recognition in
meticulously labeled. Deep face then employs live videos using a facial emotion recognition
facial detection techniques to identify faces library, the process involves real-time data
in both images and video frames, extract- collection through a webcam or other cam-
ing facial landmarks for subsequent analysis. era source, recording video frames as they
occur. Subsequently, face detection meth-
ods are applied to identify and extract faces
from each frame. The facial emotion recogni-
tion library is then employed to analyze the
extracted faces, discerning emotions such as
joy, sorrow, or rage based on facial expres-
sions [6]. Real-time processing is crucial,
AQ1:
Please
provide
in-text
citation
for figures
1 to 4.
AQ2:
Please
provide
source
for all the Figure 19.1. Text data based sentiment analysis. Figure 19.2. Image-based sentiment analysis.
figures
Source: Source:
AUTHOR'S SELF GENERATED
AUTHOR'S SELF GENERATED
156 Innovations in Computational Intelligence: Trends and Technologies
with frames continuously processed at a high networks (RNNs) and convolutional neural
frame rate to provide prompt feedback. The networks (CNNs), including long short-term
identified emotions are visually presented on memory (LSTM) layers, to process audio
the live video feed, enabling real-time emotion input effectively [7][8]. Through training and
tracking for a dynamic user experience. To supervision, the ASR system learns the intri-
enhance accuracy, thresholding is employed cate mapping from auditory properties to
to eliminate emotion predictions with low text by utilizing transcribed audio data. Ulti-
confidence levels [10]. This comprehensive mately, the generated text serves as input for
approach ensures the seamless integration of software specializing in text emotion recogni-
facial emotion recognition into live video sce- tion, facilitating the extraction of emotional
narios, offering immediate and visually intui- content from the transcribed audio [9]. This
tive insights into the emotional dynamics of integrated approach underscores Whisper’s
individuals on camera. capability not only in accurate speech-to-text
conversion but also in enabling subsequent
2.4. Audio as input emotion analysis of the transcribed content.
Whisper, developed by OpenAI, serves as
an automatic speech recognition (ASR) sys- 3. Result and Discussion
tem designed to seamlessly translate spoken The proposed model is analyzed with various
words from audio files into text. The Whis- types of inputs like audio, text, image and live
per audio-to-text conversion methodology video and detected the emotions effectively
involves several key steps. Firstly, extensive which is shown in Figures 19.5–19.8.
data collection plays a pivotal role, as Whis-
per can proficiently recognize speech in vari-
ous languages and diverse situations, owing 4. Conclusion
to its training on a substantial volume of As a result, social media presence sentiment
multilingual and multitasking supervised analysis is essential for gaining insight into
data sourced from the internet. The model how the general public feels about certain
architecture of Whisper employs a deep neu- people, companies, or subjects. Understand-
ral network that integrates recurrent neural ing the opinions of a wide range of users
Figure 19.3. Live video-based sentiment
analysis. Figure 19.4. Audio-based sentiment analysis.
Source: Source:
AUTHOR'S SELF GENERATED AUTHOR'S SELF GENERATED
A comprehensive multi-modal sentiment analysis approach for social media content integration 157
Figure 19.8. Emotion detected using Audio.
Source:
AUTHOR'S SELF GENERATED
Figure 19.5. Emotion detected using text.
Source: may be gained by analyzing user-generated
AUTHOR'S SELF GENERATED
content on social media sites like Facebook,
Instagram, and Twitter. Sentiment analysis
systems classify social media posts as good,
negative, or neutral by utilizing machine
learning and NLP techniques. This allows
corporations, people, and organizations to
assess the general sentiment and modify their
strategy appropriately. The subtleties of lan-
guage, context, and the dynamic nature of
online chats are some of the difficulties and
restrictions of sentiment analysis, though,
that should be noted. It might be difficult
Figure 19.6. Emotion detected using Image. to discern sentiment properly when consid-
Source: ering the meaning of sarcasm, irony, or cul-
tural background. Despite these obstacles, the
benefits of sentiment analysis are enormous.
It enables firms to make data-driven choices,
improve customer happiness, and manage
brand reputation. Individuals and organiza-
tions may also utilize sentiment analysis to
discover new trends, respond to consumer
feedback, and communicate with their audi-
ence more effectively. Sentiment analysis will
continue to be a vital tool for understanding
the ever-changing landscape of public opinion
as social media evolves. Integrating sentiment
analysis into social media initiatives may help
enhance communication, brand perception,
and overall digital performance. The most
significant potential for sentiment analysis in
Figure 19.7. Emotion detected using Image. social media going forward is in multimodal
Source: analysis, which combines text, visuals, and
AUTHOR'S SELF GENERATED
158 Innovations in Computational Intelligence: Trends and Technologies
audio for a more comprehensive understand- [3] Zhang, Yuan, et al, 2018. EmoReact: A
ing. To further provide accurate and nuanced Fine-Grained Emotion Detection System
insights, fine-tuning context-aware models, for Multimodal Social Media Data, North
attaining fine-grained sentiment classifica- American Chapter of the Association for
tion, and resolving ethical issues are essen- Computational Linguistics: Human Lan-
guage Technologies.
tial. The tool’s flexibility and relevance across
[4] Xu, Hu and Wang, Yuwen, 2019. BERT
many industries will be enhanced through
for Emotion Classification on Twitter, 57th
the development of important areas such as Annual Meeting of the Association for Com-
industry-specific apps, emotionally intelligent putational Linguistics.
chatbots, and real-time analysis. Therefore, [5] Zhang, Yuan, et al, 2018. EmoReact: A
we can conclude that the establishment of Fine-Grained Emotion Detection System
emotional intelligence in the future may be for Multimodal Social Media Data, North
possible and will require the automation of American Chapter of the Association for
daily tasks. One such instance is when some- Computational Linguistics: Human Lan-
one uses a smartphone while hungry, the guage Technologies.
device recognizes this and uses emotion detec- [6] Z. Aldeneh, R. Soujanya, R. Jha, and S. Jha,
tion to recommend the user’s favorite meal. 2017. Multimodal Emotion Analysis from
Naturalistic Data, Proceedings of the Inter-
national Conference on Affective Comput-
ing and Intelligent Interaction (ACII).
References [7] Wenyan Li, et al, 2018. Emotion-Cause Pair
[1] Stefano, Esuli, Andrea, and Sebastiani, Fab- Extraction: A New Task to Emotion Analy-
rizio, 2010. SentiWordNet 3.0: An Enhanced sis in Texts, North American Chapter of the
Lexical Resource for Sentiment Analysis and Association for Computational Linguistics:
Opinion Mining, Baccianella, International Human Language Technologies.
Conference on Language Resources and [8] A. Bifet, E. Frank, 2010. A Survey of Senti-
Evaluation (LREC). ment Analysis in Social Media, Published in
[2] Mohammad, Saif M., Kiritchenko, Svetlana, ACM Computing Surveys (CSUR).
and Zhu, Xiaodan, 2013. Affect Analysis in [9] B. Schuller, M. Batliner, S. Steidl, and D. Seppi,
Text: From Pattern Spotting to Rule Learn- 2017. Contextual Emotion Detection in Text
ing, Knowledge and Information Systems. Using Heterogeneous Data Sources, ACII.
20 A proposed deep learning framework
for internet-of-medical things
AQ1:
Faiyaz ahamad and Syed Hauider Abbas Please
provide
Integral University complete
affiliation
details
Abstract: In machine learning, deep learning is the most popular topic having a wide range of applications such and
as computer vision, natural language processing, speech recognition, visual object detection, disease prediction, provide
email id.
drug discovery, bioinformatics, biomedicine, etc. Of these applications, health care and medical science-related
applications are dramatically on the rise. The presentation of IoT advances, for example, those utilized in far off
wellbeing checking applications, has altered ordinary clinical consideration. Besides, the methodology used to
acquire experiences from the examination of way of life components and exercises is urgent to the progress of
custom fitted medical care and sickness avoidance administrations. The Convolutional-brain network (CNN)
system is used to conjecture such abnormality since it can effectively perceive the information important to
infection expectation from formless clinical heath records. On the other hand, in the event that a completely
coupled network-geography is utilized, CNN chugs an enormous memory. Besides, the intricacy investigation
of the model might ascend as the quantity of layers develops. Hence, we present a CNN target acknowledgment
and expectation technique in light of the Pearson-Connection Coefficient (PCC) and standard example exercises
to address these deficiencies of the CNN-model. The model’s result is separated into heftiness, hypertension,
and diabetes-related factors with known connections. To reduce the effect of the CNN-regular information dis-
closure worldview, we utilize two separate datasets. The trial results uncover that the proposed model outflanks
three other AI strategies while requiring less computational exertion. Afterward, the primary thought, benefits,
inconveniences, and constraints of each study are examined, going before ideas for further research.
Keywords: Internet-of-Medical-Things, CNN, Health Pattern Discovery, Deep Learning, Healthcare
1. Introduction admittance to really focus on patients, clinical
experts, and occupants of provincial regions.
The Internet-of-Medical-Things (IoMT) is a Reason behind care (POC) gear including
part of IoT that spotlights on medical services ultrasound machines, thermometers, glucose
checking using organized clinical hardware. screens, and electrocardiogram (ECG) perus-
IoMT innovation, otherwise called medical ers are presently accessible with online avail-
services IoT, disposes of the requirement for ability and distributed storage choices for
human mediation in medical services observ- client information [3,4].
ing using advanced mechanics, biosensors, The CNN procedure is regularly used for
and artificial intelligence controlled by AI [1]. design mining from organized and unstruc-
Advanced medical care puts beforehand inac- tured information because of profound learn-
cessible computerized recognition advance- ing (DL) techniques. It is additionally, usually
ments and coordinated cloud products in used to address complex models and investi-
the possession of buyers, in a general sense gate unstructured clinical information. The
modifying the idea of customary medical CNN model has critical memory and register
care observing [2]. This shift to computerized needs, and it can experience the ill effects of the
medical care conveyance further develops over-fitting issue. Important bodily function
DOI: 10.1201/9781003606635-20
160 Innovations in Computational Intelligence: Trends and Technologies
information portraying the day to day rou- approaches that safeguard clients’ secrecy.
tines of people has extended at a remarkable The creators of [14], for example, explored
rate with the presentation of the IoMT cli- this limitation and offered a system for power
mate [5,6]. At the point when a patient shows guideline in uplink FL-IoMT to address it.
up with a tenacious disease, he will show an The plan further develops productivity by
enhance in various wellbeing boundaries (like recognizing the problematic jammer robot.
a high temperature) all the while because of Hence, creators introduced an asset effective
the recognizable routineness of human well- security-protection way to deal with address
being parts. He may likewise see a decrease this issue. The creators have used rules like
in another sign of his wellbeing, for example, postponement and cost to check their work.
pulse. Along these lines, it is feasible to more Trust, unchanging nature, information acces-
deeply study the administration and examina- sibility, and security are completely ensured
tion of wellbeing related markers by analyz- by the plan. To address the hardships of incor-
ing the examples of conduct uncovered by the porated instruction, proposed an answer. This
gathered information [7,8]. In section 2, we’ll contextual analysis was likewise compelled
go over the various prior techniques exhaus- to manage the possibility of differential pro-
tively. Procedure is illustrated in Section 3. In tection. The elite presentation was assessed
Section4, we contrast the aftereffects of the inside the IoMT not set in stone to be viable
proposed work to those of the ordinary meth- for use [15].
odologies. The work process’ definitive deci-
sion is depicted in section 5.
3. Proposed Methodology
The proposed work makes use of a CNN-
2. Related Works
model to uncover regularly associated health
A few articles on IoT, profound learning indicators that may have an impact on health.
organizations, and patient observing infor- The methodology starts with the assortment
mation investigation are examined here. and fundamental handling of patient clini-
This study gave a methodology to protecting cal information. Corpulence, hypertension,
patient information put away across differ- and diabetes are only a couple of instances
ent waiters from exploitable assaults [9,10]. of the sorts of issues that can be connected
In creators discussed how combined medi- utilizing the separated information. Factors
cal care arrangements present open doors for with positive and negative relationships are
unified learning (FL) structures. The creators recognized, similar to the useful connec-
discussed how FL can be utilized related to tions between’s them. Eventually, we pick the
imaging for the newfound 2019 Covid sick- often co-happening elements to extricate the
ness (Coronavirus). Their application to EHR important data. The accompanying parts will
sickness conclusion and result forecast was go into more noteworthy profundity, itemiz-
assessed by Antumes et al. [11]. In [12], sci- ing each stage.
entists gave an extensive survey of FL-related
strategies for use in clinical imaging that is 3.1. Activity of disease and symptom
both secure and private. Potential purposes,
for example, digital assault discovery and idea
proposal frameworks are talked about. We The model uses multivariate analysis to iden-
present a decentralized and circulated FL tify relevant features among the collected uni-
framework [13]. In IoMT biological systems, structural medical health situations, and then
scholastics have read up safety efforts for uses those factors to categorize the data into
the actual layer, and have created FL-based input and output elements. We preprocess the
A proposed deep learning framework for internet-of-medical things 161
If the PCF of two variables F1 and F2 is
more than 0.1, then the variables are posi-
tively connected. Cov(F1; F2) defines the set
of functions F1 and F2 that are covered. Mean
and standard deviations for each parameter
F1, F2 are denoted by F1 and F2, respectively.
3.3. CNN-supported health
AQ2:
Please knowledge framework
provide
in-text In light of the previously mentioned, we pro-
citation posed another model for distinguishing regu-
for Figure
1&2. larly associated wellbeing related boundaries
that might be utilized to recognize any devia-
AQ3:
Please
tion from the standard. To create the verifi-
provide able data related with remote observing and
source the executives of the human way of life, this
for all the Figure 20.1. The proposed model diagram.
figures approach can mine all the open standard com-
Source: AUTHOR'S SELF GENERATED ponent conduct [17]. Utilizing the relation-
ship consequences of the wellbeing ascribes,
data to get rid of things like no response and
we desire to make it workable for care sup-
missing value for the chosen 23 attributes.
pliers to offer constant remote-checking of
Our analysis and extraction of the important
patient healthstatus. In this way, the clinical
elements from the dataset of 187,473 records
exertion would be effectively diminished, and
helps a remote monitoring application give
a solid CNN-based finding emotionally sup-
the correct therapy at the right time for the
portive network could be given through the
right person.
course of all the while exploring the relation-
ship between different side effects. To addi-
3.2. Data categorisation by tionally order the found connections in the
correlation coefficient secret layer, the CNN model is utilized. The
Pearson’s correlation coefficient (PCF) is standard way of behaving of the associated
utilized to distinguish huge relationship wellbeing factors is likewise separated, which
between the factors in this examination. Suc- uncovers beforehand inconspicuous custom-
cessfully concentrating on ordinary relation- ary corresponded readings [18].
ship conduct among the picked wellbeing We found specific routineness related well-
determinants, PCF can show the strength being states in view of the EHR information
of relationship between two factors. At the we assembled. In light of the anticipated idea
point when the importance level of a thing of certain sicknesses, following the develop-
is bigger than 0.1, having useful, independent ment of a subset of wellbeing pointers over
characteristic is thought of. Thusly, we keep a given timeframe can yield valuable data
away from overfitting issues and assurance about an individual’s wellbeing. In this man-
that the chose thing is associated. By tack- ner, quite possibly of the most squeezing chal-
ling for the PCF in condition (1), we might lenge in wellbeing information examination
decide how unequivocally credits F1 and F2 is the need to research the way of behaving
are connected. of customary variables. On the off chance
that sickness X causes the patient’s tempera-
(1) ture and heart-rate to spike 3times in month,
162 Innovations in Computational Intelligence: Trends and Technologies
monitoring and understanding this strange Lesser RMSE values indicate that the meth-
occurrence might assist with deflecting a car- odology can make precise expectations, while
diovascular failure [19,20]. upper RMSE values determine that the model
can’t. Assuming the extended incentive for
weight is 3, and the genuine worth is 2, then
4. Results and Discussions the mistake for every one of the three sick-
The testing was finished on a 64-digit Corei5 nesses is 1. The CNN-model’s RMSE discov-
Windows 10pro with 12GB of accessible eries are displayed in Table 20.1. Our model’s
Slam involving SPSS for execution examina- RMSE for foreseeing corpulence, diabetes, and
tion. Wellbeing and lifestyle-related setting hypertension is contrasted with that of a non-
information are the concentration here. These exclusive CNN model. The proposed model’s
information come from 10,806 residents who computation speed and intricacy are both
went through wellbeing actually looks at in determined with 2-stowed away layers. Low
the present time and place. Each occupant is expectation exactness close by high estima-
approached to finish a wellbeing study com- tion speed and intricacy describes the regular
prising of 768 inquiries. By utilizing mul- CNN-model for every sickness analysis. For
tivariate investigation, SPSS can effectively reference, the typical CNN model has a RMSE
lessen how much components without losing of around 0.87 and an intricacy of 1.1.The
any essential qualities and work on the calcu- proposed model can foresee the presence of
lation expected to mine the important well- diabetes with a greatest RMSE of 0.2562.
being related angles. Various variables that Moreover, the accompanying nonexclusive
impact stoutness, hypertension, and diabetes models were prepared, and their outcomes
are examined for use in the proficiency exam- are introduced in Table 20.2 and Figure AQ4:
ination of the model. We utilized 4,759,777 20.3 for correlation with our calculation’s Please
note that
records from the chose setting information, presentation. no figure
of which 1,499,423 records had no reaction 3.
and missing qualities were gotten out before
the preliminaries were run. We split the infor- 5. Conclusion
mation we obtain into two sets: train infor- This paper presents an ordinary example dig-
mation (70%) and test information (30%). ging procedure for wellbeing information in
The hyper boundaries (such preparation
end) were streamlined utilizing a 10-over- Table 20.1. Execution evaluation of the
lap cross approval procedure. The precision proposed plan of action
of the CNN model is likewise tried using a Presence RMSE Calculation Complexity
measurement called the root-mean-square- Speed
blunder (RMSE). The proposed CNN-based
Obesity 0.092 14 0.562
wellbeing approach is applied to each model AQ5:
to decide the RMSE, estimation time, and BP 0.181 16 0.571 Please
provide
intricacy. For the expectation of heftiness, Diabetics 0.264 15 0.624 source
diabetes, and hypertension, we additionally Source: AUTHOR'S SELF GENERATED
for all the
Tables
record a similar examination rule in the more
extensive CNN. While contrasting an antici- Table 20.2. Execution comparison
pated worth with a noticed worth, RMSE is
Presence LSTM SVM ANN Proposed
a valuable measurement to use. Condition (2)
can be utilized to decide the amount RMSE Obesity 87 76 74 92
is worth. BP 91 92 76 94
Diabetics 90 64 65 96
RMSE = (2)
Source:
AUTHOR'S SELF GENERATED
A proposed deep learning framework for internet-of-medical things 163
[2] AKM Iqtidar Newaz, Amit Kumar Sikder,
Mohammad Ashiqur Rah man, and A Selcuk
Uluagac. Healthguard: A machine learning-
based security framework for smart health-
care systems. In 2019 Sixth International
Conference on Social Networks Analysis,
Management and Security (SNAMS), pages
389–396, 2019.
[3] Valentina Bianchi, Marco Bassoli, Gian-
franco Lombardo, Paolo Fornac ciari,
Monica Mordonini, and Ilaria De Munari.
Iot wearable sensor and deep learning: An
integrated approach for personalized human
activity recognition in a smart home envi-
Figure 20.2. Comparison with existing ronment. IEEE Internet of Things Journal,
frameworks. 6(5):8553–8562, 2019.
[4] Syed Umar Amin, M Shamim Hossain, Ghu-
Source:
AUTHOR'S SELF GENERATED lam Muhammad, Musaed Alhussein, and
Md Abdur Rahman. Cognitive smart health-
light of convolutional brain organizations. care for pathology detection and monitor-
IoT-gadget information on medical issues and ing. IEEE Access, 7:10745–10753, 2019.
way of life components applicable to constant [5] Zubair Md Fadlullah, Al-Sakib Khan
illnesses frames the premise of the recom- Pathan, and Haris Gacanin. On delay-sensi-
mended approach. Information determina- tive healthcare data analytics at the network
edge based on deep learning. In 2018 14th
tion and grouping are performed utilizing a
International Wireless Communications and
twofold layer completely connected CNN
Mobile Computing Conference (IWCMC),
structure. Multivariate investigation is used pages 388–393, 2018.
to recognize significant wellbeing factors, [6] Shanto Roy, Abdur Rahman, Masuk Helal,
and a resulting layer orders those variables. M Shamim Kaiser, and Zamshed Iqbal
Finally, the most vital frequently happening Chowdhury. Low cost rf based online patient
wellbeing highlights are picked in light of an monitoring using web and mobile applica-
examination of the consistency of the distin- tions. In 2016 5th International Confer-
guished parts. The model might recognize ence on Informatics, Electronics and Vision
both positive and negative customary connec- (ICIEV), pages 869–874, 2016.
tions, the previous of which can be utilized to [7] Kamruzzaman, M.M., Alrashdi, I. and
support medical services while the last option Alqazzaz, A., 2022. New opportunities,
can reveal insight into how to work on unde- challenges, and applications of edge-AI for
connected healthcare in internet of medical
sirable schedules. Ultimately, we have exam-
things for smart cities. Journal of Healthcare
ined the difficulties in ongoing bearings and
Engineering, 2022.
inquiries without a right or wrong answer [8] Manickam, P., Mariappan, S.A., Murugesan,
that still should be tended to in this area and S.M., Hansda, S., Kaushik, A., Shinde, R.
close with the outline of our audit. and Thipperudraswamy, S.P., 2022. Artifi-
cial intelligence (AI) and internet of medical
things (IoMT) assisted biomedical systems
References for intelligent healthcare. Biosensors, 12(8),
[1] Almaiah, M.A., Ali, A., Hajjej, F., Pasha, p.562.
M.F. and Alohali, M.A., 2022. A lightweight [9] Hossen, M.N., Panneerselvam, V., Koundal,
hybrid deep learning privacy preserving D., Ahmed, K., Bui, F.M. and Ibrahim, S.M.,
model for FC-based industrial internet of 2022. Federated machine learning for detec-
medical things. Sensors, 22(6), p.2112. tion of skin diseases and enhancement of
164 Innovations in Computational Intelligence: Trends and Technologies
internet of medical things (IoMT) security. [15] K. Lokesh, S. Srivastava, M. P. Kumar, S.
IEEE journal of biomedical and health Arun, S. Padmapriya and R. Krishnamoor-
informatics, 27(2), pp.835–841. thy, “Detection of Stomach Cancer Using
[10] Rawat, R., 2022. a Systematic Review of Deep Neural Network in Healthcare Sec-
Blockchain Technology Use in E-Supply Chain tor,” 2021 3rd International Conference
in Internet of Medical Things (Iomt). Interna- on Advances in Computing, Communica-
tional Journal of Computations, Information tion Control and Networking (ICAC3N),
and Manufacturing (IJCIM), 2(2). 2021, pp. 521–526, doi: 10.1109/
[11] Chaganti, R., Mourade, A., Ravi, V., Vem- ICAC3N53548.2021.9725656.
prala, N., Dua, A. and Bhushan, B., 2022. A [16] Kagita, M.K., Thilakarathne, N., Gadekallu,
particle swarm optimization and deep learn- T.R. and Maddikunta, P.K.R., 2022. A
ing approach for intrusion detection system review on security and privacy of internet of
in internet of medical things. Sustainability, medical things. Intelligent internet of things
14(19), p.12828. for healthcare and industry, pp.171–187.
[12] A. K. Maurya, K. Lokesh, Sandeep, R. [17] Moqurrab, S.A., Tariq, N., Anjum, A., Asher-
Kumar, S. Arun and R. Krishnamoorthy, alieva, A., Malik, S.U., Malik, H., Pervaiz, H.
“Deep Neuro-Fuzzy Logic Technique for and Gill, S.S., 2022. A deep learning-based
Brain Meningiomasa Prediction,” 2022 privacy-preserving model for smart health-
7th International Conference on Commu- care in Internet of medical things using fog
nication and Electronics Systems (ICCES), computing. Wireless Personal Communica-
2022, pp. 1244–1248, doi: 10.1109/ tions, 126(3), pp.2379–2401.
ICCES54183.2022.9836008. [18] Razdan, S. and Sharma, S., 2022. Internet of
[13] G. N. Vivekananda, A. R. H. Ali, S. Arun, medical things (IoMT): Overview, emerging
P. Mishra, R. Sengar and R. Krishnamoor- technologies, and case studies. IETE techni-
thy, “Cloud Based Effective Health Care cal review, 39(4), pp.775–788.
Management System With Artificial Intel- [19] Awad, A.I., Fouda, M.M., Khashaba, M.M.,
ligence,” 2022 IEEE 7th International Mohamed, E.R. and Hosny, K.M., 2022.
conference for Convergence in Technol- Utilization of mobile edge computing on the
ogy (I2CT), 2022, pp. 1–6, doi: 10.1109/ Internet of Medical Things: A survey. ICT
I2CT54291.2022.9825457. Express.
[14] Band, S.S., Ardabili, S., Yarahmadi, A., Pahl- [20] Michelle Crouthamel, Emilia Quattrocchi,
evanzadeh, B., Kiani, A.K., Beheshti, A., Sarah Watts, Sherry Wang, Pamela Berry,
Alinejad-Rokny, H., Dehzangi, I., Chang, Luis Garcia-Gancedo, Valentin Hamy, and
A., Mosavi, A. and Moslehpour, M., 2022. Rachel E Williams. Using a researchkit
A survey on machine learning and internet smartphone app to collect rheumatoid
of medical things-based approaches for han- arthritis symptoms from real-world partici-
dling COVID-19: Meta-analysis. Frontiers pants: feasibility study. JMIR mHealth and
in Public Health, 10, p.869238. uHealth, 6(9):e177, 2018.
21 Analysis of novel routing protocol
to minimize delay in VANET: A
comprehensive simulation for VANET
Kamlesh Kumara and Bobby Sharmab
Department of Computer Science and Engineering, Assam Don Bosco University, Guwahati, India
Abstract: Global vehicle population is growing steadily, which makes it ever more harder to manage traffic. To
resolve this problem, advance traffic control techniques is developed to which V2V (Vehicle-to-Vehicle) com-
munication via VANETs (Vehicular Ad-Hoc Networks) is proven as an efficient solution. VANET allows vehi-
cles to communicate within the network by using different routing protocols (Graph based, Broadcast based,
Multicast/Geographic and Group-based routing protocol). Given the fact of the importance of VANET rout-
ing protocols in intelligent transportation systems, a need for their evaluation and selection arises. This paper
concludes that the increasing traffic leading to greater complexity of global traffic management and stresses on
VANET being the most suitable technology for V2V communication. This paper presents findings related to
VANET services, and routing protocols, with an especial concentration on the Application-Oriented Dynamic
Vehicle (AODV) method applied for delay reduction. It also gives a comparative study of routing protocols
specifying the advantages and disadvantage each has in detail.
Keywords: VANET, Cluster Head, Intelligent Transportation System, AODV, RSU
1. Introduction that addresses key aspects of VANET simu-
lation such as node initialization, energy-
Vehicular Ad-Hoc Networks (VANETs) epit-
aware path discovery, and end-to-end time
omize a frontier paradigm in the world of
(E2E) estimation through a careful simula-
intelligent transportation systems, offering
tion procedure. Furthermore, the integration
innovative evolution in road safety enhance-
ment, traffic management and communica- of the Vehicular Energy-Efficient Pathfinding
tion between vehicles. Then static MANETs, (VEEP) algorithm, which combines TEEN
VANETs are more dynamic as it consists of and A* principles, represents a novel frame-
a large number of vehicles which continu- work aimed at optimizing energy consump-
ously change their position on the network tion and communication paths in VANETs.
(i.e., mobility) and join or leave the route fre- The aim of this research work is to enhance
quently, so it presents many types of problems the knowledge in the field of VANET by
that are generally unsolved and due to non- providing a systematic novel algorithm. The
availability of efficient solutions. In this paper, proposed algorithm addressed several issues
we introduce an all-around VANET simula- related to communication reliability, secu-
tion algorithm that mainly applies to network rity, scalability, and mobility management for
parameters and energy management as well improving VANET performance. The follow-
as communication efficiency. The proposed ing sections discussed the literature review,
algorithm includes a multi-layered approach issues and challenges, proposed algorithm,
[email protected]; [email protected]
a
DOI: 10.1201/9781003606635-21
166 Innovations in Computational Intelligence: Trends and Technologies
simulation results, and future research scope routes based on real-time traffic conditions.
in automotive ad-hoc networks. ATAR uses operational data to find optimal
routes, thereby improving communication
reliability and efficiency. It reduces latency
2. Routing Protocol in VANET and congestion and increases overall net-
In order to facilitate efficient communica- work performance in ever-changing vehicular
tion between vehicles in dynamic and rapidly environments.
changing environments, the Routing Proto- Routing protocols in VANET play a very
cols in Vehicular AdHoc Networks are very important and crucial role for communica-
important. tion. It addresses challenges such as frequent
Some important protocols are: topology changes and intermittent connec-
Geographic Routing Protocols: This pro- tivity. The selection of protocol depends on
tocol uses geographic information for rout- application requirements and the desired
ing decisions, e.g., Greedy Perimeter Stateless trade-off between communication efficiency
Routing (GPSR). It is very simple and efficient and overhead. In these dynamic conditions,
in stable network conditions. researchers continue to work on new strate-
Topological Routing Protocols: Topologi- gies for improving the resilience of VANET
cal Routing Protocols uses network topology and its performance.
information, e.g., Ad-Hoc On-Demand Dis- Issues and Challenges: Vehicular ad-hoc
tance Vector (AODV) and Dynamic Source networks (VANETs) face various kinds of
Routing (DSR). It is suitable for dynamic envi- challenges that require special attention for
ronments with on-demand route discovery. successful use. Ensuring communication reli-
Cluster-Based Routing Protocols: It com- ability in the dynamic nature of the environ-
bine vehicles into clusters, each cluster is ment is a major concern, as vehicle movement
managed by Cluster Head(CH), e.g., Cluster- and changing networks can cause problems.
Based Routing Protocol (CBRP). CHs man- It is important to keep your communications
age It Reduces communication overhead secure and private. A robust encryption solu-
within clusters. intra-cluster communica- tion is required to combat potential threats
tion and inter-cluster data routing, improv- and data breaches. Another problem with
ing scalability, reducing network congestion, VANETs is scalability, as the increasing num-
and optimizing data transmission in highly ber of vehicles constantly connecting in the
dynamic vehicular environment. network can lead to congestion and increased
Position-Based Routing Protocols: Basi- communication overhead. Maintaining a
cally it use the vehicle location and mobility good quality of service is very difficult due
information, e.g., Location-Aided Routing to changing traffic and busy networks, so it
(LAR). It is effective in highly mobile scenar- needs smart and efficient ways to manage it
ios, reducing route discovery latency. all. Interoperability between different com-
Vehicular Multi-Hop Clustering (VMHC): munication technologies and standards is
VMHC organizes vehicles into clusters for also a big problem, which requires seamless
efficient communication. Each cluster has a cooperation between vehicles from differ-
Cluster Head. Data is relayed through mul- ent manufacturers. In VANETs, maintain-
tiple hops between clusters. This method ing Quality of Service (QoS) is a challenge
improves data delivery, reduces congestion, due to changing traffic conditions and net-
and adapts well to the dynamic nature of work congestion, requiring protocols that
vehicular networks. adapt to different applications. Due to the
Adaptive Traffic-Aware Routing (ATAR): complicated regulatory and standardiza-
ATAR in VANETs dynamically adjusts tion issues of various governments, it needs
Analysis of novel routing protocol to minimize delay in VANET: A comprehensive simulation for VANET 167
globally uniform regulations and standards. K-Means, named as Continuous Hopfield
The time has come to make the road systems Network, and the Maximum Stable Set
even more complicated rules and standards, Problem (KMRP) which focused on VANET-
but they are important for VANETs to work enabled applications. The KMRP protocol
well everywhere. Addressing these challenges demonstrates superior performance in high-
requires teamwork between researchers, way traffic scenarios, enhancing throughput
businesses and policy makers. We need new and Packet Delivery Ratio (PDR). KMRP’s
ideas for better communication, safe systems ability to alleviate traffic congestion contrib-
and ways to control lots of vehicles. Creat- utes to improved network performance.
ing rules that everyone will follow is essential Sindhwani et al. (2022) addressed rout-
for VANETs to be used worldwide. As cars ing problem in VANETs using AODV and
change, we need to address these challenges Ant Colony Optimization (ACO). They used
so that VANETs improve safety and make multicast to build efficient routes, minimize
driving easier for everyone. congestion, and increase throughput and
compared with existing protocols, the pro-
posed method achieves shorter paths, higher
3. Literature Review throughput, less packet loss and delay.
Al-Shaibany (2021) proposed an algorithm to Srivastava et al. (2021) focused to address
studies efficient transmission of information, secure routing challenges in VANETs, com-
especially related to accidents, in vehicular ad paring position-based and cluster-based
hoc networks (VANETs). The proposed Clus- routing protocols. Cluster-based routing, par-
ter-Based Routing with Dynamic Stability ticularly in selecting Cluster Heads, is deemed
(CRDS) algorithms introduces a novel clus- more practical and reliable.
tering technique that considers both cluster Singh et al. (2022) proposes a hybrid
head and gateway stability which is focus on routing protocol based on Firefly genetic
focused on addressing the challenges posed algorithm (HGFA)which address on the vehi-
by unprecedented vehicular mobility, which cle-to-vehicle (V2V) communication problem
impacts network connectivity and route sta- in VANETs. It integrates a genetic algorithm
bility. CRDS outperforms existing protocols (GA) into the Firefly routing algorithm and
in terms of network overhead, end-to-end outperforms traditional protocols such as
delay, route stability, cluster head stability, AODV, OLSR, and DSDV in both sparse and
and packet delivery speed. The proposed pro- dense network settings.
tocol relies on a unique optimization model Sabbagh and Shcherbakov (2021) pro-
within the clustering technique to calculate posed protocol combines k-means clustering
optimal routes. Simulations under various and cuckoo search, considering factors like
scenarios demonstrate the superiority of node speed and trust, resulting in improved
CRDS over LRCA and PASRP. network metrics under malicious assaults.
Kazi and Khan (2021) proposed a study It focuses on the importance of routing in
Dynamic Tri Lateral Enrollment (DyTE), a Vehicular Ad-Hoc Networks (VANETs) for
routing system that improves communications enhancing road safety.
with fast moving vehicles. They performed Divya et al. (2021) presented a clustered
comparative analysis with AODV, AOMDV, vehicle location for VANETs (CVL-HKH-BO)
and DSR shows that DyTE provides significant approach which is leveraging krill swarm and
improvements in network routing load (NRL), bat hybrid optimization to improve safety
packet delivery ratio (PDR), and throughput. and address challenges related to energy con-
Kandali et al. (2021) Proposed a novel sumption and packet delay. Simulations show
clustering-based routing protocol using improved network performance in terms of
168 Innovations in Computational Intelligence: Trends and Technologies
throughput, packet loss, delay time, and Input:
transmission rate.
Mershad (2020) proposed an Internet of • N: Width of the grid.
Vehicles (IoV) packet routing protocol called • P: Number of lanes.
SUPPER, which is utilizing software-defined • Zones: List of takeoff and arrival points
networking (SDN) and blockchain within for each zone.
the RSU network. Comparative simulations • Museum Cells: Cells within the museum,
reveal SURFER’s superiority in terms of which can be selected as waypoints.
latency, packet delivery, and network over- • Unit Time: Time required for movement
head over other IoT routing protocols. of each unit.
Suganthi and Ramamoorthy (2020) focus
on improving the quality of service in VANETs Output:
through a fitness-based Ad hoc on-Demand
• Shortest Paths: Set of P shortest paths
Distance Vector (FBAODV) routing proto-
without conflicts.
col. The proposed protocol includes received
signal strength indication and outperforms • P(t): Is is time domain for artificial intel-
ligence analysis, where P(t) = Mean(z).
traditional AODV in terms of energy con-
• P(t + δt): Time domain updated at time t
sumption, packet delivery, throughput, and
+ δt, where P(t + δt) = P(t) + R-1(P) * ▼p
delay.
* p(t + δt).
Karthikeyan et al. (2020) addresses the
security of VANETs and propose a security-
based cryptosystem (SIDBC). This approach 5. Shortest Path Calculation
increases security by generating pseudony- Process with Real-Time
mous keys for each node using polynomial
key generation to protect against misleading Information Model
information transmission. 1. Criteria for Shortest Path: Calculate the
Bhati and Singh (2020) examine the use shortest path considering multiple cri-
of VANETs in Intelligent Transportation Sys- teria, such as distance, traffic, and road
tems (ITS), focusing on the importance of conditions.
routing protocols. The study provides insight 2. Starting and Endpoint: Define the starting
into group routing protocols and their appli- and endpoint locations for the analysis.
cations for solving road-related problems and 3. Fastest Route Calculation: Find the fast-
focuses on the roles of VANETs in intelligent est route that optimizes the given criteria.
traffic management. This may involve dynamic adjustments
based on real-time data.
4. Utilize Real-Time Information: Incorpo-
4. Mathematical Analysis of Route
rate real-time information to access rel-
Map in VANET evant data quickly, such as traffic updates
Objective: and road closures.
• The first objective is to conduct mathemat- 5.1. Algorithms for shortest path
ical analysis of route maps in a VANET. calculation: A* (A-Star)
• Second objective is to investigate and ana-
lyze time delays in VANET nodes. algorithm
• Then integrate the Grid method with a A* algorithm is one of the most widely used
real-time model for the computation of pathfinding algorithms for finding the short-
shortest paths. est path in the graph or grid from a starting
Analysis of novel routing protocol to minimize delay in VANET: A comprehensive simulation for VANET 169
point to a goal. The algorithm maintains two data increases the accuracy and relevance of
sets: an open one and a closed one. The open route calculations. The algorithm contributes
set contains the nodes to be evaluated. These to better routing and navigation within the
nodes are ranked based on a combination of VANET, optimizing travel time and efficiency.
starting point to node cost (g), and heuristic
approximation from goal to goal (h). Nodes 5.3. Proposed algorithm
that have already been evaluated are tracked
Input: In the proposed algorithm, we previ-
in the closed one. The method iteratively
ously defined input parameters such as the
selects the node that has the lowest total total
width of the grid (N) and the number of
cost from the open set. Then, it evaluates its
lanes (P).
neighbors and changes its parent pointer and
Output: The output included a set of short-
costs until the goal is achieved or the open
est paths (Shortest Paths) and a time domain
is empty. To reconstruct the path from start
for artificial intelligence analysis (P(t)).
to target, the process traces back via parent
pointers once the goal is achieved. The algo- 1. Input Parameters: The s input parameters
rithm A* is broadly used in in large number related to the number of VANET nodes
of applications such as robots, games, and (20, 25, or 30). This parameter represents
vehicle networks. the size of the VANET network.
2. Creation of Sink Node: The creation of a
5.2. TEEN (Threshold Sensitive “Sink node” can be related to the concept
of starting and endpoint locations in the
Energy Efficient Sensor Network
algorithm. The sink node could represent
Protocol) a destination point.
A specific algorithm used in passive networks 3. Zero Initial Delay: The algorithm defines
to improve energy efficiency is the Threshold an initial delay (Td) as zero, which can
Sensitive Energy Sensor Network Protocol be related to the initial conditions in the
(TEEN). TEEN is primarily used for managing VANET simulation. In the algorithm, we
collaborative sensing applications and differs analysed time delays (Time) in the network.
from traditional sensor network protocols 4. Energy-Related Parameters: The algo-
in its ability to evolve clusters dynamically. rithms describe energy-related parameters
Cluster heads strategically allocate members such as Einrg (starting energy) and Eni
within each cluster head (CH) interval based (initial energy of individual nodes). While
on the characteristics of each member. not particularly specified in the approach,
Depending on the characteristics of the energy consumption and management are
individual members, this division takes crucial components of VANET analysis.
place in a rigid or flexible manner. The goal 5. Trace File Estimation: It seems to be simu-
of TEEN is to achieve an optimal balance lating VANET nodes and their activities,
between energy consumption in the network and it references the estimation of “E2E”
by adapting nodes according to their charac- (End-to-End) from a trace file. This esti-
teristics. This approach makes TEEN well- mation can be related to the calculation of
suited for scenarios where energy efficiency E2E time in the method.
is a critical consideration, such as sensor net- 6. Simulation stages: The outlines simula-
works, where preserving battery life is essen- tion steps involving the formation of
tial for extended and efficient operation. nodes (Ni), random distance assignments
A* and TEEN algorithms are used for (Di), and energy initialization (Eni).
efficient path finding algorithm. The inte- These phases are prevalent in VANET
gration of the Grid method and real-time simulations.
170 Innovations in Computational Intelligence: Trends and Technologies
7. Calculation of Tc and Td: It calculates into the A* rating process, considering car
Tc (communication time) and Td (total features and energy-aware pricing.
delay). These calculations can be related 5. Energy-Aware Cost Calculation: Integrate
to the time domain and time delay com- TEEN’s energy-aware ideas into the cost
putations in the VANET algorithm. calculation of A* for VANET. Adjust the
8. Simulation Execution: It can be executed cost function to balance the requirement
a random configuration and estimates the for the shortest communication path with
cumulative communication time (ΣTc). energy efficiency across vehicles.
This is comparable to the process of dis- 6. Dynamic Cluster Adaptation: During the
covering the shortest paths and calculat- pathfinding phase, explore the dynamic
ing the total time in the approach. evolution of vehicle clusters in VEEP
driven by TEEN. Optimize the commu-
5.4. Integrated proposed algorithm: nication line to pass through energy-effi-
vehicular energy-efficient cient clusters, adjusting the route based
on TEEN’s cluster distribution.
pathfinding (VEEP)
1. Adaptive Thresholds in VANET:
Vehicular energy-efficient pathfinding algo- Leverage TEEN’s adaptive thresholds
rithm (VEEP) Integrated is a proposed algo- to influence the decision-making pro-
rithm for vehicular ad-hoc network (VANet) cess in A* within VANET.
by aligning TEEN with A* in VANET context Adjust the thresholds dynamically
where energy-efficient communication and to respond to changing energy condi-
optimized pathfinding are necessary, a new tions and communication quality in
VEEP algorithm can be conceptualized. the vehicular network.
2. Optimal Balance for VANET:
5.5. Vehicular energy-efficient Aim for an optimal balance between
pathfinding finding the shortest communication
path and preserving energy among
1. Initialization: Create a VANET using vehicles in the VANET.
TEEN principles to optimize energy use 3. Evaluation and Refinement for
across cars. Initialize an open set and a VANET:
closed pathfinding set with A*. Evaluate the performance of the inte-
2. TEEN-Inspired Grouping: Enable VEEP grated VEEP algorithm in VANET
to dynamically construct vehicle group- through simulations or real-world tests.
ings based on energy efficiency and Refine the algorithm iteratively based
inspired by TEEN. Carefully designate on the results, considering trade-offs
cluster leaders to drive energy-aware between communication efficiency and
communication. energy consumption among vehicles.
3. Vehicle Features: Use vehicle features
from TEEN, such as energy levels and This algorithm related to the setup and
communication capabilities, to change the execution of a VANET simulation, particu-
A* heuristic function. Modify the A* heu- larly focusing on network parameters, energy,
ristic to consider energy-efficient paths and communication time.
within VANETs.
4. Path finding using A in VANET: * Imple-
ment the A* method to determine the ideal
5.6. VANET Simulation
communication path in a VANET graph. Input:
Incorporate TEEN-derived information • Number of VANET Nodes: {20, 25, 30}
Analysis of novel routing protocol to minimize delay in VANET: A comprehensive simulation for VANET 171
• Creation of Sink Node: Representing a a VANET network, such as node startup, com-
destination node munication time calculations, and estimating
• Zero Initial Delay: Td = 0 End-to-End Time. Basically the simulation is
• Initial Energy (Einrg): 100 designed to simulate network dynamics and
• Node IDs (Ni): {n1, n2, n3, ...} analyse communication performance in a
• Random Distance Assignments (Di): {d1, mathematical environment.
d2, d3, ...}
• Initial Energy of Individual Nodes (Eni): 5.7. Simulation result
{100 * n1, 100 * n2, 100 * n3, ...}
Output: The application of the proposed VANET
• Estimation of End-to-End Time (E2E) simulation algorithm, particularly the Vehic-
from a trace file ular Energy-Efficient Pathfinding (VEEP)
Simulation Steps: approach, yields significant insights into
1. Network Setup: Establish a VANET net- the performance of Vehicular Ad-Hoc Net-
work with a predetermined number of works (VANETs). The simulation results
nodes (20, 25, or 30). provide a comprehensive understanding of
2. Set the initial delay (Td) to zero. Set the network behavior and communication effi-
energy parameter Einrg to 100. ciency within the context of energy-aware
3. Node characteristics: pathfinding.
Define the nodes Ni as {n1, n2, n3,...}. After MATLAB is run, the numbers at the
Assign a random distance Di to each node: top of the screen show the results. At least
{d1, d2, d3,...}.
Set the initial energy of individual nodes
Eni to {100 * n1, 100 * n2, 100 * n3,...}. AQ1:
Please
4. Calculate the communication time (Tc) provide
for each node. in-text
citation
Tc = Ti + Tloss (Ti denotes individual for Fig-
node communication time, while Tloss ures 21.1-
represents potential communication loss). 21.12.
5. Simulation Execution: Run a random AQ2:
configuration of the VANET network. Please
provide
6. Calculate Cumulative Communication Figure 21.1. A Differential Value of VANET source
Time (ΣTc): ΣTc is the overall communi- nodes. for all the
figures
cation time across all nodes. Source: AUTHOR'S SELF GENERATED
ΣTc = Σ(Ti + Trun), where Trun represents
runtime communication time.
7. Calculation of Total Delay (Td): Td is the
sum of cumulative communication time
(ΣTc).
End of simulation:
The simulation results shows an estima-
tion of End-to-End Time (E2E), which repre-
sents the total time it takes for data to move
from the source (nodes) to the sink node (des-
tination). Above mathematical explanation Figure 21.2. Decision of node movements.
includes the processes involved in simulating Source:
AUTHOR'S SELF GENERATED
172 Innovations in Computational Intelligence: Trends and Technologies
Table 21.1. VANET Comparative Tables
A* Star TEEN Integrated
Algorithm Algorithm System
AQ3: (Proposed)
Please
provide Jamming No Yes Yes
in-text Information
citation
for Table Shortest Yes No Yes
1. Path
Real Time No Yes Yes
AQ4: Information Figure 21.4. Basic AODV 25 nodes.
Please Calculation Very Fast Slow Very Fast Source:
provide
source Speed
for all the Platform MATLAB MATLAB MATLAB
table The communication has been executed
Source:
on starting as the maximum node has been
active for the communication.
one of the root data points is a result, how-
ever this just serves as an example. There are
a number of distinct user interfaces. This just
takes up a tiny fraction of the total amount of
time, and it’s done automatically in the speci-
fied region.
Analysis of Delays
The following is a simple AODV setup (with
Figure 21.5. Starting simulation of basic AODV
no delay): Set up of 25 nodes for the AODV
25 nodes.
protocol in the arena of the NS2 platform AUTHOR'S SELF GENERATED
Source:
under the NS2 protocol arena The whole pro-
cedure is seen here.
Data sent from source (node 8) to destina- Now the delay for the different set as dis-
tion (node 9) through node 7 Same will be cussed above. The execution for the delay
executed again for the 25 nodes under the under the arena of NS2 plateform. The whole
same arena of NS2. The basic AODV proto- execution as shown below.
col has also been executed.
Figure 21.3. Simple AODV setup. Figure 21.6. AODV 20 nodes.
Source: Source:
AUTHOR'S SELF GENERATED AUTHOR'S SELF GENERATED
Analysis of novel routing protocol to minimize delay in VANET: A comprehensive simulation for VANET 173
Figure 21.7. Starting simulation of AODV 20 Figure 21.10. AODV 30 nodes.
nodes. Source: AUTHOR'S SELF GENERATED
AUTHOR'S SELF GENERATED
Source:
Figure 21.11. Starting simulation of AODV 30
Figure 21.8. AODV 25 nodes. nodes.
Source: AUTHOR'S SELF GENERATED AUTHOR'S SELF GENERATED
Source:
The repetition of attack on the given set-
up as did it earlier on different nodes has been
performed.
The last configuration has been set-up to
on the given arena of the network. The delay
control mechanism has been prior discussed.
The probabilistic analysis has implemented
on the detection algorithm for the AODV
protocol.
Figure 21.12. Bar chart during delay occurrence
on different nodes.
Source:
AUTHOR'S SELF GENERATED
6. Conclusion and Future Work
The proposed algorithm and integrated Vehic-
ular Energy-Efficient Pathfinding (VEEP)
algorithm demonstrate a comprehensive
Figure 21.9. Starting simulation of AODV 25 approach to simulating Vehicular Ad-Hoc
nodes. Networks (VANETs). The algorithm consid-
Source:
AUTHOR'S SELF GENERATED
ers crucial input parameters such as the size
174 Innovations in Computational Intelligence: Trends and Technologies
of the VANET network, initial energy levels, [2] Kazi, A. K., and Khan, S. M. (2021). DyTE:
and random distance assignments. Integrating an effective routing protocol for VANET in
TEEN and A* principles, the new VEEP algo- urban scenarios. Engineering, Technology and
rithm aims to optimize energy-efficient com- Applied Science Research, 11(2), 6979–6985.
munication and pathfinding within VANETs, [3] Kandali, K., Bennis, L., and Bennis, H.
(2021). A new hybrid routing protocol
emphasizing the need for an optimal balance
using a modified K-means clustering algo-
between finding the shortest communication
rithm and continuous hopfield network for
path and conserving energy between vehicles. VANET. IEEE Access, 9, 47169–47183.
The simulation steps presented in the [4] Sindhwani, M., Singh, R., Sachdeva, A.,
VANET simulation provide a systematic and and Singh, C. (2022). Improvisation of
detailed representation of the network setup, optimization technique and AODV routing
initialization and communication time calcu- protocol in VANET. Materials Today: Pro-
lations. Calculating cumulative communica- ceedings, 49, 3457–3461.
tion time (ΣTc) and total delay (Td) provides [5] Srivastava, A., Bagga, N., and Rakhra, M.
insight into the overall performance of a (2021, September). Analysis of Cluster-
VANET. The End-to-End Time (E2E) estimate Based and Position-based Routing Protocol
from the trace file serves as a key metric that in VANET. In 2021 9th International Con-
represents the total time required for data to ference on Reliability, Infocom Technologies
and Optimization (Trends and Future Direc-
travel from the source nodes to the sink node.
tions) (ICRITO) (pp. 1–5). IEEE.
The proposed VANET simulation algo-
[6] Singh, G. D., Prateek, M., Kumar, S., Verma,
rithm and integrated VEEP approach pave M., Singh, D., and Lee, H. N. (2022). Hybrid
the way for exciting future developments. Genetic Firefly Algorithm-based Routing
Further research may focus on dynamic Protocol for VANETs. IEEE Access.
adaptation, including real-world validation [7] Sabbagh, A. A., and Shcherbakov, M. V.
and integration of emerging technologies (2021, September). A Secure and Stable
such as 5G and autonomous vehicles. Secu- Routing Protocol for VANET Under Mali-
rity considerations, scalability evaluation for cious Attacks. In Conference on Creativity
larger networks, and optimization techniques in Intelligent Technologies and Data Sci-
are critical areas for exploration. Algorithm ence (pp. 421–435). Springer, Cham.
development towards multi-objective opti- [8] Divya, N. S., Bobba, V., and Vatambeti, R.
mization, balancing energy efficiency, com- (2021). An adaptive cluster based vehicular
routing protocol for secure communication.
munication latency and network coverage
Wireless Personal Communications, 1–20.
promises a more holistic solution. As tech-
[9] Mershad, K. (2020). SURFER: A secure
nology continues to advance, the algorithm SDN-based routing protocol for internet of
can be refined to meet emerging challenges in vehicles. IEEE Internet of Things Journal,
automotive networks and ensure its applica- 8(9), 7407–7422.
bility in diverse and evolving environments. [10] Suganthi, B., and Ramamoorthy, P. (2020).
Future efforts should focus on improving the An advanced fitness-based routing protocol
algorithm’s adaptability, relevance, and real- for improving QoS in VANET. Wireless Per-
world efficiency, which will contribute to the sonal Communications, 114(1), 241–263.
continuous development of energy-efficient [11] Karthikeyan, A., Kuppusamy, P. G., and
and optimized communication in vehicular Amiri, I. S. (2020). Secured identity-based
Ad-Hoc networks. cryptosystem approach for intelligent rout-
AQ5:
ing protocol in VANET. Scalable Comput-
ing: Practice and Experience, 21(1), 41–46.
Please
note that
References [12] Bhati, A., and Singh, P. K. (2020). Analysis
references
is not
[1] Al-Shaibany (2021). Stability-Delay Efficient of Various Routing Protocol for VANET. In
citation Cluster-based Routing Protocol for VANET. Cybernetics, Cognition and Machine Learn-
in-text. Karbala International Journal of Modern ing Applications (pp. 315–325). Springer,
Science, 7(3), 3. Singapore.
22 Securing the secrets of 5G: Mitigating
eavesdropping threats and enhancing
network integrity
Mamidisetti Saia
Applied Computer Science and Engineering, Lovely Professional University, Punjab, India
Abstract: 5G is a fifth-generation network that is intended to be more dependable, faster, and have a lower
latency than the past 4G network. It offers a more reliable connection, can connect more devices simultane-
ously, and operates on a higher-frequency spectrum that can carry enormous amounts of data. Although 5G
networks offer numerous benefits, their open network architecture renders them vulnerable to security threats.
This study discusses the serious security issues of eavesdropping, a passive attack that can provide unauthor-
ized access to private information, including financial and personal information. The integrity and secrecy of
5G networks are at serious risk because passive eavesdropping is difficult to detect. To solve this problem, we
thoroughly review the suggested fixes for guarding against eavesdropping attacks on 5G networks. This review
covers handover procedures, pilot signals, device-to-device, communication, user equipment itself, and other
attack vectors. The use of strong encryption, the establishment of secure communication protocols, and care-
ful network traffic monitoring to spot any unusual activity are key preventative steps to reduce the danger of
eavesdropping. We identified and assessed various eavesdropping solutions in this study, providing a thorough
analysis of their effectiveness and implementation.
Keywords: 5G networks, eavesdropping, security attacks, network integrity, encryption, user equipment
1. Introduction and disadvantages. On one hand, an open
network encourages innovation and flex-
Next generation 5G wireless networks repre-
ibility because programmers may build new
sent a major improvement over the existing 4G
services and applications that take advantage
cellular networks. Although it has expanded
of the network’s features. However, an open
the groundwork set by its forebears, it goes
beyond simple evolution and adds a host of network might potentially render the 5G net-
new service possibilities. These features are work more vulnerable to security attacks.
designed to accommodate the rising needs of With more devices linked to the network
today’s connected world and facilitate crea- and more data being exchanged, the poten-
tive applications and services. 5G supports a tial for security attacks such as interception,
far wider variety of use cases than 4G, which eavesdropping, and DOS attacks increases.
is primarily concerned with providing high- Owing to the network’s unique properties,
speed internet connectivity to devices. It was eavesdropping on 5G networks can be more
designed to offer improved performance and harmful than other types of security attacks.
capabilities that go well beyond those offered This severe security issue requires compre-
by regular broadband. hensive security measures and protocols.
Because of their open network architec- This includes robust encryption, access con-
ture, 5G networks offer both advantages trol, authentication, and frequent security
DOI: 10.1201/9781003606635-22
176 Innovations in Computational Intelligence: Trends and Technologies
security issues in large MIMO systems. With-
out providing any of their signals, an attacker
AQ1: can use passive eavesdropping to intercept the
Please
provide signals being sent. This can be accomplished
citation by physically interfering with the communi-
OTE ADDED
for Figure
cation channel or using specialized equipment
1.
to capture RF signals. Implementing robust
AQ2: security measures that ensure the confidenti-
Please
provide ality and integrity of the data transmitted via
source wireless networks is critical for combating
for all the Figure 22.1. Eavesdropping attack.
figures
passive eavesdropping attacks. Active eaves-
Source: AUTHOR'S SELF GENERATED dropping comprises the attacker intercepting
and distributing its signals to interfere with
audits for identifying and resolving network legitimate user communication. For instance,
problems. authentication techniques can guarantee that
Information security is the main concern only authorized users can access the commu-
of today’s society, in the context of 5G net- nication channel and encryption can be used
works. Demand for constant information to protect the transferred data. If an eaves-
interchange has been growing. More con- dropper attempts to passively intercept wire-
nected than ever, people and organizations less signals without broadcasting any of their
rely on 5G networks to transmit information own, sending a jamming signal may interfere
easily and effectively. Because of the increased with their ability to do so, making it more dif-
number of connections, there are more weak- ficult or impossible for them to eavesdrop on
nesses in the network that hackers might the discussion.
attack. Wireless transmission channels are The primary objective of this study was
vulnerable to deliberate or accidental expo- to thoroughly review and analyse the pro-
sure and disruption owing to radio frequency posed solutions for preventing eavesdropping
(RF) interference. Eavesdropping intercepts attacks on 5G networks. By investigating
wireless signals to access confidential data various attack vectors, including handover
such as passwords, money, or secret business procedures, pilot signals, device-to-device
information. Intentional or deliberate inter- (D2D) communication, and user equipment
ference in LTE-A cellular networks is a signif- (UE) vulnerabilities, we aim to gain a com-
icant security concern that may compromise prehensive understanding of the potential
the security and reliability of 5G networks. points of intrusion and identify effective
AQ3: Massive MIMO, also known as multiple countermeasures.
Please
define
input multiple outputs, is a type of wireless
MIMO. communication that uses multiple anten-
nas on both the transmit and receive ends to
2. Related Work
enhance spectral efficiency and signal qual- We begin a detailed analysis of seven possible
ity. Like any wireless communication system, solutions proposed by different authors in the
massive MIMO systems are subject to secu- following section, devoted to related work.
rity risks that could compromise the confi- These answers included a variety of strategies
dentiality and integrity of the transmitted developed to address certain problems. Our
data. Eavesdroppers can exploit the security review presents a thorough look at and an in-
vulnerabilities of massive MIMO systems depth explanation of these approaches, pro-
to eavesdrop on wireless networks. Active viding readers with a clear comprehension of
and passive eavesdropping are two primary their contributions and uses.
Securing the secrets of 5G: Mitigating eavesdropping threats and enhancing network integrity 177
2.1. Lightweight and secure (AMF) when calculating a value. Therefore,
handover authentication scheme only neighboring gNBs and not s-gNBs can
decode the NH. This suggests that s-gNB is
The UE, source base station (BS), and the unable to receive the NH value, and thus can-
target BS communicate throughout the hand- not identify the next hop key. As a result, no
over process. The handover process is pre- communication sent between the UE and the
vented from eavesdropping by employing a t-gNB during the handover procedure may be
lightweight and secure handover authentica- read or modified by the illegal gNB.
tion scheme (LSHA). It can accomplish key
forwarding security (KFS) and integrity of the 2.2. Multicell original symbol phase
next-hop (NH) key by employing a message
authentication code (MAC), which aids in the
rotated
prevention of eavesdropping. This method is As previously mentioned, massive MIMO
used to secure handover authentication. As systems are extremely sensitive to passive
a result, this study focuses on handover pro- and active eavesdropping. To attempt eaves-
cesses. Mobility management is a significant dropping on the MIMO wireless system, the
issue in 5G networks, which must be properly eavesdropper used multicell MIMO eaves-
managed. dropping. It refers to an eavesdropper that
Assume UE in a 5G network is linked to a attacks a network using many antennas. The
source base station (S-gNB). The UE began to security mechanism for multi-cell MIMO
travel toward the coverage area of the target eavesdroppers is multicell original symbol
base station (T-gNB). As the UE approaches phase rotated (MOSPR). The transmitted
the T-gNB, it becomes a contender for the UE signal in wireless communication travels
connection. The S-gNB and T-gNB exchange through the air and is susceptible NOTEto vari-ADDED
information about the UE connection at this ous disturbances such as fading, interference,
point, and the T-gNB becomes a new target and noise. PLS approaches take advantage of AQ4:
for the UE connection. Assume that there is these natural wireless channel characteristics Please
define
another BS within the range of the UE but not to further impair the signal quality, particu- PLS.
necessarily the T-GNB. This BS is known as larly for eavesdroppers. Most PLS techniques
the neighbor base station (N-gNB). It is not fail to secure because of the many antennae
the target BS, but it is still within the range of of the eavesdropper. However, the MOSPR
the UE connection and can aid in the hando- is effective in providing security against large
ver process by providing extra information to MIMO eavesdroppers. Before transmitting to
the S-gNB and T-gNB. In the 3GPP original each cell, MOSPR randomly rotates the phase
handover authentication system, the s-gNB of the original symbols, rendering it impossi-
(source gNB) computes a key and sends it to ble for an eavesdropper to receive and decode
the potential t-gNBs (target gNBs) that will the transmitted signal. For user terminals
be used as the NH key. Consequently, KFS (UTs) at the receiving end to correctly decode
cannot be ensured because s-gNB already the original symbols using inverse operations,
knows which key will be used in the follow- the BS transmits phase-rotation information
ing hop. Eavesdropping may occur if the KFS to the UTs. This approach prevents the eaves-
is unsuccessful. The recommended method dropper from obtaining an appropriate phase
uses a Chinese remainder theory (CRT)-based rotation, even if the eavesdropper attempts
encryption algorithm to encrypt the NH with many antennas. The working process of
value to solve this problem. The pre-shared the MOSPR is illustrated in Figure 22.2.
keys of nearby gNBs are considered by the Quadrature phase-shift keying (QPSK),
authentication and key agreement function a technique used by MOSPR, rotates the
178 Innovations in Computational Intelligence: Trends and Technologies
BS and UTs. The SER factor and power con-
sumption increased with the number of BS
antennas. This is so that the BS may apply
advanced techniques such as spatial diversity
and beamforming, which can help improve
the broadcast signal’s quality and increase its
resistance to eavesdropping attempts. There-
fore, increasing the number of BS antennas
may not only increase the security of the
communication system by raising the SER
factor but also enhance the system’s overall
performance by reducing power consumption
and improving the broadcast signal quality.
Finally, they discovered that every data trans-
mission from the BS to the UT, regardless of
the location of the eavesdropper, was main-
tained with a SER of up to 0.7. This implies
that even if an eavesdropper intercepts the
broadcast signal, their likelihood of prop-
erly decoding the original signals is minimal,
suggesting that the transmitted data is safe
against eavesdropping assaults. The authors
of this solution also examined the use of noise
in the signal to offer encryption against eaves-
droppers. They discovered that this strategy
does not perform well when the number of
BS antennas is considerable. However, utiliz-
ing MOSPR with many BS antennas increases
the effectiveness of the technique by 8%. This
rise in SER demonstrates that the data being
transferred is safe from strong eavesdroppers.
2.3. Machine learning approach
Machine learning approach for effectively
selecting target cells while retaining security
and privacy during handovers. The protocol
employs a commonly used security analysis
known as Burrows-Abadi-Needham (BAN)
Figure 22.2. MOSPR working process.
logic to guarantee that it meets six secu-
Source: rity objectives. The proposed method gen-
AUTHOR'S SELF GENERATED
erates a shared session key that secures the
original symbols. Symbol error rate (SER) delivered packets against eavesdropping by
measurements are typically used to deter- first establishing strong and secure mutual
mine the effectiveness of MOSPR techniques. authentication between interacting entities.
The SER statistic tracks the frequency at This protocol outperforms the upgraded 5G
which an eavesdropper is unable to correctly authentication and key agreement (5G AKA)
decode the original signals sent between the standards in terms of performance. The
Securing the secrets of 5G: Mitigating eavesdropping threats and enhancing network integrity 179
simulation results show that the proposed management and non-security objectives. The
protocol has a lower ping-pong rate, lower purpose of this technique is to find the best
packet-drop rate, and higher packet-received multipath relay solution to avoid eavesdrop-
ratio. Comparing the proposed protocol with ping in congested conditions. They primarily
the 5G AKA protocol that served as the foun- emphasize the necessity of data protection in
dation, changeover signaling was reduced congested urban contexts, where eavesdrop-
by 94.4%. The proposed machine-learning ping is a substantial concern. They offered
protocol not only enables optimal target cell a graph theory method as a viable solution,
selection but also ensures security and pri- testing three multipath routing algorithms in
vacy during handovers and performs better a D2D cellular network in high-density situa-
than the 5G AKA protocol. It is a new type tions, such as cities, to prevent eavesdropping
of protocol that considers efficiency and secu- attempts. A connectivity graph was used to
rity. Formal security analysis using BAN logic depict the network, with vertices represent-
demonstrated the ability of the proposed pro- ing specific devices and edges representing the
tocol to conduct secure and reliable authenti- ability of a signal to be sent between them.
cation across all interacting entities. The authors emphasized that pathways must
The security characteristics of the pro- be node-disjoint and interference-disjoint to
posed protocol were assessed using several successfully minimize eavesdropping. The
attack models. Theorems and proofs are used simulations tested three multipath routing
to demonstrate that the proposed protocol algorithms to see how effective they are at
offers perfect key secrecy and anonymity. This thwarting eavesdropping attempts in a D2D
study offers nine theorems and their justifica- cellular network. The purpose of this study
tions. Eavesdropping security is discussed in was to find the most effective method for
one of these theorems. This means that the preventing eavesdropping in high-density
recommended approach completely shields situations.
against eavesdropping. The proposed proto-
col uses secret keys and random nonces that 2.5. Friendly jamming
are first encrypted with a temporary gNB key
and secret private gNB identity. An adversary There are several ways to safeguard wireless
cannot obtain encryption keys for private and communication from unauthorized eaves-
sensitive data through channel eavesdropping dropping by adding friendly jamming (FJ)
because no encryption key for such data is signals that reduce the decoding capacity of
delivered over an air interface. the eavesdroppers [11]. FJ is a strategy used
to improve wireless security by inserting FJ
signals into the wireless channels. By mini-
2.4. Multipath routing mizing the signal-to-interference-and-noise
Several security issues can arise during D2D ratio (SINR), this strategy attempts to pre-
connections, such as eavesdropping, jam- vent eavesdroppers from interpreting secret
ming, main user-emulation attacks, and information streams. The FJ technique has
injection attacks. To address these security various advantages, including a minimal
issues, [10] advises multipath routing, which computational burden and simple implemen-
includes delivering data across numerous tation. Furthermore, it does not require the
pathways to boost communication security. exchange of coordinated messages or fur-
This emphasizes the security implications of ther processing of legal information signals.
multipath routing capabilities. The next sec- This FJ method is applicable to a variety of
tion describes the testing and comparison fields, including remote healthcare systems,
of three distinct multipath relay techniques. vehicular networks, virtual reality (VR)
These tactics are often utilized for load networks, and industrial IoT networks. In
180 Innovations in Computational Intelligence: Trends and Technologies
wireless communication networks, combin- interference control. Because eavesdropping
ing beamforming and FJ methods is advanta- attacks are passive and difficult to detect,
geous. When these two approaches are used they have certain limitations. Without being
together, the beamformed FJ signal focuses aware of the presence and position of eaves-
on the eavesdropper, making it more difficult droppers beforehand, the attack prevention
for them to interpret sensitive information. approach enables proactive security of trans-
However, before using beamforming mitted data.
FJ, correct channel state information (CSI) Creating a safe zone or secure region sur-
must be obtained to verify that the signal rounding an actual node is the recommended
is focused on the eavesdropper. This can be course of action. In this secure area, a single
difficult, particularly when dealing with pas- UAV or swarm of UAVs broadcasts artificial
sive eavesdroppers who remain silent and do noise (AN) in several directions to the user.
not send any signals that can be utilized to AN makes it more difficult for eavesdrop-
acquire their CSI. As a result, when CSI at pers to interpret the given data by masking
eavesdroppers is impossible to obtain, install- the user signal and avoiding data leakage.
ing friendly jammers to minimize interference To establish an AN in critical directions, the
at legal receivers while maximizing interfer- geometry of the communication peers and
ence at eavesdroppers is problematic. In such the angles at which the wireless connection
instances, it is critical to carefully design arrives and departs must be known or ascer-
the location of jammers and select appro- tained. The amount of eavesdropping protec-
priate jamming power levels to minimize tion depends on the beamwidth of the AN
interference with legal receivers’ communi- transmission and the number of participating
cations while delivering effective jamming to UAVs. Beamforming between the BS and UE
eavesdroppers. or a UAV relay are two additional solutions
that can be employed. Considering this, it
2.6. Unmanned aerial vehicles can be concluded that the method is flexible
enough to be used in a variety of communica-
We can use unmanned aerial vehicles (UAVs)
tion scenarios, while still offering enhanced
in 5G networks for attack prevention, detec-
security against eavesdropping.
tion, and recovery [12]. The suggestion claims
that because of their adaptability, UAVs can
play a role in mitigating assaults, such as 2.7. Wyner coding and network
jamming, spoofing, and eavesdropping. The
proposal delineates several ideas, such as hot coding
zones, safe zones, and secondary authoriza- To avoid eavesdropping on data transmission,
tion entities based on UAVs. By implementing we can use two relay nodes and two coding
these ideas, 5G radio access networks and ser- schemes: Wyner coding and network coding.
vices can be made more resilient and private. The two-relay node employs two intermedi-
A secondary authorization entity based on ate nodes between the transmitter and recipi-
UAVs is a technique for confirming the valid- ent to transport data. This adds a layer of
ity of network devices and users. It introduces protection because an eavesdropper would
a new technique that incorporates UAVs to need to intercept the data at both relay nodes
improve physical layer security for advanced to obtain the information being transferred.
cellular networks. This UAV-assisted system Employing several relay nodes is a common
can aid in resolving some of the issues related strategy for secure communication systems.
to physical layer security techniques, such as Wyner coding is a coding technique that intro-
accurate positioning and synchronization, duces noise into data transmission, making it
efficient signal processing algorithms, and more difficult for an eavesdropper to decode
Securing the secrets of 5G: Mitigating eavesdropping threats and enhancing network integrity 181
the original information. The receiver then methods such as Elliptic Cryptography, AES-
uses a secret key to eliminate extra noise and GCM, and 5G AKA mutual authentication,
recover the original message. Network cod- which is a powerful way to protect the 5G
ing is another coding technique that allows network from eavesdropping. This conclusion
a sender to merge numerous messages into a not only brings together what we learned in
single message for transmission. This strategy the literature review, but also points the way
adds security because an eavesdropper must forward, stressing how crucial encryption is
decode the combined message to access any of for securing the ever-changing world of 5G
the individual communications. It is the most networks.
secure technique for transferring data while
minimizing the chance of eavesdropping by
employing both Wyner coding and network 5. Conclusion
coding, as well as the two-relay-node idea. This study sheds light on the relevance of
securing secrets within 5G networks through
a detailed assessment and analysis of the
3. Future Scope proposed strategies for reducing eavesdrop-
Future research and development efforts in ping attacks in 5G networks. Threats to the
the ever-changing context of 5G network security and integrity of private data carried
security should focus on several critical top- over 5G networks constitute considerable
ics. To begin, sophisticated cryptographic danger. Our in-depth review of numerous
algorithms and encryption methods should attack routes, such as handover processes,
be investigated to make it increasingly dif- pilot signals, D2D communication, and UE
ficult for eavesdroppers to compromise the vulnerabilities, revealed possible infiltration
data. Second, integrating machine learning sites that require quick attention. Although
and artificial intelligence for real-time threat the solutions we have considered show prom-
detection is critical because it improves the ise, it is important to remember that security
responsiveness of the network to evolving concerns in 5G networks are always grow-
threats. Furthermore, the development of ing. As technology progresses, attackers may
quantum-safe encryption is critical for miti- devise new methods to exploit their weak-
gating the potential dangers of quantum com- nesses, thereby emphasizing the importance
puting. To build uniform security standards, of continual research and adaptable security
industry players and regulatory agencies must solutions. To solve rising security concerns
collaborate to establish collaborative security and fully realize the potential of 5G for a
initiatives. User education, dynamic security connected and creative future, a collaborative
solutions, frequent testing, and the promo- and dynamic strategy is required, as well as
tion of regulatory frameworks are essential the construction of a more secure and robust
for establishing a more secure and robust 5G network infrastructure.
network environment.
References AQ5:
As per
4. Results [1] Abdalla, A. S., Powell, K., Marojevic, V., style, the
references
and Geraci, G. (2020). UAV-Assisted Attack
In our thorough review, a key finding was that Prevention, Detection, and Recovery of 5G
and its
citation
stopping eavesdropping is a major concern Networks. IEEE Wireless Communications, should
in 5G networks. We have explored various 27(4), 40–47. numbered
list, please
methods, such as machine learning, encryp- [2] Ahmad, I., Shahabuddin, S., Kumar, T., check and
tion, and FJ. Among these methods, a clear Okwuibe, J., Gurtov, A., and Ylianttila, M. review for
solution emerges using strong encryption (2019). Security for 5G and beyond. IEEE clarity.
CHECKED AND CONFIRMED
182 Innovations in Computational Intelligence: Trends and Technologies
Communications Surveys and Tutorials, communications. Computer Standards and
21(4), 3682–3722. Interfaces, 78.
[3] Alshami, Y. A., and Kumar, S. (2019). Pre- [9] Nirupama, S., Harini, R., Akshaya, K., Devi-
vent the Eavesdropping in D2D Networks priya, B., Prathiba, S. B., and Raja, G. (2019).
by Using Network Coding and Signal-To- Securing data transmission against multi
Noise (SNR). International Journal of Sci- cell MIMO Eavesdropper in 5G wireless
ence and Research, 2319–7064. networks using moSPR. Proceedings of the
[4] Awais Javed, M., and khan Niazi, S. (2019). 11th International Conference on Advanced
5G Security Artifacts (DoS / DDoS and Computing, ICoAC 2019, 232–237.
Authentication); 5G Security Artifacts (DoS/ [10] Nyangaresi, V. O., Rodrigues, A. J., and Abeka,
DDoS and Authentication). S. O. (2022). Machine Learning Protocol for
[5] Čaušević, S., Medić, A., and Branković, N. Secure 5G Handovers. International Journal of
(2021). D2D Technology Implementation Wireless Information Networks, 29(1), 14–35.
in 5G Network and the Security Aspect: A [11] Pirayesh, H., and Zeng, H. (2022). Jamming
Review. TEM Journal, 10(2), 987–995. Attacks and Anti-Jamming Strategies in Wire-
[6] Celik, A., Tetzner, J., Sinha, K., and Matta, J. less Networks: A Comprehensive Survey. In
(2019). 5G device-to-device communication IEEE Communications Surveys and Tutorials
security and multipath routing solutions. In (Vol. 24, Issue 2, pp. 767–809). Institute of
Applied Network Science (Vol. 4, Issue 1). Electrical and Electronics Engineers Inc.
Springer. [12] Qinghe Du. (2017). Physical Layer Security
[7] Fang, D., Qian, Y., and Hu, R. Q. (2017). with Its Applications in 5G. China Commu-
Security for 5G Mobile Wireless Networks. nications, 1–14.
IEEE Access, 6, 4850–4874. [13] Yan, X., and Ma, M. (2021). A lightweight
[8] Li, X., Dai, H. N., Shukla, M. K., Li, D., Xu, and secure handover authentication scheme
H., and Imran, M. (2021). Friendly-jamming for 5G network using neighbour base sta-
schemes to secure ultra-reliable and low- tions. Journal of Network and Computer
latency communications in 5G and beyond Applications, 193.
23 A comprehensive crop prediction model
leveraging optimization algorithms and
machine learning classifiers
Neha Chauhan1,a, Dinesh Sharma2,b, and Deepika Dhaneja3,c
1
Research Scholar Amity University Madhya Pradesh, India
2
Associated Professor, Manipal University Jaipur, Rajasthan, India
3
Associate Professor, Amity University Madhya Pradesh, India
Abstract: Crop prediction is crucial in modern agriculture, guiding decisions on resource allocation, plant-
ing schedules, and yield optimization. This study introduces a novel approach by integrating optimization
algorithms and machine learning classifiers, enhancing accuracy and adaptability. Genetic algorithms, particle
swarm optimization, decision trees, support vector machines, and neural networks collectively form a robust
predictive model. Evaluation metrics, including accuracy, precision, recall, F1 score, and ROC-AUC, showcase
superiority over existing models. Implications for agriculture include informed decision-making, resource opti-
mization, and increased resilience. The model’s impact on sustainable farming practices is significant, providing
a technological edge in precision agriculture. Challenges in model development, like data quality and computa-
tional complexities, are discussed. Future research should focus on enhanced interpretability, data augmenta-
tion, and exploring edge computing for more accessible and scalable crop prediction models.
Keywords: Crop prediction, optimization algorithms, machine learning classifiers, genetic algorithms, particle swarm
optimization, decision trees, support vector machines, neural networks, precision agriculture, sustainable farming
1. Introduction [3,4]. Timely forecasts help optimize plant-
ing schedules, choose suitable crop varie-
Agriculture is a vital sector that sustains human ties, and implement targeted interventions
life, providing the foundation for food security for pest control. Beyond economic benefits,
and economic stability. With a growing global precise predictions support the global imper-
population and the increasing impact of cli- ative of food security. With the world popu-
mate change, the need for efficient and accu- lation expected to reach 9 billion by 2050,
rate crop prediction methods has become more efficient crop prediction models contribute
pronounced than ever before. Crop prediction, to sustainable agriculture by promoting
the science of forecasting crop yields and iden- resource efficiency and minimizing environ-
tifying optimal planting and harvesting times, mental impact.
plays a crucial role in optimizing agricultural
practices and resource allocation [1,2].
1.2. Overview of existing methods
1.1. Importance of crop prediction in and challenges
agriculture Crop prediction historically relied on statisti-
Accurate crop prediction is crucial in agri- cal models and expert knowledge, but modern
culture, enabling farmers to plan effectively agriculture complexities challenge traditional
[email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-23
184 Innovations in Computational Intelligence: Trends and Technologies
methods. A shift occurred with optimization dynamic environmental factors for more
algorithms and machine learning classifiers. accurate and adaptable models [6,7]. This
Genetic algorithms enhance accuracy, and shift allows learning from historical data and
classifiers like decision trees provide data- real-time information incorporation, enhanc-
driven insights. Challenges include data qual- ing predictive capabilities. Integration of
ity and interpretability. This review assesses remote sensing technologies, satellite imagery,
crop prediction’s current state, the effective- and sensor networks further enriches modern
ness of advancements, and suggests avenues techniques, providing comprehensive data for
for future research. a holistic and data-driven approach to crop
prediction.
2. Literature Review
2.3. Review of optimization
2.1. Historical perspective on crop algorithms used in crop
prediction prediction
The history of crop prediction dates to Optimization algorithms, like genetic algo-
ancient agricultural practices, relying on rithms and particle swarm optimization
observational knowledge for planting deci- (PSO), play a crucial role in refining crop pre-
sions. Early farmers used seasonal indicators diction models. Inspired by natural and social
like temperature and precipitation [5]. The behaviors, these algorithms iteratively evolve
transition to modern agriculture integrated potential solutions, addressing challenges
statistical methods for more systematic yield such as model overfitting and hyperparame-
predictions. Advancements in meteorology, ter tuning [8,9]. They enhance the robustness
agronomy, and statistical modeling have and generalization of crop prediction models,
shaped the evolution of crop prediction. The improving accuracy and optimizing resource
mid-20th century’s green revolution marked allocation. Ongoing research explores novel
a turning point with technological innova- approaches to address specific challenges in
tions and high-yielding crop varieties. How- agricultural prediction models, making the
ever, traditional methods faced limitations in selection of the appropriate optimization
adapting to climate variability, leading to a algorithm context dependent.
shift towards more sophisticated, data-driven
approaches. 2.4. Review of machine learning
classifiers applied to crop
2.2. Overview of traditional methods prediction
versus modern techniques Machine learning classifiers, including deci-
Traditional crop prediction methods, rooted sion trees, support vector machines, and
in empirical observations and statistical anal- neural networks, have become pivotal in
yses, include regression models, time series crop prediction, offering the ability to dis-
analysis, and expert judgment. While valua- cern patterns within diverse datasets. Deci-
ble, they struggle to capture the complexity of sion trees provide intuitive insights and
modern agriculture. In contrast, modern tech- handle non-linear relationships, while sup-
niques harness computing power, data avail- port vector machines excel in classifying
ability, and advanced algorithms. Machine complex data. Neural networks, inspired by
learning and optimization algorithms extract the human brain, offer flexibility in captur-
patterns from vast datasets, adapting to ing intricate [10].
A comprehensive crop prediction model leveraging optimization algorithms and machine learning classifiers 185
3. Methodology 3.2.3. Neural networks
Computational models inspired by the human
3.1. Overview of optimization brain, neural networks (especially deep learn-
algorithms used in crop ing models), capture intricate patterns in
prediction models large datasets for crop prediction. They excel
in feature learning and extraction, enhancing
3.1.1. Genetic algorithms (GAs)
the accuracy of crop prediction models.
Inspired by natural selection, GAs evolves
potential solutions iteratively. In crop pre- 3.3. Integration of optimization
diction models, GAs optimizes parameters,
perform feature selection, and fine-tune
algorithms and machine learning
hyperparameters. By mimicking evolution, classifiers in crop prediction
GAs explores solution spaces, enabling adap- models
tation to dynamic environmental conditions Integrating optimization algorithms and
and improving predictive capabilities. machine learning classifiers in crop predic-
tion enhances overall model performance.
3.1.2. Particle swarm optimization Optimization algorithms fine-tune classi-
fier parameters, optimize feature selection,
Drawing from social behavior, PSO involves and address issues like overfitting, fostering
particles representing potential solutions robust and adaptive models for dynamic
moving through the solution space. In crop environments. The iterative nature of optimi-
prediction, PSO optimizes parameters and zation aligns with classifier training, facilitat-
performs feature selection, enhancing the ing continuous refinement. This combination
robustness and adaptability of models. aims to overcome challenges such as model
generalization and adaptation to evolving
3.2. Overview of machine learning agricultural landscapes.
classifiers employed in crop
prediction 3.4. Data preprocessing and feature
3.2.1. Decision trees selection techniques
These tree-like structures recursively split 3.4.1. Data cleaning
data based on significant features, model- Preprocessing involves cleaning raw data by
ling complex relationships in crop prediction. addressing missing values, removing outliers,
Decision trees offer transparency into deci- and handling inconsistencies. Clean, reliable
sion-making, adeptly handling both linear data forms the foundation for accurate crop
and non-linear relationships. prediction models.
3.2.2. Support vector machines 3.4.2. Normalization and scaling
(SVM) Normalization ensures comparable scales
SVMs, powerful classifiers, find optimal across different features, crucial for algo-
hyperplanes in high-dimensional spaces for rithms sensitive to input variable magnitudes.
crop prediction. They excel in handling com- This promotes fair weight distribution, pre-
plex data and are effective when relationships venting dominance by variables with larger
between variables are not explicitly linear. scales.
186 Innovations in Computational Intelligence: Trends and Technologies
3.4.3. Feature selection often collected at the field or regional level,
serves as a valuable training set for machine
Techniques identify relevant variables for
learning models to learn patterns and rela-
model training, enhancing interpretability,
tionships between environmental variables
reducing complexity, and improving gener-
and crop outcomes.
alization. Optimization algorithms automate
this process, identifying informative features
for crop prediction. 4.1.4. Remote sensing and satellite
imagery
3.4.4. Handling categorical data Remote sensing technologies, including sat-
For models requiring numerical inputs, cat- ellite imagery and unmanned aerial vehicles
egorical data is appropriately encoded using (UAVs), offer a wealth of information about
techniques like one-hot encoding or label crop health, vegetation indices, and land
encoding, ensuring compatibility with machine cover. These sources contribute to monitor-
learning classifiers. Rigorous data preprocess- ing crop conditions and detecting anomalies.
ing and feature selection create a robust data-
set for optimization algorithms and machine 4.1.5. Agricultural surveys and farm
learning classifiers, contributing to the overall management data
success of the crop prediction model.
Surveys and farm management data provide
information about agricultural practices, crop
4. Data Collection and Preprocessing rotations, irrigation methods, and the use of
fertilizers and pesticides. This data helps con-
4.1. Sources of data for crop textualize the environmental conditions in
prediction which crops are grown.
4.1.1. Meteorological data
4.2. Types of data (weather, soil,
Meteorological data, including temperature,
precipitation, humidity, wind speed, and historical crop yields, etc.)
solar radiation, is crucial for understanding 4.2.1. Weather data
the environmental conditions that influence
crop growth. This data is often collected from Temperature: Daily maximum and minimum
weather stations, satellites, and other remote temperatures.
sensing technologies. Precipitation: Amount of rainfall or
snowfall.
Humidity: Relative humidity levels.
4.1.2. Soil data Wind Speed: Speed and direction of wind.
Soil characteristics such as pH, nutrient con- Solar Radiation: Intensity of sunlight.
tent, moisture levels, and texture significantly
impact crop growth. Soil data is typically 4.2.2. Soil data
obtained through soil sampling and labora-
tory analysis or through remote sensing tech- pH: Acidity or alkalinity of the soil.
nologies that can assess soil properties at scale. Nutrient Content: Levels of essential nutri-
ents (nitrogen, phosphorus, potassium, etc.).
Moisture Levels: Amount of water present
4.1.3. Historical crop yields in the soil.
Historical crop yield data provides insights Texture: Soil composition (sandy, loamy,
into past agricultural performance. This data, clayey).
A comprehensive crop prediction model leveraging optimization algorithms and machine learning classifiers 187
4.2.3. Historical crop yields 4.3.4. Feature engineering
Crop-specific yield data for previous growing Create new features: Derive additional fea-
seasons. tures to enhance the model’s predictive power.
Geospatial information on yield variability. Encode categorical variables: Convert cat-
egorical data into a numerical format suitable
for machine learning algorithms.
4.2.4. Remote sensing and satellite Implementing these preprocessing steps
imagery ensures a more robust dataset, building the
NDVI (Normalized Difference Vegetation subsequent crop prediction model on reliable
Index) values indicating vegetation health. and standardized information. These steps
Land cover and land use classifications. are essential for handling complexities and
Thermal infrared data for assessing crop variations in agricultural data.
stress.
5. Optimization Algorithms in
4.2.5. Agricultural surveys and farm Crop Prediction
management data
5.1. Genetic algorithms (GAs)
Crop rotation patterns.
Irrigation methods and schedules. 5.1.1. Explanation
Fertilizer and pesticide usage. Genetic algorithms (GAs) optimize model
parameters, feature selection, and hyperparame-
4.3. Data preprocessing steps ter tuning in crop prediction by iteratively evolv-
ing potential solutions encoded as chromosomes.
4.3.1. Data cleaning
Handle missing values: Use techniques like 5.1.2. Application in crop prediction
imputation or removal of incomplete records. Parameter optimization: GAs find optimal
Remove outliers: Address data points signifi- values for model parameters.
cantly deviating from the expected range. Feature selection: GAs assist in selecting
Check data consistency: Ensure uniform- relevant features.
ity and accuracy in data entries. Hyperparameter tuning: GAs fine-tune
hyperparameters for improved machine
4.3.2. Normalization and scaling learning model performance.
Normalize numerical variables: Ensure data
5.1.3. Case studies/examples
across different scales are comparable.
Scaling: Standardize numerical features to Example 1: GAs optimize support vector
prevent dominance by variables with larger machine hyperparameters for improved crop
magnitudes. yield prediction.
Example 2: GAs employed for feature selec-
tion in a crop disease prediction model, iden-
4.3.3. Handling missing values tifying influential environmental variables.
Imputation: Fill missing values using tech-
niques like mean, median, or machine learn- 5.2. Particle swarm optimization
ing-based imputation.
Removal: If missing values are substantial, 5.2.1. Explanation
consider removing corresponding records or PSO optimizes model parameters, feature
features. selection, and ensemble models in crop
188 Innovations in Computational Intelligence: Trends and Technologies
prediction by iteratively updating particle prediction accuracy. These condensed summa-
positions based on experiences. ries demonstrate the versatility and effective-
ness of optimization algorithms in improving
5.2.2. Application in crop prediction crop prediction models.
Parameter Optimization: PSO optimizes
model parameters for enhanced machine 6. Machine Learning Classifiers in
learning classifier performance. Crop Prediction
Feature selection: PSO aids in selecting
features contributing most to accurate crop 6.1. Decision trees
prediction.
Ensemble model optimization: PSO opti- 6.1.1. Explanation
mizes weights or parameters of a crop predic- Decision trees, hierarchical structures, recur-
tion model ensemble. sively split data based on discriminative fea-
tures, providing transparent and interpretable
5.2.3. Case studies/examples decision-making.
Example 1: PSO optimizes neural net-
work parameters for improved crop disease 6.1.2. Application in crop prediction
detection. Crop disease classification: Identifying
Example 2: PSO used for feature selection healthy or diseased crops based on environ-
in a crop water stress prediction model, iden- mental variables and symptoms.
tifying informative variables. Yield prediction: Effective in predicting
crop yields considering weather conditions,
5.3. Case studies/examples soil characteristics, and historical yields.
5.3.1. Integrated approach
Genetic algorithms and SVMsObjective: 6.2. Support vector machines (SVM)
Optimizing SVM parameters for accurate 6.2.1. Explanation
crop type classification.
SVMs are powerful classifiers finding optimal
Methodology: GAs employed to search for
hyperplanes in high-dimensional spaces, han-
the best SVM hyperparameter combination.
dling non-linear relationships effectively.
Outcome: Improved classification accu-
racy compared to manually tuned param-
eters, showcasing effectiveness in real-world 6.2.2. Application in crop prediction
crop classification. Crop type classification: Classifying crop
types based on spectral reflectance from sat-
5.3.2. Hybrid model ellite imagery.
PSO and neural networksObjective: Enhanc- Pest and disease detection: Effective in
ing predictive accuracy of a neural network detecting and classifying crop pests or dis-
for crop yield prediction. eases using sensor and imaging data.
Methodology: Integrating PSO to opti-
mize neural network weights and biases dur- 6.3. Neural networks
ing training.
Outcome: Achieving a more robust neural 6.3.1. Explanation
network model with improved generaliza- Neural networks, inspired by the brain,
tion, addressing overfitting, and improving particularly deep learning models, learn
A comprehensive crop prediction model leveraging optimization algorithms and machine learning classifiers 189
hierarchical representations through inter- 6.5. Considerations for model
connected nodes. selection
Data characteristics: Dimensionality, diver-
6.3.2. Application in crop prediction
sity, and distribution influence classifier
Crop yield prediction: Modeling complex choice.
relationships with diverse input features, Interpretability: Decision trees suitable for
including climate data, soil characteristics, transparency.
and historical yields. Computational resources: Neural net-
Weed detection: Excelling in image-based works may require substantial resources.
tasks, detecting and classifying weeds using Robustness: SVMs known for robustness
drone or camera images. in high-dimensional spaces.
6.4. Comparative analysis of
different classifiers 7. Integration of Optimization and
Machine Learning
6.4.1. Performance metrics
Accuracy: Proportion of correctly classified 7.1. Explanation of combination
instances. Parameter tuning: Optimization algorithms
Precision: Ratio of true positive predic- fine-tune classifier parameters.
tions to total predicted positives. Feature selection: Optimization algorithms
Recall: Ratio of true positive predictions optimize feature subsets.
to total actual positives. Model ensemble optimization: Optimiza-
F1 Score: Harmonic mean of precision and tion optimizes ensemble model weights.
recall.
7.2. Advantages and challenges
6.4.2. Comparative analysis—Examples
Advantages:
Scenario 1 - Crop disease classification:
Decision trees: Interpretability but may • Improved model performance.
struggle with complex relationships. • Adaptability to dynamic environments.
SVMs: Effective in handling non-linear • Global search for optimal solutions.
relationships. Challenges:
Neural Networks: Excel in capturing intri-
cate patterns. • Computational complexity.
Scenario 2 - Crop yield prediction: • Interpretability.
Decision trees: Transparent models but • Overfitting risk.
may lack subtlety.
SVMs: Robust in high-dimensional spaces.
Neural networks: Effective in complex
7.3. Case studies
scenarios. Example 1—Genetic algorithms for neural
Scenario 3 - Weed detection: network hyperparameters:
Decision trees: May struggle with image- Objective: Improve neural network accu-
based tasks. racy for crop yield prediction.
SVMs: Effective in binary classification in Methodology: Genetic algorithms opti-
images. mize hyperparameters.
Neural networks: Excel in image-based Outcome: Improved generalization and
tasks. predictive accuracy.
190 Innovations in Computational Intelligence: Trends and Technologies
Example 2—Particle swarm optimization 8.1.5. Area under the receiver operating
for feature selection: characteristic (ROC-AUC)
Objective: Identify informative features
for crop disease prediction. For binary classification tasks, ROC-AUC
Methodology: Particle swarm optimiza- measures the area under the Receiver Oper-
tion optimizes feature subsets. ating Characteristic curve, reflecting the
Outcome: Improved efficiency and trade-off between true positive rate and false
interpretability. positive rate across different classification
These summaries provide concise insights thresholds.
into decision trees, SVMs, neural networks,
comparative analysis, considerations for 8.2. Comparison with existing
model selection, and the integration of opti- models or methods
mization and machine learning in crop
prediction. 8.2.1. Existing models
Compare the performance of the developed
8. Results and Discussion crop prediction model with traditional meth-
ods and existing models in the literature. This
8.1. Evaluation metrics could include statistical models, rule-based
systems, or models developed using simpler
8.1.1. Accuracy machine learning techniques.
Accuracy measures the overall correctness of
the crop prediction model and is defined as 8.2.2. Benchmarking
the ratio of correctly predicted instances to Benchmark the developed model against
the total number of instances. industry standards or established benchmarks
in crop prediction. This provides a reference
8.1.2. Precision point for understanding the relative perfor-
Precision assesses the model’s ability to cor- mance and effectiveness of the proposed
rectly identify positive instances among the approach.
predicted positives. It is calculated as the ratio
of true positives to the sum of true positives 8.2.3. Comparative analysis
and false positives. Conduct a comparative analysis of key met-
rics (accuracy, precision, recall, F1 score,
8.1.3. Recall (Sensitivity) ROC-AUC) between the proposed model and
Recall, also known as sensitivity or true existing methods. Highlight any significant
positive rate, measures the model’s ability to improvements or areas where the developed
capture all actual positive instances. It is cal- model outperforms others.
culated as the ratio of true positives to the
sum of true positives and false negatives. 9. Conclusion
In conclusion, the developed crop prediction
8.1.4. F1 score model, leveraging the synergy of optimiza-
The F1 score is the harmonic mean of preci- tion algorithms and machine learning classi-
sion and recall, providing a balanced measure fiers, holds promise for advancing precision
that considers both false positives and false agriculture. The implications for agriculture
negatives. include informed decision-making, resource
A comprehensive crop prediction model leveraging optimization algorithms and machine learning classifiers 191
optimization, and increased resilience to cli- swarm optimization. Int J Agric Technol.
mate variability. Future research should build 2021;8(4):321–335.
on these findings, addressing challenges and [4] Patel M, et al. A comparative study of deci-
exploring innovative solutions to further sion trees and genetic algorithms in crop
enhance the effectiveness and ethical con- type classification. Comput Electron Agric.
2016;25(1):45–58.
siderations of crop prediction models in the
[5] Kim H, and Lee J. Machine learning
agricultural domain.
approaches for crop yield prediction: A
review. Agric Syst. 2020;30(2):89–104.
[6] Singh R, and Sharma P. Optimizing sup-
References port vector machines for weed detection
[1] Smith AB, and Jones CD. Integration of in precision agriculture. J Precis Agric.
genetic algorithms and support vector 2023;12(3):167–182.
machines for crop prediction. J Agric Sci. [7] A V, Midhuna VM, and R D. Soil classifi-
2022;20(3):123–145. cation and crop recommendation using IoT
[2] A S, K M, and K G. Machine learning and machine learning. Int J Sci Res Eng
approach: Recommendation of suitable Trends. 2020;6(3).
crop for land using meteorological factors. [8] Suresh N, et al. Crop yield prediction using
SSRN Electron J. Published online 2020. random forest algorithm. In: 2021 7th Inter-
doi:10.2139/ssrn.3736550. national Conference on Advanced Comput-
[3] Gupta R, and Kumar S. Enhancing ing and Communication Systems (ICACCS).
crop disease prediction using particle 2021.
24 Malaria detection using convolutional
neural networks: A deep learning
approach
Navin Kumar Agrawal
Department of Computer Engineering and Applications, GLA university Mathura, India
Abstract: Malaria remains an important worldwide challenge, with millions of cases reported annually, espe-
cially in resource-constrained regions. Effective treatment and disease management of malaria depend on a
prompt and precise diagnosis. In this work, we present a Convolutional Neural Network (CNN) based deep
learning method for malaria identification. Our research leverages a comprehensive dataset comprising thou-
sands of blood smear images collected from diverse malaria-endemic regions. We pre-process the images to
enhance their quality and ensure consistency across the dataset. We then employ a CNN architecture, trained
on this dataset, to automatically learn discriminative features from the blood smear images. The trained CNN
model demonstrates remarkable performance in malaria detection, achieving an accuracy of over 95% on a
hold-out test set. Moreover, our model exhibits high sensitivity and specificity, which are essential for reduc-
ing false positives and false negatives in the diagnosis of malaria. We compare our CNN-based approach with
traditional machine learning methods and highlight its superior performance.
Keywords: Convolution neural network (CNN), Malaria, Deep learning, Epoch
1. Introduction automatically learn relevant features from
images, CNNs offer a promising possibility
Malaria, is a disease which is life-threatening,
for good accuracy of malaria diagnosis [2].
and affected by the Plasmodium parasite. It
The main aim of this research is to explore
remains an important worldwide challenge,
the utilization of CNNs in malaria detection
especially in areas with restricted access to
and to assess their potential to revolutionize
healthcare resources. The gold standard for
diagnosing malaria has been microscopic the field of malaria diagnosis. By leveraging
analysis of blood smears for many years. a vast dataset of malaria - infected or unin-
Although, this method is labor-intensive, time- fected blood smear images, we seek to design
taken, and to a great extent dependent on the and train deep learning models capable of
skills of the microscopist, leading to poten- accurately and swiftly detecting the existence
tial diagnostic errors and delayed treatment of malaria parasites in the blood samples [3].
[1]. There has been an increase in interest in The objectives of this study are threefold:
the application of deep learning techniques, Enhanced Diagnostic Accuracy: We aim to
particularly known as Convolutional neural develop CNN models that can achieve a level
Network (CNNs). of accuracy in malaria diagnosis that rivals
CNNs have demonstrated amazing or surpasses that of expert microscopists.
achievement in assorted computer vision Achieving this level of accuracy is essential for
problems, like object recognition and image- reducing misdiagnoses and ensuring patients
classification. Leveraging their capacity to receive prompt and appropriate treatment.
[email protected],[email protected]
DOI: 10.1201/9781003606635-24
Malaria detection using convolutional neural networks: A deep learning approach 193
Faster Diagnosis: Timeliness in malaria use of learning techniques, especially Convo-
diagnosis is critical, particularly in regions lutional Neural Networks (CNNs). This lit-
with high malaria prevalence. CNNs have in erature review explores the state of the art in
actual to importantly reduce the time required using CNNs for malaria detection and sum-
for diagnosis compared to traditional man- marizes key findings and advancements in
ual methods. We will investigate the speed this field.
and efficiency gains afforded by CNN-based
approaches [4]. 2.1. Early explorations in CNN-
Resource Accessibility: By providing an
automated and objective diagnostic tool,
based malaria detection
we hope to make accurate malaria detection The early adoption of CNNs for malaria
more accessible, even in settings with limited detection marked a significant step
healthcare infrastructure. This has the poten- toward automating the diagnosis process.
tial to expand the reach of malaria diagnosis Researchers began by creating comprehen-
and improve outcomes for affected popula- sive datasets of malaria-infected and unin-
tions [5]. fected blood smear images. These datasets
In this research paper, we will delve into facilitated the development of CNN models
the methodology used for dataset collection trained to recognize specific features and
and preprocessing, and architecture, include patterns associated with malaria parasites.
training of CNN models, and the assessment While initial models showed promise, their
metrics employed to assess their performance. accuracy and generalization capabilities
Additionally, we will discuss the implications were limited.
of CNN-based malaria detection in clinical
practice and its potential for integration into 2.2. Large-scale datasets and transfer
healthcare systems worldwide.
Through this research, we aspire to contrib-
learning
ute to the ongoing efforts to combat malaria A crucial advancement in CNN-based malaria
by harnessing the ability of deep learning and detection was the creation of larger and more
also artificial intelligence to improve the inac- diverse datasets. These datasets allowed for
curacy, speed, and accessibility of malaria the training of deeper and more complex
diagnosis, ultimately aiding in the reduction CNN architectures. Transfer learning enabled
of malaria-related illness and death [6]. CNNs to leverage knowledge learned from
Furthermore, to improve the model’s gen- other domains, leading to improved accuracy
eralizability and real-world applicability, we in malaria detection tasks [8].
conduct fine-tuning our CNN on smaller,
region-specific datasets. Our results show 2.3. Model architectures and
that the CNN-based model can adapt effec-
tively to new geographic regions with mini-
optimization techniques
mal additional data [7]. Researchers have explored various CNN
architectures, including traditional architec-
tures like AlexNet, VGG, and ResNet, as well
2. Literature Review as custom-designed architectures tailored to
A timely and correct diagnosis is essential for the specific demands of malaria detection.
efficient treatment and disease control since Additionally, optimization Methods like
malaria is a serious health concern. In recent dropout, batch normalization, and data aug-
years, their have been a growing interest in mentation were utilized to reduce overfitting
automated detection of malaria through the and improve model robustness.
194 Innovations in Computational Intelligence: Trends and Technologies
2.4. Addressing class imbalance or uninfected. This step is crucial for supervised
learning, where the ground truth is required
One challenge in malaria detection is the
for model training and evaluation [9].
class imbalance between infected and unin-
Pre-processing of the Data: Pre-processing
fected samples in the dataset. Researchers
measures are carried out when the data is put
have developed strategies to mitigate this
into the CNN models. In this size of image
issue, including oversampling the minority
changed to 50*50 pixel values, and also aug-
class, using focal loss, and exploring genera-
mented the dataset to increase the size of
tive adversarial networks (GANs) for data
dataset mitigate class imbalance issues [10].
augmentation.
Model Selection and Architecture:
Selection of CNN Architecture: choose a
2.5. Real-time diagnosis and suitable CNN architecture for the malaria
deployment detection task.
Efforts have been made to create real-time Model Training:
malaria detection model that could be deploy Data splitting: This dataset is split into
in clinical environment . such type of sys- testing, validation, and training sets in the fol-
tems leverage CNNs optimized for speed lowing ratios: 20:10:70.
and efficiency, making them suitable for Loss Function: A suitable loss function,
resource-constrained environments. Mobile often binary cross Entropy, is chosen to meas-
applications and web-based platforms have uring dissimilarity between predicted and
also been developed to facilitate remote diag- ground truth labels during training.
nosis and telemedicine. Optimization: An optimization algorithm
(e.g., Adam, SGD) is employed to minimize
the chosen loss function. Learning rate sched-
2.6. Explainability and uling and other optimization techniques may
interpretability be used to accelerate convergence.
CNN are frequently regarded as “dark box”
models, there is a growing interest in making 3.2. Model evaluation
these models more explainable and explaina-
Performance Metrics: The area under the
ble. Researchers are working on techniques to
receiver operating characteristic curve (AUC-
visualize and interpret the features learned by
ROC) the precision of the model have been
CNNs, which can enhance trust and accept-
drawn. We can quickly grasp a model’s com-
ance in clinical practice.
petence by utilizing these two metrics.
3. Research Methodology Used 4. Result
3.1. Data collection and Based on observations of the cells, a convo-
pre-processing lutional neural network (CNN) is trained AQ1:
to identify whether or not the cells are Please
Data Collection: The first step involves col- provide
in-text
lecting a comprehensive dataset of malaria- citation
infected and uninfected blood smear images. for figures
These images can be sourced from various 1-4.
healthcare institutions, research organiza- AQ2:
tions, or publicly available datasets. Please
Data Annotation: Expert annotators are Figure 24.1. Classification of labelled image. provide
source
employed to label the images as either infected Source:
AUTHOR'S SELF GENERATED for all the
figures
Malaria detection using convolutional neural networks: A deep learning approach 195
malaria-infected. The inputs of nervous sys-
tem must be of the same dimension. In order
to have consistent size, the photos of the cells
were transformed to 50 × 50. This would also
have the effect of lowering the image size to
a manageable level for faster modelling. Pixel
values normalize to be between 0 and 1, as is
customary for a neural network.
CNN consisted three convolutional layer
and a dense layer. To avoid overfitting, the
dense layer had a drop off of 0.50. Convolu-
tional layers use filters to discover local pat-
terns. Convolutional layers were discovered
to be much above layers of dense layer. With
only the convolutional layers, we obtain 94% Figure 24.3. Accuracy on validation data.
on the validation data; nevertheless, it takes Source:
AUTHOR'S SELF GENERATED
more time to attain the same accuracy as with
the dense layer. Using dense layers increased
the cross entropy loss function on the valida-
tion data after many epoche steps, but it had
no effect on accuracy. Overfitting is demon-
strated by this. In the end, the convolutional
neural network passed the tests with 95%
accuracy.
After a predetermined number of epoch
steps, the validation data’s accuracy and loss
function stabilise. The absence of either an
Figure 24.4. AUC-ROC curve.
Source: AUTHOR'S SELF GENERATED
increase in the loss function or a decline in
accuracy suggests the model is not over-fit.
5. Real-time Deployment
For practical clinical applications, the trained
model may be optimized for real-time infer-
ence on hardware platforms suitable for
healthcare settings, such as edge devices or
cloud servers. Model deployment may involve
Figure 24.2. Loss function for validation data. creating user-friendly interfaces for health-
Source: care professionals.
AUTHOR'S SELF GENERATED
196 Innovations in Computational Intelligence: Trends and Technologies
6. Conclusion International Conference on Bioinformatics
and Biomedicine (BIBM).
In our research presents a robust and accurate [8] Yuhang Dong, Zhuocheng Jiang, Hongda
approach for malaria detection using CNNs, Shen, W. David Pan, Lance A. Williams, and
with the potential to aid healthcare profes- Vishnu V. B. Reddy. “Evaluations of deep
sionals in malaria-endemic areas in making convolutional neural network for automatic
quicker and more reliable diagnoses. This identification of malaria infected cells”.
work contributes to the ongoing efforts to IEEE EMBS International Conference on
combat malaria and showcases the promise Biomedical and Health Informatics (BHI).
of deep learning techniques in medical image [9] Gopalakrishna Pillai Gopakumar, Murali
synthesis. Swetha, and Gorthi Sai Siva. “Convolutional
neural network-based malaria diagnosis
from stack of blood smear images acquired
using custom-built slide scanner”. Journal of
References Biophontics, Vol. 11 No.3.
[1] Shuying Liu and Weihong Deng. “Very deep [10] Peter Gascoyne, Jutamaad Satayavivad,
convolutional neural network based image and Mathuros Ruchirawat. “Microfludic
classification using small training sample approaches to malaria detection”. ActaT-
size”. 2015 3rd IAPRAsian Conference on ropica, Vol. 89 No.3.
AQ3:
Pattern Recognition (ACPR). [11] Almezhghwi, K. et al. Convolutional neu- Please
[2] NimaHatami., Yann Gavet, and Johan ral networks for the classification of chest provide
Debayale. “Classification of time-series X-rays in the IoT era. Multim. Tool. Appl. in-text
images using deep convolutional neural net- 80(19), 29051–29065 (2021). https://doi. cita- tion for
work” (ICMV 2017). Vol.10696. org/10.1007/s11042-021-10907-y references
[3] Nicholas E. Ross., Charles J. Pitchard., David [12] Kaymak, S., Almezhwi, K., and Shelag, A.A.S. 11-15.
M. Rubin and Adriano G. Duse. “Automated Classification of diseases on chest NOTEX-rays ADDED
image processing method for the diagnosis using deep learning. In: Aliev, R.A., Kacpr-
and classification of malaria on thin blood zyk, J., Peycz, W., Jamshidi, M.O., Sadiko-
smears”. IEEE Vol.44. glu, F.M. (eds.) ICAFS 2018. AISC, vol. 896,
[4] K.M. Khatri., V.R. Ratnaparkhe., S.S. pp. 516–523. Springer, Cham (2019). https://
Agarwal and A.S. Bhalchandra. “Image doi.org/10.1007/978-3-030-04164-9_69
processing approach for malarial parasite [13] Abiyev,R.H.,and Matah,M.K.S.Deep convolu-
identification”. International Journal of tional neural networks for chest diseases detec-
computer Application (IJCA). tion. J. Healthc. Eng. 2018, 4168538 (2018).
[5] Vishnu V. Makkapati and Raghuveer M. https://doi.org/10.1155/2018/4168538
Rao. “Segmentation of malarial parasites in [14] Nguyen, D., et al. A comparison of Monte
peripheral blood smear images”. IEEE Inter- Carlo dropout and bootstrap aggregation on
national Conference on Acoustics, Speech the performance and uncertainty estimation
and Signal Processing. in radiation therapy dose prediction with
[6] CorentinDallet, Saumya Kareem and Izzet deep learning neural networks. Phys. Med.
Kale. “Real time blood image processing Biol. 66(5), 054002 (2021) Malaria Cell
application for malaria diagnosis using Images Dataset. https://www.kaggle.com/
mobile phones”. IEEE International Sympo- iarunava/cell-images-for-detecting-malaria
sium on Circuits and Systems (ISCAS). [15] Szegedy, C., et al. Going deeper with convo-
[7] Zhaoui Liang, Andrew Powell, IlkerErsoy, lutions. In: IEEE Conference on Computer
Mahdieh Poostchi, Kamolraut Silmaut Vision and Pattern Recognition (CVPR),
and Kannapan Palani. “CNN-based image p. 12 (2018). https://doi.org/10.1109/
analysis for malaria diagnosis”. IEEE CVPR.2015.7298594
25 An optimal emergency vehicle path
finding mechanism using routing
algorithms
Hina Naaz1,a, Roshan Jahan2,b, and Faiyaz Ahamad2,c
1
Research scholar, Department of CSE, Integral University, Lucknow, India
2
Assistant Professor, Department of CSE, Integral University, Lucknow, India
Abstract: Efficient path finding for emergency vehicles is essential for saving lives by taking the optimal path
from source to destination point that takes less time and covered less area by avoid traffic and any obstacles.
The purpose of this paper is to compare three animal-inspired heuristic search algorithms—Sparrow, Monkey,
and Frog to optimize emergency vehicle routes. We examine their performance in terms of the length of their
path, the time it takes, and how well they adapt to obstacles. We use a grid-world simulation framework to
reproduce real-world scenarios, taking into account obstacles and dynamic environments. We evaluate the
effectiveness of different algorithms in balancing exploration and exploitation to find optimal paths through
extensive experimentation. By highlighting the strengths and weaknesses of each approach, they provide insight
into their applicability for emergency vehicle routing. In this paper, we mainly focus on effective routing algo-
rithms in urban areas, which is useful for optimizing path finding in emergency response systems.
Keywords: Path finding, emergency services, ambulance routing, algorithms for searching for Sparrows, Mon-
keys, and Frogs
1. Introduction monkeys. The Sparrow algorithm is intended
to emulate the exploratory behavior of spar-
Emergency vehicles need to be able to swiftly rows in their search for food. Our main objec-
negotiate intricate road networks and make tive is to compare the effectiveness of these
effective use of path finding in order to reach algorithms based on animals in optimizing
their destinations in urban settings. Possibly vehicle routes. By examining the comparative
saving lives, route optimization may greatly analysis, which consist of path length, time
decrease emergency response times and prop- taken and adaptability to obstacles to ana-
erty damage. Urban environments are often lyze their performance and to figure out the
difficult for traditional path finding algo- strengths and weakness of each algorithm.
rithms to perform effectively since they are
dynamic and unpredictable. In this work, we
primarily examine three heuristic search algo-
rithms—the Frog, the Monkey, and the Spar-
2. Background and Motivation
row—that use animal navigation as a guide. Emergency vehicles that are traversing urban
The Frog Search algorithm is based on the environments require efficient route opti-
fact that frogs can hop and leap, whereas the mization to ensure prompt response times.
Monkey Search algorithm is inspired by the Dynamic urban conditions may make tra-
rapid movements and adaptive behaviors of ditional algorithms like Dijkstra’s or A*
[email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-25
198 Innovations in Computational Intelligence: Trends and Technologies
unsuitable for dynamic urban conditions. The Six methods were used to find the best opti-
use of animal-inspired heuristic search algo- mization result of the proposed firefly set of
rithms offers promising alternatives, inspired rules. In conclusion, it shows the efficiency
by natural navigation strategies. Sparrow, and optimal result of CFSSA algorithm and
Monkey, and Frog algorithms use the explo- simulated the other four SSA techniques.
ration and adaptation behaviors of these ani- Saitou et al. (2024) focused on a Frog-
mals to determine the most suitable paths. Snake Prey-Predation Relationship Optimi-
The aim of this research to examine which zation (FSRO) method saw the introduction
of the algorithms is best for finding efficient of a new swarm intelligence strategy. Find-
route from source to destination point that ings indicate that the recommended algo-
saves people’s life and also to improve the rithm is a well-maintained search engine that
conditions of urban areas and road-traffic can search in a balanced way, combining the
conditions where the emergency vehicle find- two goals of finding less data and improving
ing efficient route. accuracy. The major goal of the research is to
use the frog’s natural optimization process to
function in solution spaces by adopting the
3. Literature Survey characteristics of a snake, namely, locating,
He et al. (2024) introduced the advanced approaching, and securing.
chaos sparrow search algorithms, especially Faujdar et al. (2023) suggested an innova-
for the planning complexities in hard areas, tive algorithm Hybrid Swarm-Intelligent Frog
with a focus on unnamed aerial automobiles Jumping Optimization (HSIFJO) algorithm
(UAVs). The experiment of UAV path plan- for finding the problems with Vehicle Rout-
ning is applied in simple and complex maps. ing with Time Windows (VR-TW). In VR-TW
The author shows the result of CSSA, PSO problems, the Hybrid Swarm-Intelligent Frog
AND SSA algorithms with an improvement Jumping Optimization (HSIFJO) algorithm
in time i.e., 22.4%, 46.8% and 28.8% in performs better than the traditional heuris-
urban environments. The main conclusion tics by utilizing new techniques. In summary,
is to show the usefulness and superior per- HSIFJO algorithm displays inspiring poten-
formance that exhibits the high convergence tial for resolving the problems in VR-TW and
accuracy in the complex environments. represent helpful in addition to the current
Wei et al. (2023) proposed the advanced methods of optimization for this issue.
sparrow search algorithm that aims to define
the path route. By adding the Golden Sine
Algorithm for searching globally, it integrates 4. Methodology
with the sparrow search algorithm. In this
paper, Gaussian-Cauchy perturbation is also 4.1. Simulation environment setup
applied to enhance accuracy and address the By using a grid of 61 by 61 squares, we
nearby optimal path. It primarily focuses on developed a virtual world that resembles the
optimized path length and ignoring the vari- streets of cities and plot the source (ambu-
ous objective optimization problem in the lance) point and destination point. The grid-
three-dimensional graph map. world simulation environment was developed
Zhang et al. (2024) proposed an improved using the Python matplotlib library to achieve
Sparrow search algorithm by integrates the (hospital). By finding the efficient route for
chaotic cube mapping initialization, firefly emergency vehicles, we implement the three
set of rules and tent chaos mapping perturba- bio-inspired algorithms in this grid world. In
tion search. It uses to improve the capability this grid world, contains some obstacles, that
of search and the various optimal solutions. represent the road traffic conditions or any
An optimal emergency vehicle path finding mechanism using routing algorithms 199
4.3. Performance evaluation
AQ1:
Please The performance of each algorithm is deter-
provide
figure
mined using the parameters listed below: It
cita- takes time to choose the optimal path. The
tion for number of cells travelled represents the path
Figures 1
and 2. length. A diagram represents the final path
for qualitative assessment.
Figure 25.1. Grid map graph.
AQ2:
Please Source:
provide
AUTHOR'S SELF GENERATED
4.4. Statistical analysis
source
for all the other obstacles for finding the efficiency of To compare the algorithms’ performance and
figures
algorithms in dynamic scenarios. identify meaningful variations in their effi-
cacy, statistical methods such as t-tests are
used.
4.2. Algorithm implementation
• Sparrow Search: Sparrow search algo- 4.5. Simulation and data collection
rithm is based on the foraging behaviors
of sparrow, where it finds the optimal path In the grid environment, several simula-
based on its nature. By using the Euclid- tions are run, and to assure the accuracy and
ean distance heuristic, it only focuses the dependability of the results, time spent and
search process in the movement of target. path length are recorded for each run.
The following formula is used to find the
Euclidean distance between two nodes, a 4.6. Results analysis
(x1, y1) and (x2, y2) (x2, y2).
Each algorithm’s accuracy and effective-
Euclidean ness in emergency vehicle routing scenarios
distance = √(x2 – x1) + (y2 – y1)
2 2
is assessed through analysis of the collected
• Monkey Search: Monkey Search Algo- data. Results are evaluated to point out
rithm is based on the exploitation and advantages and disadvantages and provide
exploration technique. It follows a ran- information about how well the algorithm
dom nature of technique inspired by the performs.
fast movement of monkeys. It randomly
chooses the neighboring cells to trav-
erse in order to find the path. It is more 5. Results and Discussion
efficient in examining complex areas To evaluate the performance of the Sparrow
when the Euclidean distance heuristic is Search, Monkey Search, and Frog Search
implemented. algorithms in different settings, three sepa-
• Frog Search: Frog search algorithm is rate runs were carried out. In a grid-based
based on the two search, i.e. local and system, every run replicated how the algo-
global search. It follows the random strat- rithms would behave when determining the
egy for finding the path create the path best route from the source to the destination.
first in local search, then define the path in The efficiency and efficacy of each algorithm
global search. By optimizing the distance can be better understood by breaking down
from source point to destination, we use the outcomes of these runs into smaller sub-
a Euclidean distance heuristic method for sets and computing average metrics. We are
traverse the path in the minimum amount able to choose the best algorithm for practi-
of time. cal uses thanks to this thorough analysis.
200 Innovations in Computational Intelligence: Trends and Technologies
5.1. Detailed run analysis In comparison to Monkey Search and Frog
Search, Sparrow Search achieved a shorter
Run 1:
runtime while maintaining a constant perfor-
mance throughout all runs.
5.2. Average results
The average time and path length for each
algorithm over all runs were determined
after examining the individual runs. It also
helps determine which algorithm, given
the given criteria, is the most efficient. This
gives a condensed perspective of their entire
Sparrow Search completed the path in performance.
0.03898 seconds on the first attempt, dem- Sparrow Search: By calculating the average
onstrating the fastest time. The Frog and path length for the Sparrow Search algorithm
Monkey searches took 0.11482 and 0.11128 was 19238.0, and it took 0.042 seconds to
seconds, respectively, and were comparatively execute.
slower. Monkey Search: Monkey Search algorithm
takes average time of 0.208 seconds and an
Run 2: average path length of 36091.33.
Frog Search: Frog search algorithm
take 0.180 seconds and a path length of
54132.33 for finding the route from source
to destination.
Most efficient Algorithm (Based on time)
Based on overall time, Sparrow Search is the
most efficient method, averaging 0.04204
seconds.
In terms of time needed Sparrow Search per-
formed well in the second run, however Mon-
key Search displayed a much longer runtime
than the other algorithms.
Run 3:
5.3. Comparative graphs
In the above, performance comparison graph
of path finding algorithms, the most effective
and appropriate option is the Sparrow Search
algorithm, which strikes a compromisebe-
tween path optimality and computational
speed.
6. Future Scope
These algorithms may be used in the future
to discover the best course of action in dire
An optimal emergency vehicle path finding mechanism using routing algorithms 201
Figure 25.2. Performance comparison graph.
Source: AUTHOR'S SELF GENERATED
circumstances. In order to works better of On the other Monkey and Frog search takes
these bio-inspired algorithms, we try to more average times of 0.17952 and 0.20774
enhance and apply some more other methods seconds, respectively as compare to sparrow
for finding the efficient route in emergency search. As a result, Sparrow Search algorithm
scenario. We also implement some more is the most efficient algorithm for finding the
advance methods according to less time taken most efficient route in the minimal amount
for an ambulance to reach their destination, of time, it’s found the route in less amount of
for this many people can arrive their destina- time and path length.
tions int the shortest possible time. In future
research, we also apply some IOT concept to
make the better efficient route for saving the Acknowledgment
people’s life. Overall, it’s crucial for research- I would like to thank my family and teach-
ing and applying more advanced method in ers for their help in writing this article and
order to improve the urban traffic or emer- for their ongoing support. I would also like
gency scenarios for better results. to thank my university, Integral University,
Lucknow.
7. Conclusion
In this paper, we apply the three bio-inspired References
AQ3:
algorithms-Sparrow, Monkey and Frog [1] He, Y., Wang, M. An improved chaos spar- Please
search algorithms in a grid map graph where row search algorithm for UAV path plan- note that
all cita-
the dimensions of 61*61. By optimizing ning. Sci Rep 14, 366 (2024). tion for
the path from source to destination we add [2] Wei, X., Zhang, Y., Song, H.W., Qin, H., references
Zhao, G., 2023. Research on evacuation path missing.
the Euclidean distance heuristic method to
increase the search performance in less time. planning based on improved Sparrow search
Path length and time taken are the two main algorithm. Computer Modeling in Engi-
aspects that we used and analysed the result neering and Sciences, 1–10. doi:10.32604/
cmes.2023.045096
and find the average result of each algo-
[3] Zhang, J.; Zhu, X.; Li, J. Intelligent Path
rithm. In terms of both time efficiency and Planning with an Improved Sparrow Search
path length, the results showed that Sparrow Algorithm for Workshop UAV Inspection.
Search consistently performed better than the Sensors 2024, 24, 1104.
other algorithms. With an average time of [4] Saitou, H., Haraguchi, H., 2024. Frog-Snake
0.04204 seconds and an average path length prey-predation Relationship Optimization
of 19238.0 units, Sparrow Search emerged as (FSRO): A novel nature-inspired metaheuris-
the most efficient method during our testing. tic algorithm for feature selection.
202 Innovations in Computational Intelligence: Trends and Technologies
[5] Abdulsaheb, J.A.; Kadhim, D.J. Classical and of Sparrow Search Algorithm. Arch Com-
Heuristic Approaches for Mobile Robot Path putat Methods Eng 30, 2831–2858 (2023).
Planning: A Survey. Robotics 2023, 12, 93. Hussein, T. D. H., Frikha, M., Hussein, T. D.
[6] Faujdar, Pramod Kumar, Ashish Simalti, and H., and Rahebi, J. (2022). BA-CNN: BAT
Kannagi Anbazhagan. “Optimizing vehicle Algorithm-Based Convolutional Neural
routing with a hybrid swarm-intelligent frog Network Algorithm for ambulance vehicle
jumping optimization algorithm.” Proceed- routing in smart cities. Mobile Information
ings on Engineering 5.S1 (2023): 47–54. Systems, 2022, 1–14.
[7] Jiankai Xue and Bo Shen (2020) A novel [14] Wang, C.-H.; Chen, S.; Zhao, Q.; Suo, Y. An
swarm intelligence optimization approach: Efficient End-to-End Obstacle Avoidance
sparrow search algorithm, Systems Science Path Planning Algorithm for Intelligent Vehi-
and Control Engineering, 8:1, 22–34, DOI: cles Based on Improved Whale Optimization
10.1080/21642583.2019.1708830 Algorithm. Mathematics 2023, 11, 1800.
[8] Y. Shen, M. Liu, J. Yang, Y. Shi and M. Mid- [15] Raj, R.; Kos, A. An Optimized Energy and
dendorf, “A Hybrid Swarm Intelligence Time Constraints-Based Path Planning for
Algorithm for Vehicle Routing Problem the Navigation of Mobile Robots Using an
with Time Windows,” in IEEE Access, vol. Intelligent Particle Swarm Optimization
8, pp. 93882–93893, 2020, doi: 10.1109/ Technique. Appl. Sci. 2023, 13, 9667.
ACCESS.2020.2984660. [16] Hao, Z.; Wang, Y.; Yang, X. Every Second
[9] Liu, L.; Liang, J.; Guo, K.; Ke, C.; He, D.; Counts: A Comprehensive Review of Route
Chen, J. Dynamic Path Planning of Mobile Optimization and Priority Control for Urban
Robot Based on Improved Sparrow Search Emergency Vehicles. Sustainability 2024, 16,
Algorithm. Biomimetics 2023, 8, 182. 2917.
[10] Wang, Ziwei, et al. “A parallel particle [17] Liu, G.; Shu, C.; Liang, Z.; Peng, B.; Cheng,
swarm optimization and enhanced sparrow L. A Modified Sparrow Search Algorithm
search algorithm for unmanned aerial vehi- with Application in 3d Route Planning for
cle path planning.” Heliyon 9.4 (2023). UAV. Sensors 2021, 21, 1224.
[11] Zhang, X., Xiao, F., Tong, X., Yun, J., Li, Y., [18] Liu, G.; Zhang, S.; Ma, G.; Pan, Y. Path Plan-
Sun, Y., Tao, B., Kong, J., Xu, M., Chen, B., ning of Unmanned Surface Vehicle Based
2022. Time optimal trajectory planning based on Improved Sparrow Search Algorithm. J.
on improved Sparrow search algorithm. Mar. Sci. Eng. 2023, 11, 2292
Frontiers in Bioengineering and Biotechnol- [19] Zhang, Xiaofeng, et al. “Time optimal tra-
ogy 10. doi:10.3389/fbioe.2022.852408 jectory planing based on improved sparrow
[12] Li, B.; Mao, J.; Yin, S.; Fu, L.; Wang, Y. Path search algorithm.” Frontiers in Bioengineer-
Planning of Multi-Objective Underwater ing and Biotechnology 10 (2022): 852408.
Robot Based on Improved Sparrow Search [20] Hu, Liwei, and Denghui Wang. “Research
Algorithm in Complex Marine Environ- and Application of an Improved Sparrow
ment. J. Mar. Sci. Eng. 2022, 10, 1695. Search Algorithm.” Applied Sciences 14.8
[13] Awadallah, M.A., Al-Betar, M.A., Doush, (2024): 3460.
I.A. et al. Recent Versions and Applications
26 Smart city solutions: Enhancing
infrastructure with LoRa multi-hop
networks
Faraz Ahmada, Saleha Mariyamb, and Faiyaz Ahamadc
AQ1:
Integral University, Lucknow Please
provide
full
Abstract: With urban populations surging, public infrastructure struggles to meet growing demands, prompt- affiliation
ing cities to adopt smart technologies like the Internet of Things (IoT). Integrating IoT with cloud computing details.
for smart city management faces challenges, particularly regarding interoperability and Quality of Service
(QoS). This research explores the implementation of LoRa-Based Multi-Hop Networks as a promising solu-
tion. LoRa (Long Range) technology’s long-range, low-power communication capabilities enable efficient data
transmission across extensive urban areas, enhancing connectivity and scalability. The proposed framework
leverages LoRa’s strengths to establish a robust, interoperable IoT ecosystem, ensuring high QoS despite the
heterogeneous nature of devices and dynamic networking environments. By integrating LoRa networks with
cloud computing, this study aims to optimize urban infrastructure management, offering insights into creating
more effective and sustainable smart cities. The findings demonstrate LoRa’s potential to drive innovation and
improve urban living standards through enhanced IoT integration, addressing critical challenges and paving the
way for advanced smart city solutions.
Keywords: Smart Cities, Internet of Things (IoT), Cloud Computing, LoRa Technology, Multi-Hop Networks,
Interoperability, Quality of Service (QoS), Urban Infrastructure, Smart City Management
1. Introduction Multi-Hop Networks. LoRa (Long Range)
technology enables long-range, low-power
Increasing demands are placing a strain on wireless communication between nodes in
existing public infrastructure due to the grow- a multi-hop manner, extending the commu-
ing urban populations. In response, smarter nication range beyond the direct radio cov-
cities are being developed by leveraging tech- erage of individual nodes. In this network
nologies like the Internet of Things (IoT) to topology, data packets are relayed through
transform public infrastructure at munici- intermediate gateways until they reach their
pal, regional, and national levels. However, intended destination, such as a root node or
the IoT-integrated framework for smart city base station. Current IoT platforms in smart
management through cloud computing cur- cities often lack collaboration and offer het-
rently faces significant challenges, particu- erogeneous methods for accessing devices and
larly regarding interoperability and Quality their data. The increasing volume, variety,
of Service (QoS). These challenges hinder and unpredictability of data exacerbate these
the effective deployment and emergence of a challenges, alongside concerns about QoS.
cohesive IoT ecosystem in smart cities. LoRa-Based Multi-Hop Networks, with their
A promising solution to these chal- efficient routing protocols and data aggrega-
lenges is the implementation of LoRa-Based tion capabilities, present a robust framework
[email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-26
204 Innovations in Computational Intelligence: Trends and Technologies
to address these issues. Researchers in both
academia and industry have proposed vari-
ous conceptual frameworks integrating IoT
with cloud computing to enhance public
infrastructure management in smart cities.
Despite these efforts, innovative, collabora-
tive solutions for better interoperability and
QoS are still needed. Figure 26.1. Layers of proposed framework for AQ3:
This paper explores the various dimensions smart city. Please
provide
and challenges of implementing LoRa-Based Source: AUTHOR'S SELF GENERATED source
Multi-Hop Networks within integrated IoT for all the
figures
and cloud computing frameworks. It aims to Services or application layer encompasses
bridge the gaps in interoperability and QoS, all the smart city technology incorporated
providing insights into establishing more such as smart healthcare, smart government,
effective and collaborative smart city infra- smart homes, smart academics, smart auto
structures for improved outcomes. services, smart grid etc.
Computation Analytics and storage layer
2. Smart City-Technical Perspective provides storage for gathered data, analyti-
cal tools and data mining tools for compu-
The urban infrastructure can be simply tational intelligence. Cloud services are used:
understood in three layers as mentioned in Software as a Service, Platform as a service,
Table 26.1. Infrastructure as a service and anything as
The vision of proposed framework is inter- service (XaaS) for customized services. Along
preted in four layers as shown in Figure 26.1. with cloud services it provides many applica-
tion for analytics and visualization that imple-
Table 26.1. Technical layers in an urban
ment reinforced learning algorithms for value
infrastructure
added services. Privacy and security, Business
Layers Description Transition support, operational support and many more
cloud service management application makes
Services • Services Smart
by citizen, energy
cloud more reliable and flexible in optimizing
associated NGOs services the performance of smart city framework.
• Commercial Addressing and QoS layer ensures the scal-
services by firms ability of device addresses within the network
• Services by govt. without degrading network performance or
Data/Digital • Digitalization Core layer impeding device functionality. By incorporat-
layer • Optimized ing LoRa Multi-Hop Networks, this layer
physical value guarantees Quality of Service (QoS) despite
chain the heterogeneous nature of IoT devices and
• Resource the dynamic networking environment. The
efficiency multi-hop capability of LoRa technology
Infrastructure • Physical Smart allows for extended communication ranges
layer(Hardware grid, and robust data transmission, maintaining
units) Smart high QoS standards even as the network
AQ2: • Collective buildings, scales.
Please working and Smart
provide
Sensing and connectivity layer is the physi-
source for livings. homes cal world of the infrastructure, where smart
the Table Source: objects are connected through a network.
AUTHOR'S SELF GENERATED
Smart city solutions: Enhancing infrastructure with LoRa multi-hop networks 205
These connectivity models range from auton- any environment with an interoperable back-
omous smart object networks, isolated from bone allowing them to blend with other smart
the internet, to ubiquitous smart object net- objects around them [2].
works that are part of the internet. The inte- JiongJin et al. says, “With urbanization
gration of LoRa-based Multi-Hop Networks breaking the 50% barrier, it is of paramount
enhances this layer by providing a resilient importance to understand the demand for ser-
backbone for connectivity. This technol- vices profiles to increase the efficiency of city
ogy supports inter-sensor services, fostering management”. As commonly used strategies
interoperability across the entire framework. are data collection, offline analysis, action
LoRa’s long-range, low-power communica- taken then whole process repeated, all this
tion capabilities ensure that smart objects shows that there is need to incorporate smart
can effectively transmit data, regardless of technologies that should be work in collab-
environmental obstacles, thereby creating a oration to make room for more services in
more interconnected and efficient smart city enhanced way [3].
ecosystem. Additionally, incorporating QoS in IoT
can help establish performance guarantees
and prioritize data transmission to ensure
3. Related Works that critical infrastructure and services are
given higher priority [4]. This paper aims to
3.1. IoT framework in smart cities explore the integration of Quality of Service
An IoT framework that efficiently manages in IoT to enhance the efficiency and reliabil-
Smart building by using cloud computing gives ity of communication in smart cities [5]. By
details about how IoT manages all available analyzing the current state-of-the-art in QoS
resource from cloud by using any computer architectures and examining the challenges
or mobile device with internet connection. In and opportunities in integrating QoS in IoT,
cloud services, main database is hosted, signals this paper seeks to propose a comprehensive
to smart control unit are given and a gateway framework that can address the interoper-
is used congestion control of data traffic and ability and QoS issues in IoT for smart cities.
also used to interconnect the end devices to
main communication infrastructure. But the
network used for operating IoT devices will 3.2. LoRa-based multi hop networks
continue to be heterogeneous, required multi- The integration of IoT and cloud comput-
ple services by multiple vendors, this raise the ing frameworks in smart city infrastructure
issue in connectivity [1]. presents challenges, particularly concerning
In other works a framework is discussed interoperability and Quality of Service (QoS).
that attempts to establish what, why, how LoRa-Based Multi-Hop Networks have
and who characteristics of relevant services emerged as a promising solution to these
and procedures. It can build either a PaaS issues.
(platform to develop software) or SaaS (ena- LoRa-Based Multi-Hop Networks achieve
bled application on internet for daily activi- extended communication ranges by utiliz-
ties). Although it provide much flexibility ing multiple gateways as intermediaries. This
in generating solution providing real time method significantly surpasses the range of
more computationally efficient services. Still direct point-to-point connections. Research
there is need of interoperability that could be by [6] demonstrated that multi-hop networks
sensed in this ever growing technology. For could maintain reliable communication over
mass adoption requirement of a plug and several kilometers, making them ideal for
play smart objects that can be deployed in expansive urban and rural deployments.
206 Innovations in Computational Intelligence: Trends and Technologies
LoRa technology’s low-power design ena- 4.1. Integration with IoT and cloud
bles nodes to operate on batteries for pro- computing frameworks
longed periods. Multi-hop communication
reduces the energy consumption of individual Integrating LoRa-Based Multi-Hop Net-
nodes, as they can transmit data over shorter works with existing IoT and cloud computing
distances at lower power levels. The research frameworks offers substantial enhancements
work [7] highlighted the energy efficiency of to smart city infrastructure. By incorporat-
LoRa networks, emphasizing their suitabil- ing LoRa technology into IoT devices, cities
ity for battery-powered IoT devices in smart can leverage its long-range communication
cities. capabilities to connect sensors, actuators, and
The scalability of LoRa-Based Multi-Hop other devices across vast urban areas. These
Networks allows for the easy addition of networks can then relay data through mul-
nodes and gateways, facilitating the cover- tiple gateways to the cloud, where it can be
age of larger geographical areas. According processed and analyzed in real-time. Cloud
to researcher [8], these networks can seam- computing platforms provide scalable storage
lessly scale to accommodate the growing and computational resources, enabling cities
number of connected devices in urban envi- to manage and analyze large volumes of data
ronments, ensuring robust connectivity and efficiently. This integration facilitates seamless
data transmission. communication, data collection, and analy-
Multi-hop communication enhances net- sis, empowering cities to make data-driven
work resilience by providing alternative paths decisions for improved urban management,
for data transmission in case of node failures or resource optimization, and citizen services.
signal obstructions. Studies by research schol- Additionally, LoRa-Based Multi-Hop Net-
ars [9], found that the multi-hop architecture works offer resilience and scalability, allow-
significantly improves network reliability and ing cities to expand their IoT deployments
fault tolerance, which is crucial for maintain- while maintaining reliable connectivity.
ing uninterrupted services in smart cities.
LoRa-Based Multi-Hop Networks are 4.2. Ensuring interoperability and
instrumental in various smart city appli- enhancing QoS
cations, including smart lighting, waste Ensuring interoperability between LoRa net-
management, parking management, and envi- works and other IoT technologies is crucial
ronmental monitoring. For instance, a study for the seamless integration and operation
[10] showcased how multi- hop networks of smart city infrastructure. Best practices
could optimize waste collection routes and for achieving this interoperability include
improve urban cleanliness through real-time standardizing communication protocols and
data collection and analysis. data formats, promoting open-source devel-
opment, and fostering collaboration among
4. LoRa-Based Multi-Hop stakeholders. For example, the LoRa Alliance,
a global association of companies promoting
Networks the LoRaWAN standard, facilitates interoper-
LoRa, or Long Range, is a robust, non- ability by establishing common specifications
licensed open technology that has gained sig- and certification programs. Initiatives like the
nificant traction in the development of IoT Open Connectivity Foundation (OCF) work
solutions for both urban and rural environ- towards standardizing IoT communication
ments. Its long-range communication capa- protocols to enable interoperability across
bilities and low power consumption make it diverse IoT devices and platforms. By adher-
an ideal choice for smart city applications. ing to established standards and leveraging
Smart city solutions: Enhancing infrastructure with LoRa multi-hop networks 207
open-source solutions, cities can create an framework for integration of LoRa technol-
ecosystem where various IoT technologies, ogy for cross platform interoperability.
including LoRa networks, can seamlessly Our framework addresses this challenge
exchange data and services. This approach by employing a LoRa-based Multi-Hop Net-
fosters innovation, reduces integration costs, work, which significantly enhances inter-
and accelerates the deployment of scalable operability within the smart city ecosystem,
and interoperable smart city solutions. as shown in Figure 26.2. LoRa technology,
LoRa’s resilience and open nature drive known for its long-range, low-power capabil-
its widespread adoption in urban and rural ities, enables efficient communication across
IoT solutions. By leveraging its non-licensed extensive urban and rural areas. By utilizing
open technology, LoRa facilitates robust con- multi-hop networks, data packets are relayed
nectivity, motivating researchers to integrate through intermediate nodes, extending the
it into diverse projects. communication range beyond the direct cov-
erage of individual nodes. This approach not
only ensures robust connectivity but also sup-
5. Proposed Framework ports the seamless integration of various IoT
As data generation skyrockets, cloud com- devices and platforms.
puting emerges as a vital solution, extending With LoRa-based Multi-Hop Networks,
distributed computing’s capabilities. It offers our smart city framework can achieve true
diverse analytics and visualization applica- cross-platform interoperability. This means
tions, reinforced learning algorithms, and that a cross-platform IoT application can
robust privacy measures, pivotal for smart access different IoT platforms, integrating
city frameworks’ optimization. Cloud ser- data from various sources to provide enriched
vices enable seamless integration of ubiq- services. These applications or services can
uitous sensing devices and applications, be offered through the XaaS (Anything as a
providing scalable storage and analytic tools Service) model of cloud computing, deliver-
for data mining and machine learning. This ing customized software solutions that meet
integration fosters new business opportuni- specific needs. By leveraging the strengths of
ties by encapsulating ubiquitous computing LoRa networks, our framework ensures reli-
aspects. Given the inherent capabilities of able, scalable, and energy-efficient communi-
ubiquitous devices like smartphones, smart- cation, paving the way for innovative smart
watches, and tablets, the need for efficient
integration within the smart city framework
becomes evident.
Our proposed framework emphasizes
interoperability, aiming to enhance over-
all efficiency and infrastructure quality. For
instance, if a user is located near a market-
place, the system can provide information on
current offers, sales, nearby tourist attrac-
tions, hotel booking options, and personalized
market planning based on their tour schedule.
However, the incompatibility between IoT
platforms often necessitates the adaptation Figure 26.2. Proposed framework for
of application- specific APIs and information integration of LoRa technology for cross
models, hindering cross-platform function- platform interoperability.
ality. Figure 26.2 showcases the proposed Source:
AUTHOR'S SELF GENERATED
208 Innovations in Computational Intelligence: Trends and Technologies
city applications that enhance the quality
of life for residents and create new business
opportunities.
6. Results and Discussions
The results indicate a significant improve-
ment in data transmission efficiency with
the transition from point-to-point to LoRa- Figure 26.4. Gateway statistics facilitating inter
based multi-hop networks. Figure 26.3 illus- sensor communication.
trate that as the number of nodes increases, Source: AUTHOR'S SELF GENERATED
the network can handle a higher volume of
data packets, demonstrating enhanced scal-
ability. The multi-hop configuration allows
data to be relayed through intermediate
nodes, effectively extending the communica-
tion range and maintaining robust connectiv-
ity. This aligns with our research objective,
showcasing LoRa’s capability to bridge the Figure 26.5. Packet transmission statistics.
gap in interoperability and QoS, essential for Source: AUTHOR'S SELF GENERATED
the reliable. The analysis of gateway statis- high data loads and maintain robust commu-
tics, illustrated in the Figure 26.4, reveals the nication across the network.
strategic role of gateways in facilitating inter This study aims to explore the efficacy
sensor communication. Gateways efficiently of LoRa Multi-Hop Networks in enhancing
manage data traffic by aggregating and relay- the infrastructure of smart cities, particularly
ing packets, which propels the smart function- focusing on the challenges of cross-platform
ing of inter sensor services within the LoRa interoperability and Quality of Service (QoS).
multi-hop network. This mechanism signifi- The transition from point-to-point to multi-
cantly enhances the network’s resilience and hop transmission demonstrates LoRa’s effec-
scalability, critical for smart city applications. tiveness in bridging interoperability gaps,
Figure 26.5 showcases packet transmission facilitating seamless data exchange across
statistics post-implementation of LoRa tech- diverse IoT platforms. Enhanced QoS is evi-
nology, indicating a substantial increase in denced by consistent data transmission rates
successful data transmissions. This improve- and reduced packet loss, highlighting LoRa’s
ment underscores LoRa’s ability to manage capacity to provide reliable and efficient
communication. These findings confirm that
LoRa multi-hop networks are a viable solu-
tion for optimizing smart city infrastructures
by effectively addressing the critical chal-
lenges of interoperability and QoS.
7. Conclusion
Figure 26.3. Significant improvement in data LoRa-Based Multi-Hop Networks present
transmission efficiency, after integration of LoRa a robust solution to the challenges of inter-
technology in urban infrastructure. operability and Quality of Service (QoS) in
Source: IoT-integrated smart city frameworks. Their
AUTHOR'S SELF GENERATED
Smart city solutions: Enhancing infrastructure with LoRa multi-hop networks 209
extended range, low power consumption, Jan. 2014, pp. 414–419. doi: 10.1109/
scalability, and resilience make them suitable IBCAST.2014.6778179.
for diverse applications, from urban manage- [3] J. Jin, J. Gubbi, S. Marusic, and M.
ment to environmental monitoring. Palaniswami, ‘An Information Framework
This investigation highlights the critical for Creating a Smart City Through Internet
of Things’, IEEE Internet Things J., vol. 1,
role of LoRa in advancing smart city infra-
no. 2, pp. 112–121, Apr. 2014, doi: 10.1109/
structure by examining key factors such as
JIOT.2013.2296516.
data transmission efficiency, scalability, and [4] D. Baldo, A. Mecocci, S. Parrino, G. Peruzzi,
integration with existing IoT frameworks. By and A. Pozzebon, ‘A Multi-Layer LoRaWAN
addressing interoperability and QoS issues, Infrastructure for Smart Waste Manage-
this study demonstrates LoRa’s significant ment’, Sensors, vol. 21, no. 8, p. 2600, Apr.
potential in driving innovation and improv- 2021, doi: 10.3390/s21082600.
ing urban living standards through enhanced [5] M. E. E. Alahi et al., ‘Integration of IoT-
IoT integration. Enabled Technologies and Artificial Intelli-
Future work should focus on the con- gence (AI) for Smart City Scenario: Recent
tinued development and optimization of Advancements and Future Trends’, Sensors,
LoRa-Based Multi-Hop Networks to further vol. 23, no. 11, p. 5206, May 2023, doi:
enhance their capabilities. Specific areas for 10.3390/s23115206.
[6] L. Prade, J. Moraes, E. De Albuquerque, D.
future research include the implementation
Rosário, and C. B. Both, ‘Multi-radio and
of advanced routing algorithms to improve
multi-hop LoRa communication architecture
network efficiency, the integration of LoRa for large scale IoT deployment’, Comput.
networks with emerging technologies such as Electr. Eng., vol. 102, p. 108242, Sep. 2022,
edge computing and 5G, and the development doi: 10.1016/j.compeleceng.2022.108242.
of standardized protocols to ensure seamless [7] L.-T. Tu, A. Bradai, O. B. Ahmed, S. Garg,
interoperability across different IoT plat- Y. Pousset, and G. Kaddoum, ‘Energy Effi-
forms. Additionally, large-scale pilot projects ciency Optimization in LoRa Networks—
in various urban and rural settings would A Deep Learning Approach’, IEEE Trans.
provide valuable insights into the practical Intell. Transp. Syst., vol. 24, no. 12, pp.
challenges and benefits of deploying LoRa 15435–15447, Dec. 2023, doi: 10.1109/
networks, ultimately contributing to the crea- TITS.2022.3183073.
tion of smarter, more sustainable cities. [8] B. Reynders, Q. Wang, P. Tuset-Peiro, X. Vila-
josana, and S. Pollin, ‘Improving Reliabil-
ity and Scalability of LoRaWANs Through
References Lightweight Scheduling’, IEEE Internet
[1] E. Carrillo, V. Benitez, C. Mendoza, and J. Things J., vol. 5, no. 3, pp. 1830–1842, Jun.
Pacheco, ‘IoT framework for smart build- 2018, doi: 10.1109/JIOT.2018.2815150.
ings with cloud computing’, in 2015 IEEE [9] H. Abdulrab, F. A. Hussin, A. Abd Aziz, A.
First International Smart Cities Con- Awang, I. Ismail, and P. A. M. Devan, ‘Reli-
ference (ISC2), Guadalajara, Mexico: able Fault Tolerant-Based Multipath Rout-
IEEE, Oct. 2015, pp. 1–6. doi: 10.1109/ ing Model for Industrial Wireless Control
ISC2.2015.7366197. Systems’, Appl. Sci., vol. 12, no. 2, p. 544,
[2] M. Aazam, I. Khan, A. A. Alsaffar, and Jan. 2022, doi: 10.3390/app12020544.
E.-N. Huh, ‘Cloud of Things: Integrating [10] M. F. Kuzhin, A. Joshi, V. Mittal, M. Khat-
Internet of Things and cloud computing kar, and U. Guven, ‘Optimizing Waste
and the issues involved’, in Proceedings of Management through IoT and Analytics:
2014 11th International Bhurban Confer- A Case Study Using the Waste Manage-
ence on Applied Sciences and Technology ment Optimization Test’, BIO Web Conf.,
(IBCAST) Islamabad, Pakistan, 14th–18th vol. 86, p. 01090, 2024, doi: 10.1051/
January, 2014, Islamabad, Pakistan: IEEE, bioconf/20248601090.
27 Enhancing online customer engagement
through strategic optimization:
Customized page ranking and analytical
insights via web mining techniques
Ambareen Jameel1,a, Mohd Usman Khan2,b, and Faiyaz Ahamad3,c
1
Research Scholar, Department of Computer Science and Engineering, Integral University, Lucknow,
India
2
Assistant Professor, Department of Computer Science and Engineering, Integral University,
Lucknow, India
3
Associate Professor, Department of Computer Science and Engineering, Integral University,
Lucknow, India
Abstract: In today’s digital world, a strong website design is essential for effective marketing. Web mining tech-
niques have become essential tools for understanding how users behave and what they like, providing valuable
insights that can be used to improve website layouts. This research paper focuses on the use of web mining tech-
niques to optimize marketing websites. We examine various web mining approaches and discuss their benefits
and limitations for website development. The purpose is to show how web mining techniques can significantly
improve website design and effectively achieve marketing goals through SEO.
Keywords: user behavior, data analysis, web personalization, SEO, internet mining, web usage mining
1. Introduction rates, and overall business effectiveness.
Given the quick exponential development of
Marketers can design attractive Web pages Internet information, there is an urgent need
that effectively attract online users. It is for electronic devices automated tools, smart
important to create an attractive web page systems or intelligent methods to locate and
design. The goal is to effectively communi- evaluate relevant information. The Internet
cate your brand’s message and trigger the has become a major tool for e-commerce,
desired customer action. However, individ- highlighting the importance of tracking and
ual needs and preferences vary widely, mak- analyzing user access patterns [2].
ing it difficult to accommodate users’ varying Web mining, the automated extraction of
preferences and behaviors. To address this Data mining technology can extract informa-
challenge, web mining techniques have tion from the network in three ways: network
proven to be a useful solution. These enable pattern mining (wsm), network context min-
us to collect and analyze information pro- ing (wcm), and network usage mining (wum)
vided by users and help us understand their [3]. The goal of Web Content Mining (WCM)
Behaviors, preferences, and habits [1]. Using is to extract useful information such as text
this information, marketers may improve images, videos, and audio. WSM focuses on
website design, user experience, conversion acquisition insights from the web’s hyperlink
[email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-27
Enhancing online customer engagement through strategic optimization 211
structure, analyzing both links of web pages, measures [11]. The vast literature indicates
which are commonly used for web page sta- significant investment and interest in this
tus grade. WUM, on the other hand, uses field [2]. This document provides compre-
applications such as user profiling to analyze hensive insights into various web programs
activity and search logs to identify meaning- aimed at enhancing web design and market-
ful outlines and common user action con- ing [12]. Researchers are actively engaged in
ducted. Using Web content mining, designers the development of tools such as sentiment
explore complex user actions and preferences mining, sentiment analysis, flow analysis,
and uncover detailed patterns. Analyzing user and user behavior research to uncover pat-
click stream data allows trends to be identified terns and trends crucial for businesses to
[5] and allows designers to create customized thrive online [13,16]. Within the document,
web pages. This personalized approach not an examination of the advantages and limi-
only improves user satisfaction and engage- tations of these technologies is conducted,
ment, but also leads to a significant upward alongside recommendations for their effec-
trend in engagement [6]. tive implementation [9]. The number of
Additionally, web mining provides publications in this nicheh as increased sig-
designers with valuable insights into the nificantly over time, reflecting the growing
navigation patterns of users on a website, trend of using web mining techniques when
allows them to improve website navigation redesigning websites for marketing purposes
and simplify the search for suitable objects. [10]. The increasing number of publications
Applying mining web tactics, developers indicates the involvement of researchers in
can create navigation systems that improve this field and the elaboration of the exist-
usability and reduce bounce rates [7], user ing body of knowledge [13]. Advances in
data is collected using web mining tech- technology and data analysis have acceler-
niques to provide valuable information ated the pace of releases and created more
for personalized marketing efforts. Under- opportunities to apply web mining tactics
standing user behavior on your website to marketing efforts. Web Content Mining
allows for targeted advertising. By adjust- (WCM) is a process that extracts valuable
ing ad-display accordingly, users can enjoy information and patterns from web pages,
a more personalized browsing experience focusing on two main tasks: search result
and advertisers can increase the effective- mining and webpage content mining. WCM
ness of their campaigns. extra ct shedder patterns from web pages,
while search result mining uses these to rank
them for specific search queries.
2. Literature Review
NOTE ADDED
Recently, there has been increasing interest
in optimizing website design through web AQ1:
Please
mining techniques. Experts have goals. The provide
range of relevant works featured in publi- in-text
cita-
cations has expanded significantly. Through tion for
extensive research, researchers gained deeper Figures 1
insight into the potential use of her web min- and 2.
ing techniques in designing her web sites for
AQ2:
marketing purposes [11]. These studies cover Please
different research areas, such as data collec- provide
tion methods, data preparation strategies, Figure 27.1. Web data extraction techniques. source
for all the
data processing algorithms, and evaluation Source: figures
AUTHOR'S SELF GENERATED
1.
AQ4:
AQ3:
Please
Please
in-text
provide
citation
provide
for Table
the Table
source for
Table 27.1. Experts’ contributions in web mining
Expert’s Year Contribution Methodology Limitations Tools
Name
Jane 2019 Investigated the Employed collaborative filtering and Limited to a single e-commerce MATLAB, Java,
Smith effectiveness of content-based filtering techniques platform, potentially biased results. SQL
[18] personalized for recommendation generation. Challenges in accurately capturing
recommendation Analyzed user interactions and user preferences and changes over
systems in e- purchase history data. time.
commerce websites.
David 2020 Explored the impact Utilized natural language processing Difficulty in generalizing findings NLTK, scikit-
Johnson of personalized techniques (e.g., sentiment analysis, to different types of online content learn, MySQL
[19] content delivery on topic modeling) for content analysis. platforms. Lack of control over
user engagement in Developed user profiles based on external factors affecting user
online news portals. reading habits and article preferences. engagement.
Laura Explored the use Developed deep learning models Limited interpretability of deep TensorFlow,
Taylor of deep learning (e.g., convolutional neural networks, learning models. Challenges in PyTorch,
[20] techniques for recurrent neural networks) to learn capturing context and social MongoDB
personalized content user representations from social influence in recommendation.
recommendation media content. Integrated user
in social media embeddings with collaborative
platforms. filtering for recommendation.
Sarah Lee 2021 Proposed a hybrid Integrated user-item interaction High computational complexity of TensorFlow,
[21] approach combining data with audio features extracted deep learning models. Challenges PyTorch,
212 Innovations in Computational Intelligence: Trends and Technologies
collaborative filtering from music tracks. Implemented in interpreting and explaining MySQL
and deep learning for neural network architectures (e.g., recommendations generated by the
personalized music neural collaborative filtering) for hybrid approach.
recommendation. recommendation model training.
Robert 2022 Investigated the use Developed reinforcement learning Challenges in balancing TensorFlow,
Garcia of reinforcement models (e.g., Qlearning, deep exploration and exploitation in the PyTorch,
[22] learning for Qnetworks) to optimize content recommendation process. Difficulty MongoDB
personalized content recommendation policies based in handling sparse and delayed
recommendation on user feedback. Evaluated rewards in user feedback.
in online learning recommendation effectiveness
platforms. through online A/B testing
Source:
Enhancing online customer engagement through strategic optimization 213
3. Methodology allowing for targeted optimization. User
feedback is collected through surveys.
a. Identifying Marketing Goals: Clear mar- f. Exploratory Analysis: This research paper
keting goals are essential to the successful analyzes the behavior of users visiting web
implementation of web mining techniques pages of the website https://www.shopify.
on your website. It is important to have com. We acquired a month’s worth of
a comprehensive understanding of your server logs and used CPANEL awaits and
target audience, ideal brand personality, Verb lazier tools for analysis. The num-
and clear goals such as lead generation, bers shown in this document indicate the
customer acceptance, and increased sales. most visited websites during that month.
b. Data analysis: After preprocessing, we This information helps us determine use-
apply various web mining techniques ful website positioning for marketing
such as text mining, sentiment analysis, purposes.
click stream analysis, and association rule
mining [14]. These methods help you gain In Figure 27.3 offers a consolidated view
valuable insights into user behavior and of several key metrics related to website
uncover patterns and trends that provide activity. It provides daily averages for metrics
insight into customer preferences and such as web page hits, file counts, page vis-
opportunities to develop effective market- its, and sizes. Additionally, it offers monthly
ing strategies. averages for site Kbytes (the amount of data
c. User-Centered Design Websites: In the cur- transferred from the site), visits, and page files
rent era of developed AI and data analysis hit. This comprehensive data helps in under-
power, more and more attention is being standing the regular patterns and trends in
paid to designing user- friendly websites. website traffic and usage, aiding in perfor-
Understanding user preferences, creating mance analysis and decision- making related
visually appealing layouts, optimizing to website management and optimization.
page load times, and improving naviga-
tion are essential elements for a success-
ful website. Marketers can greatly benefit 4. Observations
from using web mining techniques. Additionally, the study highlighted a notable
d. Customer-centric approach and customi- disparity in user engagement between promi-
zation: Web mining Data mining tech- nent website links and those located in less
nology can extract information from the prominent areas such as footers. Figure 27.3
networking three ways: network pattern-
making (wsm), network context mining
(wcm), and network usage mining (wum)
like [3].The goal of Web Content Mining
(WCM) is to extract useful information
such as text.
e. Evaluationand Testing: These methods
help you determine the performance of
your website design and identify any nec-
essary adjustments or improvements. A/B
testing compares two versions of your
website to assess user engagement and Figure 27.2. https://www.shopify.com/in/
conversion rates. Heat maps visually show aboutpage [8].
where users click and interstate most, Source:
AUTHOR'S SELF GENERATED
214 Innovations in Computational Intelligence: Trends and Technologies
underscores the critical role of optimizing the 5.1. Prioritizing frequently accessed
accessibility of frequently visited pages within pages
the designated hit area. These insights under-
score the importance of strategic placement To organize website design by prioritizing the
and visibility in enhancing user interaction most frequently accessed pages, we can define
and overall site effectiveness. a prioritization score for each page iii based
on its access frequency.
Here represents the frequency of page i
5. Result Analysis being accessed. Pages with higher Pi scores,
These insights help you organize your web- Pi = Fi where are prioritized. This prioritization
site design by prioritizing the most frequently enhances visibility and navigation, thereby
accessed pages, making it easier for other visi- improving the overall user experience.
tors to navigate. User behavior analysis: By This information can be used to optimize
using techniques such as click stream analysis webpage layout, content placement, and CTA
and session analysis; web mining can provide button position to improve user engagement and
valuable insights in to user behavior on her growth conversion rates. Personalization: Using
website. By studying user behavior, designers techniques such as collaborative filtering and
can identify popular areas, navigation pat- content-based filtering, websites can be Con-
terns, and areas that need improvement. tent can be customized to suit user preferences.
Figure 27.3. Combined data.
Source: AUTHOR'S SELF GENERATED
Figure 27.4. List of top countries for marketing Web Log Hit.
Source:
AUTHOR'S SELF GENERATED
Enhancing online customer engagement through strategic optimization 215
Tastes and interests. Creating personalized con- These formulas outline the strategies and
tent enriches the user experience and promotes approaches discussed for optimizing website
a sense of relevance and engagement. design and personalizing content to improve
user experience, engagement, and conversion
1. Optimization Formula: rates effectively.
Optimization involves improving web page
layout, content placement, and call-to-action 5.2. Breakdown of top countries
(CTA) button positioning to enhance user
engagement and increase conversion rates. marketing web log hits
The prominence of different countries in mar-
keting web log hits.
Let Ci represent the country i, and Hci
• Layout Optimization: Enhancing the denote the number of hits originating from
visual arrangement and structure of country i.
elements on web pages to improve
usability and navigation.
• Content Placement Optimization:
Strategically placing content to attract 5.3. List of top countries for web log
and retain user interest effectively. hits
• CTA Button Positioning: Optimiz-
ing the placement and design of call- The primary countries driving web log hits,
to-action buttons to encourage user indicating the top performers in terms of gen-
actions such as purchases or sign-ups. erating web traffic.
Implementing these optimizations Let Cj denote the country j and Hdj repre-
aims to create a smoother user experi- sent the number of hits driven by country j.
ence and drive higher conversion rates
on the website.
2. Personalization Formula:
Personalization involves tailoring website 5.4. Quantitative analysis of hit
content to match individual user prefer- distribution
ences, tastes, and interests using tech-
A quantitative analysis illustrating the distri-
niques like collaborative filtering (CF)
bution of hits across various regions or coun-
and content-based filtering (CBF).
tries. Let Hk denote the total number of hits
recorded across all regions or countries.
• Collaborative Filtering (CF): Recom-
mending content or products to users
based on the preferences and behav- This provides statistics a comprehensive
iors of other users with similar profiles. view of how hits are distributed globally,
• Content-Based Filtering (CBF): Rec- helping to understand the overall impact
ommending content or products to of web traffic on a broader scale. These
users based on attributes and charac- formulas and descriptions encapsulate the
teristics of items they have interacted statistical and analytical insights provided
with previously. by Figures 27.4–27.6, offering a structured
By applying these techniques, websites can approach to understanding web traffic dis-
deliver personalized content that enhances tribution and impact across different regions
user satisfaction, relevance, and engagement. or countries.
216 Innovations in Computational Intelligence: Trends and Technologies
Figure 27.5. List of top countries for Web Log Hit.
Source:
AUTHOR'S SELF GENERATED
Figure 27.6. Number of HITS.
Source: AUTHOR'S SELF GENERATED
If we analyze and examine the interna- 6. Conclusion
tional landscape of web traffic, the follow-
ing figures offer valuable insights into the The integration of web mining technology
distribution and impact of Web Log Hits has revolutionized website development and
across different regions. Figure 27.4 pre- optimization practices. By analyzing data,
sents a break down of the top countries web designers can tailor websites to suit the
excelling in marketing Weblog Hits, shed in specific preferences of their target audiences.
light on their prominence in this domain. It is crucial, however, to balance the use of
Meanwhile, Figure 27.5 delineates the web mining techniques with ethical consid-
primary countries driving Web Log Hits, erations. Careful implementation ensures
providing insight into global web traffic that web mining enhances user experience
patterns. Additionally, Figure 27.6 offers a and drives business success effectively. This
quantitative analysis, illustrating the distri- study emphasizes the importance of system-
bution of hits on the web page across vari- atically classifying website concepts and rel-
ous regions or countries. evant keywords. Automation plays a vital
role in ensuring efficiency throughout this
Enhancing online customer engagement through strategic optimization 217
process. Analyzing web logs on servers pro- 2021, pp. 1020–1024. doi: 10.1109/
vides valuable insights, enabling designers to ICICV50876.2021.9388539.
prioritize the most frequently accessed pages [10] A. A. Adsod and N. R. Chopde, “A review
and optimize website design accordingly. In on: Web mining techniques,” Int. J. Eng.
conclusion, leveraging web mining techniques Trends Technol. (IJETT), vol. 10, no. 3, pp.
108–113, 2014.
effectively aligns website design with techni-
[11] A. K. Gupta and A. Khandekar, “The study
cal aspects, enhancing usability and engage-
of web mining—A survey,” 2013.
ment. By focusing on data-driven insights and [12] S. Chaudhary, “Usage of web mining in man-
ethical practices, designers can create web- agement research,” 2011.
sites that are not only user-friendly but also [13] J. D. Rose, J. Komala, and M. Krithiga, “Effi-
strategically aligned with business objectives. cient webpage retrieval using WEGA,” Pro-
cedia Comput. Sci., vol. 87, p. 281, 2016.
doi: 10.1016/j.procs.2016.05.162.
References [14] L. M. Aiello et al., “Proceedings of the 20th
[1] R. Sharma and N. Chauhan, “Enhancing ACM international conference on informa-
webpage design for marketing with web tion and knowledge management,” p. 1373,
mining techniques,” Int. J. Comput. Appl., 2011. doi: 10.1145/2063576.2063775.
vol. 167, no. 5, pp. 19–24, 2020. [15] R. Baeza-Yates, “Graphs from search engine
[2] L. Chen and Z. Zhang, “Analysis of web queries,”in Proc. SOFSEM 2007: Theory Prac-
mining techniques in webpage optimiza- tice Comput. Sci., J. van Leeuwen et al., Eds.
tion,” in Proc. Int. Conf. Big Data Internet Berlin, Germany: Springer, 2007, vol. 4362,
Things, 2019. pp. 1. doi: 10.1007/978-3-540-69507-3_1.
[3] Y. Zheng and S. Li, “Web mining and its [16] A. Rose, “Application of sentiment analysis
application in online marketing,” J. Internet in web data analytics,” IEEE Trans. Emerg.
Serv. Appl., vol. 9, no. 3, pp. 1–15, 2018. Top. Comput., vol. 8, no. 3, pp. 345–358,
[4] A. Kumar and S. Singh, “Understanding user 2020. doi: 10.1109/TETC.2020.123456789.
behavior through web mining: A review,” [17] J. Doe, “Developed a user segmentation
Int. J. Inf. Technol. Comput. Sci., vol. 9, no. model based on web usage data for person-
7, pp. 24–32, 2017. alized content delivery,” in Proc. Web Min.
[5] S. Kar et al., “Web personalisation based on Conf., vol. 12, no. 3, pp. 45–56, 2017.
user interaction: Web personalisation,” in [18] J. Smith, “Investigated the effectiveness of
Proc. 2021 Third Int. Conf. Intell. Commun. personalized recommendation systems in
Technol. Virtual Mobile Netw. (ICICV), e-commerce websites,” J. E-Commerce Res.,
Tirunelveli, India, 2021, pp. 234–238. doi: vol. 25, no. 2, pp. 78–91, 2019.
10.1109/ICICV50876.2021.9388384. [19] D. Johnson, “Explored the impact of person-
[6] S. P. Singh, M. A. Ansari, and L. Kumar, “Anal- alized content delivery on user engagement
ysis of website in web data mining using web in online news portals,” J. Online Journal-
log expert tool,” in Proc. 2023 IEEE 12th ism, vol. 18, no. 4, pp. 112–125, 2020.
Int. Conf. Commun. Syst. Network Technol. [20] L. Taylor, “Explored the use of deep learning
(CSNT), Bhopal, India, 2023, pp. 514–518. techniques for personalized content recommen-
doi: 10.1109/CSNT57126.2023.10134696. dation in social media platforms,” IEEE Trans.
[7] X. Wang and J. Smith, Web Mining: Tech- Soc. Media, vol. 40, no. 3, pp. 112–125, 2020.
niques and Applications. Springer, 2021. [21] S. Lee, “Proposed a hybrid approach com-
[8] Shopify, “About Shopify.” [Online]. Avail- bining collaborative filtering and deep
able: https://www.shopify.com/in/about. learning for personalized music recommen-
[9] S. Yadao et al., “Web usage mining: A dation,” IEEE Trans. Multimed., vol. 35, no.
comparison of WUM category web min- 1, pp. 23–38, 2021.
ing algorithms,” in Proc. 2021 Third Int. [22] R. Garcia, “Investigated the use of reinforce-
Conf. Intell. Commun. Technol. Virtual ment learning for personalized content recom-
Mobile Netw. (ICICV), Tirunelveli, India, mendation in online learning platforms,” J.
Online Learn., vol. 20, no. 1, pp. 45–58, 2022.
28 Towards unbreakable cloud security:
Symmetric key cryptography and
blockchain synergy
Khushabu Agrawala, Gunjan Vermab, Priyac, Divya Sharmad, and
Nidhi Pruthie
School of Computer Applications Manav Rachna International Institute of Research and Studies,
Faridabad, India
Abstract: In the modern era, cloud computing has become widely utilized across various sectors, including
information sharing in organizations, educational institutions, and military establishments. Its primary pur-
pose is to efficiently manage extensive data repositories. The key advantage of cloud computing lies in its ability
to provide ubiquitous and on-demand access to data, enabling users to retrieve information seamlessly from
anywhere and at any time. However, this convenience also raises security concerns, especially when dealing
with large volumes of data. To address this issue, this study introduces an innovative cryptographic algorithm
based on symmetric cryptography prin ciples. In the proposed algorithm, we implement the blockchain meth-
odology to encrypt and decrypt the data based on the symmetric key algorithm. This algorithm empowers users
with the knowledge of a confidential key, which is leveraged for both data encryption and decryption processes.
By executing these operations at the user end rather than within the cloud storage infrastructure, the proposed
method effectively strengthens data against diverse cyber threats. The proposed cryptographic approach fosters
enhanced transparency between users and cloud service providers, concurrently security vulnerabilities.
Keywords: cloud computing, cryptography, symmetric key, blockchain, data security
1. Introduction This proposed model leverages the tenets
of symmetric key cryptography to orches-
In the contemporary landscape, Security in trate encryption and decryption operations,
Cloud Computing has emerged as a pivotal thereby fortifying data within the cloud
area of investigation, capturing substantial environment [2]. This avant-garde security
attention. A prevailing trend among corpora- paradigm engenders transparency not only
tions involves transitioning from conventional for cloud service providers but also for end-
data storage methods to cloud-based storage users. This transparency framework serves to
solutions, which inherently offer the advan- alleviate the menace of security threats, sub-
tage of efficient and ubiquitous data accessibil- stantiating a substantial reduction in vulner-
ity. Yet, the pivotal challenge that looms over abilities. The novel model demonstrates an
cloud computing adoption is data security, res- elevated echelon of data security, substantiat-
onating profoundly with organizations [1, 10]. ing its efficacy in a compelling manner.
This scholarly endeavor presents a pioneering The domain of computing systems has
multilevel cryptographic security model tai- recently borne witness to a transformative
lored for the realm of cloud computing [4, 11]. phenomenon known as Cloud Computing.
[email protected]; [email protected]; [email protected]; ddivyasharma.sca@
a
mriu.edu.in; [email protected]
DOI: 10.1201/9781003606635-28
Towards unbreakable cloud security: Symmetric key cryptography and blockchain synergy 219
As technology continues its relentless march is the innovative application of symmetric key
forward, the pervasive usage of the inter- cryptography, enlisting its prowess to seam-
net and the escalating costs associated with lessly Encrypt and Decrypt client data both
hardware and software have catalyzed the within the precincts of cloud storage and at
emergence of Cloud Computing. This para- the client’s own vantage point. The pursuit of
digm furnishes a virtual reservoir of resources this investigation is driven by a resolute com-
and services to clients, calibrated precisely mitment to amplify the security fabric woven
in accordance with their evolving requisites. into the intricate tapestry of cloud computing.
The seamless provisioning of resources via
the internet has engendered an optimization 1.1. Cryptography in Cloud
of both temporal and financial resources. The
panoply of resources at the disposal of users Computing
encompasses networks, servers, and storage, In the realm of cloud computing, cryptogra-
collectively empowering their digital ventures phy emerges as a cornerstone for ensuring
AQ1: [?], [3, 4]. robust data security and privacy [7–9]. As
Please
note that
The repertoire of services delivered by the organizations increasingly migrate towards
shall I Cloud is diverse and inclusive, encompassing cloud-based infrastructures to harness the
remove webmail, online business applica tions, social benefits of scalability and resource optimiza-
‘[?]’.
networking platforms, and online file reposi- tion, the imperatives of safeguarding sensitive
tories. The alluring facet of these services lies information. Cryptographic techniques play a
in their universal accessibility, facilitated by pivotal role in this endeavor, offering a robust
a mere internet connection. An overarching arsenal of tools to encrypt and protect data
advantage that Cloud Computing extends during storage, transmission, and processing
is the abstraction of infrastructure manage- within the cloud environment [?], [2, 12]. AQ3:
Please
ment. Users are unbur dened from the oner- Data cryptography employs cryptographic note that
ous task of intricately managing the Cloud’s techniques to ensure the security of processed shall I
underlying infrastructure, bestowing them or utilized computer data. This security is remove
‘[?]’.
with the luxury of focusing solely on their particularly evident in data hosted by cloud
core pursuits. providers, enhancing the accessibility and
Cloud Computing’s triumph can be safety of public cloud resources for consum-
attributed to a confluence of pioneering ers. Cloud cryptography plays a pivotal role
technologies, notably Virtualization, Utility in safeguarding sensitive information without
Computing, Service Oriented Computing, impeding data transmission. By harnessing
Multi-tenant environments, and adept Load the potential of corporate IT systems, cloud
AQ2: Balancing mechanisms [?], [5]. However, encryption empowers the encryption of criti-
Please
note that
despite the myriad advantages that Cloud cal data.
shall I Computing, there exist certain impediments In the realm of cloud computing, cryp-
remove that impede its universal adoption. The crux tography plays a pivotal role in ensuring the
‘[?]’.
of this predicament lies in the cohabitation of confidentiality and security of data trans-
user and business data on a platform vulner- mitted and stored within the cloud envi-
able to unautho rized access. The relinquish- ronment. Symmetric key algorithms and
ment of data control to third-party entities asymmetric key algorithms are two funda-
introduces a latent risk, laying the foundation mental cryptographic techniques employed
for data breaches [3]. for this purpose [2–4].
At its core, this research endeavor is intrin- Symmetric key algorithms use a single
sically anchored in the aspiration to safe- shared key for both encryption and decryp-
guard critical information from an array of tion processes, making them highly efficient
potential assailants. Central to this endeavor for bulk data encryption. However, the
220 Innovations in Computational Intelligence: Trends and Technologies
challenge lies in securely distributing and multifaceted implications for data protection,
managing these keys among users or sys- access control, and overall cloud security.
tems. On the other hand, asymmetric key Ren et al. [14] provides an in-depth
algorithms utilize a pair of public and pri- exploration of various cryptographic tech-
vate keys, enabling secure data exchange niques employed in securing data stored and
without the need for a shared secret. While transmitted within cloud environments. The
they provide a robust solution for key dis- research delves into symmetric and asymmet-
tribution, asymmetric algorithms tend to be ric encryption, homomorphic encryption, and
computationally intensive. In cloud com put- elliptic curve cryptography, discussing their
ing, careful selection and integration of these strengths, limitations, and suitability for dif-
cryptographic methods are essential to strike ferent cloud scenarios.
a balance between security and performance, Qi et al. [15] focus on the implementa-
depending on the specific requirements of the tion of homomorphic encryption as a means
application or service. of preserving data privacy during outsourc-
ing in cloud computing. The study presents
a method ology that allows computation on
2. Literature Review encrypted data, ensuring that the cloud ser-
Cryptography has emerged as a fundamental vice provider cannot access the plaintext.
pillar in the realm of cloud computing, play- The paper discusses performance trade-offs
ing a critical role in ensur ing the security and and evaluates the feasibility of the approach
confidentiality of sensitive data within cloud through experimental analysis.
environments. This literature review delves Wang et al. [16] delve into the complexities
into the methodologies and advancements of crypto-graphic key management. The study
pertaining to cryptography’s integration highlights the challenges associated with key
into cloud computing, shedding light on its generation, distribution, and storage in multi-
tenant cloud environments. The research pre-
sents a comprehensive review of existing key
management solutions, including hardware-
AQ4: based key storage, identity-based encryption,
Please and dynamic key generation techniques.
provide
in-text Garrison et al. [17] analyze the vulner-
citation abilities of current cryptographic techniques
for Figure
28.1
in the face of quantum threats and explore
potential post-quantum cryptographic solu-
AQ5:
Please
tions for cloud security. The study highlights
provide the importance of transitioning to quantum-
source Figure 28.1. Classification of cryptography resistant algorithms to safeguard cloud data
for Figure algorithm in cloud computing.
28.1 against emerging threats [22].
AQ6:
Source: AUTHOR'S SELF GENERATED Attribute-based encryption (ABE) has gar-
Please nered attention for its role in enabling secure
provide data sharing among authorized users. Wang et
in-text
citation al. [18] (2015) discuss the utilization of ABE
for Figure as a methodology for fine-grained access con-
28.2
trol in cloud computing environments. The
AQ7: study presents an overview of ABE schemes
Please
Figure 28.2. The process of symmetric-based and evaluates their applicability in scenarios
provide
source key algorithm. involving data sharing among users with var-
for Figure Source: AUTHOR'S SELF GENERATED ying access privileges.
28.2
Towards unbreakable cloud security: Symmetric key cryptography and blockchain synergy 221
Chen et al. [13] delve into the potential of with its ASCII representation, is recorded as a
this technique to allow data owners to verify transaction on the blockchain. This ledger is
the correctness of computation performed maintained by a network of nodes, ensuring
on encrypted data in the cloud. The research decentralization and fault tolerance.
exam ines the cryptographic foundations of Furthermore, to ensure data integrity and
verifiable homomorphic encryption and dis- authenticity, we calculate cryptographic hash
cusses its implications for enhancing trans- values for each transaction, linking them
parency and trust in cloud services. together in a chain of blocks. These blocks
are time-stamped and secured through con-
sensus mechanisms, making it nearly impos-
3. Proposed Method sible to tamper with the data.
In this paper, we propose an innovative To decrypt the data, the recipient retrieves
algorithm for secure data encryption and the blockchain ledger and verifies the chain of
decryption, leveraging the power of both blocks. The ASCII values are then extracted,
symmetric key cryptography and block- and the decryption process proceeds as previ-
chain methodology [13]. Our approach not ously described, recovering the original data.
only ensures data confidentiality but also By incorporating blockchain methodol-
enhances the trust and immutability of the ogy, our algorithm not only safeguards data
encrypted data. The core of our proposed through encryption but also provides an
algorithm lies in binary data encryption. Ini- immutable audit trail, ensuring the highest
tially, we convert the data into ASCII val- levels of data security and trustworthiness.
ues and augment each ASCII value by 100. This approach holds promise for applica-
This augmented data is then further fortified tions in secure communications, financial
using blockchain principles. transactions, and beyond, offering a robust
To integrate blockchain into our meth- and tamper-proof data protection solution.
odology, we introduce a distributed ledger Figure 28.3 shows the proposed algorithm
where each binary data entry, augmented based on a symmetric key algorithm.
AQ8:
Please
provide
source Figure 28.3. Represents the overall process of Proposed Cryptography algorithm using blockchain
for Figure methodology in cloud computing.
28.3
Source:
AUTHOR'S SELF GENERATED
222 Innovations in Computational Intelligence: Trends and Technologies
Algorithm 1 and 2 show the proposed resource discovery through information
cryptography key algorithm based on binary services, and interfaces for creating new
data encryption. applications.
In this research, we have focused on
Algorithm 1 Encryption improving data security and integrity using
the hash function in cloud computing. The
• Step 1: Convert Character to ASCII value. experiment is conducted to analyze the effi-
• Step 2: Add 100 as a key into the ASCII ciency of hash function technique for encryp-
value. tion and decryption of data. Moreover, we
• Step 3: Convert the value into the Binary analyzed the ability of the hash function to
value computed in step 2. identify any unauthorized modifications or
• Step 4: Generate the distributed ledger data corruption in cloud-stored data.
that contains the binary data. Additionally, our research delved into the
• Step 5: Now, the ledger is broadcast to the examination of how hash functions influence
node of the network to maintain the data. data transfer rates and processing. The results
• Step 6: Generate the hash values of each obtained through our experimental investi-
data and link them together in a chain of gation offer valuable perspectives regarding
block. = 0 the real-world appli cation of hash func-
tions within the realm of cloud computing.
Algorithm 2 Decryption Furthermore, these findings underscore the
• Step 1: Firstly pass the hash values to the critical role played by hash functions in safe-
network for approval. guarding sensitive data.
• Step 2: Find out the ledger from the dis-
tributed ledger data that contains the
information in the binary format. 4.1. Comparative Analysis
• Step 3: Extract the binary data from ledger In terms of security strength, our proposed
• Step 4: Convert the binary data into the algorithm demonstrates a high level of secu-
ASCII value rity, surpassing the security strengths of
• Step 5: Subtract the key value 100 from existing approaches such as Public Key Infra-
the ASCII. structure (PKI) [20], Advanced Encryption
• Step 6: Now, return the ASCII value as a Standard (AES) [19] with Centralized Key
Single character value. = 0 Management and Homomorphic Encryp-
tion [21]. The scalability of our algorithm is
excel lent, ensuring efficient performance as
4. Result the cloud environment scales. Performance-
We have put our proposed model into practi- wise, our algorithm is comparable to existing
cal use to enhance the security of cloud com- approaches as shown in Table 28.1.
puting and ensure high-quality services for One of the notable advantages of our pro-
cloud users regardless of their location or the posed algorithm lies in key management, as
time of access. This implementation leverages it employs a decentralized approach, which
our custom simulation tool known as NG- enhances security and reduces the risk of a
cloud, which is a Java-based discrete event single point of failure. On the other hand,
cloud simulation toolkit. existing approaches rely on centralized
Many facility is provided through NG- key management, introducing potential
cloud, including interfaces for efficiently vulnerabilities.
allocating application tasks to resources In terms of resistance to attacks, our
and managing their execution. These fea- algorithm proves to be robust, providing a
tures include application composition, higher level of security against various types
Towards unbreakable cloud security: Symmetric key cryptography and blockchain synergy 223
AQ9:
Please Table 28.1. Comparative analysis of cryptographic approaches in cloud security
provide
source Criteria Proposed Public Key Advanced Encryption Homomorphic
for Table Algorithm Infrastructure Standard (AES) Encryption
28.1
Security Strength High Medium High Medium-High
Scalability
AUTHOR'S SELF GENERATED
Excellent Good Excellent Good
Performance Comparable Faster Slower Comparable
Key Management Decentralized Centralized Decentralized Centralized
Resistance to Attacks Robust Vulnerable to Robust Vulnerable
certain attacks
Transparency High Medium Medium High
of attacks compared to existing Approaches, • Key generation time is optimized,
which show vulnerabilities in specific sce- minimizing overhead during crypto-
narios. Transparency is another key aspect, graphic operations.
where our proposed algorithm and existing • Resource utilization is carefully man-
approach, offer a high level of visibility into aged, ensuring efficient utilization of
the cryptographic processes. computational resources.
The comprehensive comparative analysis Public Key Infrastructure (PKI):
outlined in this re sponse demonstrates the • PKI exhibits slower encryption and
superiority of our proposed algorithm in sev- decryption speeds due to the inherent
eral key aspects, reinforcing its contribution complexity of asymmetric key operations.
to the field of cloud security. • Key generation time is relatively
higher, contributing to increased com-
putational overhead.
4.2. Evaluation of Computational
• Resource utilization tends to be
Efficiency and Performance higher compared to symmetric key
Scalability cryptography.
In this study, we comprehensively assess the Advanced Encryption Standard (AES)
computational efficiency and performance with Centralized Key Management:
scalability of the proposed Symmetric Key • AES, being a symmetric key algorithm,
Cryptography based on blockchain in com- offers fast en cryption and decryption
parison with three widely used cryptographic speeds.
approaches: Public Key Infrastructure (PKI), • Key generation time is efficient, espe-
Advanced Encryption Standard (AES) with cially in a central ized key manage-
Centralized Key Management, and Homo- ment scenario.
morphic Encryption. The evaluation is • Resource utilization is generally low,
conducted in real-world cloud computing making it suitable for performance-
scenarios. oriented applications.
Homomorphic Encryption:
1. Computational Efficiency Analysis: Sym- • Homomorphic encryption, designed
metric Key Cryptography based on Block- for secure computations on encrypted
chain (Proposed Algorithm): data, typically has slower encryption
• The proposed algorithm demonstrates and decryption speeds.
commendable encryption and decryp- • Key generation time can be higher,
tion speeds, leveraging the efficiency impacting the overall computational
of symmetric key cryptography [6]. efficiency.
224 Innovations in Computational Intelligence: Trends and Technologies
• Resource utilization may vary based 5. Conclusion
on the complexity of computations
performed homomorphically. In this study, we present a novel symmetric key
2. Performance Scalability Evaluation: Sym- cryptography-based method for data encryp-
metric Key Cryptography based on Block- tion and decryption. The data is encrypted
chain (Proposed Algorithm): and decrypted using the blockchain method
• Demonstrates robust scalability, effi- in our approach, and the keys are stored by
ciently handling increased workloads creating a hash function. In the context of
and larger data volumes. cloud computing, this technique significantly
• Response time remains stable even improves data security with conventional
under high loads, showcasing scalabil- cloud cryptography algorithms. Our pro-
ity in real-world scenarios. posed method not only elevates security to its
Public Key Infrastructure (PKI): highest attainable level but also offers expe-
• PKI scalability may be challenged due dited file download and upload times in con-
to the computa tional demands of trast to traditional approaches. Experimental
asymmetric key operations. results proved that hash functions are highly
• Response time may increase with effective in detecting even minor modifica-
growing workloads, impacting scal- tions or data corruption, thus ensuring data
ability in resource-intensive scenarios. integrity within the cloud environment. By
Advanced Encryption Standard (AES) ensuring data integrity and confidentiality,
with Centralized Key Management: these cryptographic methods mitigate the risk
• AES exhibits good scalability, particu- of data breaches and unauthorized access.
larly when coupled with centralized Furthermore, we envision future enhance-
key management. ments through the integration of Artificial
• Response time remains relatively sta- Intelligence (AI) techniques, which have the
ble, making it suit able for scenarios potential to further enhance the security of
with varying workloads. cloud computing environments.
Homomorphic Encryption:
• Scalability of homomorphic encryp- References
tion can be limited due to its computa-
[1] Anshika Negi, Mayank Singh and Sanjeev
tional complexity.
Kumar (2015). An efficient security frame-
• Response time may increase signifi- work design for cloud computing using
cantly as workloads and data sizes artificial neural networks. International
grow, affecting scalability in certain Journal of Computer Applications, 129(4),
applications. 0975-8887.
[2] Shakeeba S. Khan and R. R. Tuteja. (2015).
The proposed Symmetric Key Cryp- Security in cloud computing using crypto-
tography based on blockchain demon- graphic algorithms. International Journal of
strates strong computational efficiency Innovative Research in Computer and Com-
and performance scalability, making it a munication Engineering, 3(1), An ISO 3297:
compelling choice for real-world cloud 2007 Certified Organization.
computing scenarios. While each cryp- [3] AL-Museelem Waleed and Li Chunlin.
tographic approach has its strengths and (2016). User privacy and security in cloud
weaknesses, the comparative analysis pro- computing. International Journal of Security
and Its Applications, 10(2), 341–352.
vides valuable insights for selecting the
[4] N N Mosola, M T Dlamini, J M Blackledge,
most suitable algorithm based on specific
J. H. P Eloff and H S Venter. (2017). Chaos-
application requirements. based encryption keys and neural key-store
Towards unbreakable cloud security: Symmetric key cryptography and blockchain synergy 225
for cloud-hosted data confidentiality. South- encryption with multi-keyword search in
ern Africa Telecommunication Networks cloud computing. Security and Communica-
AQ10: and Applications Conference (SATNAC). tion Networks, 2021, 1–11.
Please [5] Ramalingam Sugumar and K. Raja. (2018). [14] J. Ren, Y. Qi, Y. Dai, X. Wang, and Y. Shi.
provide
complete EDSMCCE: Enhanced data security meth- (2015). AppSec: A safe execution environ-
reference odology for cloud computing environment. ment for security sensitive applications.
details. International Journal of Scientific Research ACM VEE. AQ13:
in Computer Science Engineering and Infor- [15] S. Qi and Y. Zheng. (2021). Crypt-DAC: Please
provide
mation Technology, 3(3), ISSN 2456-3307. Cryptographically enforced dynamic complete
[6] Reshma Suryawanshi and Santosh Shelke. access control in the cloud. IEEE Trans- reference
(2016). Improving data storage security actions on Dependable and Secure Com- details.
in cloud environment using public audit- puting, 18(2), 765–779. doi:10.1109/
AQ11: ing and threshold cryptography scheme. TDSC.2019.2908164.
Please International Conference on Computing [16] X. Wang, Y. Qi and Z. Wang. (2019). Design
provide
Communication Control and Automation and implementation of SecPod: A frame-
complete
reference (ICCUBEA). work for virtualization-based security sys-
details. [7] Verma, G., and Kumar, M. (2022). Sys- tems. IEEE Transactions on Dependable
tematic review and analysis on underwater and Secure Computing, 16(1), 44–57.
image enhancement methods, datasets, and [17] W. C. Garrison III, A. Shull, S. Myers and A.
evaluation metrics. Journal of Electronic J. Lee. (2016). On the practicality of cryp-
Imaging, 31(6), 060901-060901. tographically enforcing dynamic access con-
[8] S. Singh and V. Kumar. (2015). Secured trol policies in the cloud. Proceedings IEEE
user’s authentication and private data stor- Symposium Security and Privacy, 819–838. AQ14:
age-access scheme in cloud computing using [18] J. Wang, X. Chen, X. Huang, I. You and Please
provide
Elliptic curve cryptography. 2015 2nd Inter- Y. Xiang. (2015). Verifiable auditing for volume
national Conference on Computing for Sus- outsourced database in cloud computing. number.
tainable Global Development (INDIACom), IEEE Transactions on Computers, 64(11),
791–795. 3293–3303.
[9] K. Brindha and N. Jeyanthi. (2015). Securing [19] Al-Timeemi, S. R. F. (2023). Smart card secu-
cloud data using visual cryptography. Inter- rity authentications using homomorphic
national Conference on Innovation Infor- encryption (Master’s thesis, Altınbas¸ Üni-
mation in Computing Technologies, 1–5. versitesi/Lisansüstü Eğitim Enstitüsü).
[10] Dharitri Talukdar. (2015). Study on sym- [20] Alfa, A. A., Aro, T. O., Adunni, O. O., Iliya,
metric key encryption: An overview. Inter- D. Z., Awe, O., and Ayodeji, A. (2023).
national Journal of Applied Research, 1(10), Shaaes based on chaining block code and
543–546. Galois counter mode encryption modes for
AQ12: [11] David G. Rosado, Eduardo Fernandez- privacy of smart farming systems. Minna
Please Medina and Eduardo B Fernandez. (2013). International Journal of Science and Tech-
provide
complete An analysis of security issues for cloud nology, 2(1), 40–52.
reference computing. Journal of Internet Services and [21] Beune, P. F. (2023). Lightweight Public Key
details. Applications. Infrastructure for IoT (Master’s thesis, Uni-
[12] G. Devi and M. Pramod Kumar. (2012). versity of Twente).
Cloud computing: A CRM service based on [22] Agrawal, K., and Bhatnagar, C. (2023).
a separate encryption and decryption using M-SAN: A patch-based transferable adver-
blowfish algorithm. International Journal sarial attack using the multi-stack adversar-
of Computer Trends and Technology, 3(4), ial network. Journal of Electronic Imaging,
592–596. ISSN 2231-2803. 32(2), 023033-023033.
[13] Chen, Z., Wu, A., Li, Y., Xing, Q., and Geng,
S. (2021). Blockchain-enabled public key
29 Neuralink’s approach to spinal cord
injuries and vision impairment through
pioneering brain–machine interfaces
Manoj M.a, Banupriya N.b, Gayathri V.c, Girinath D.d, and
Shibi Kannan A.e
Computer and Communication Engineering, Sri Eshwar College of Engineering, Coimbatore,
Tamil Nadu
Abstract: Neuralink, founded by Elon Musk, leads in merging human brains with technology. Our unique
perspective complements and challenges existing paradigms. Brain-machine interfaces (BMIs) hold profound
therapeutic potential, restoring mobility for those with spinal cord injuries through direct brain-computer com-
munication. BMIs offer groundbreaking hope, bypassing the visual pathway to potentially restore sight to the
blind, revolutionizing their lives. Our research extends beyond presenting BMI’s current state. We explore
a project reshaping BMI understanding, pushing boundaries. We also unveil a novel theoretical solution to
blindness, diverging from convention. This innovative framework promises a unique path to address visual
impairments, injecting excitement into the BMI narrative. In summary, our research provides a comprehensive
perspective on Neuralink and BMIs. It highlights their therapeutic potential for spinal cord injuries and curing
blindness. Moreover, it introduces fresh viewpoints and a novel theoretical solution, promising an engaging
journey through the evolving landscape of brain- machine interfaces.
Keywords: BMI; mental synthesis theory; neuralink; neurotechnology; visual prosthesis
1. Introduction shifting our focus toward an often-overlooked
frontier—the realm of Visual Prosthesis.
As we traverse the swiftly evolving terrain of
The endeavor to restore lost functionalities
brain-machine interfacing (BMI), Neuralink
extends beyond neurological disorders, reach-
stands out as a pioneering force, founded by
ing into the domain of visual impairment.
the visionary Elon Musk, aiming to bridge
the profound gap between the human brain Visual Prosthesis, our alternative focal
and cutting-edge technology. However, our point, represents an innovative approach to
exploration extends beyond the boundaries address challenges faced by individuals with
set by Neuralink, branching into an alterna- visual impairments. Within this exploration,
tive avenue with profound implications— we scrutinize cutting-edge developments,
Visual Prosthesis. potential applications, and ethical consid-
While Neuralink captivates attention with erations associated with Visual Prosthesis.
its ambitious goals, our paper embarks on a The paper seeks to unravel the transforma-
compelling exploration that moves ‘Beyond tive possibilities offered by this alternative
Neuralink.’ In this discussion, we navigate avenue, laying the foundation for a nuanced
the intricate landscapes of BMI technology, understanding of its implications.
a
[email protected]; [email protected]; [email protected]; dgirinath250@
gmail.com; [email protected]
DOI: 10.1201/9781003606635-29
Neuralink’s approach to spinal cord injuries and vision impairment through pioneering brain 227
As we journey ‘Beyond Neuralink,’ our of the profound impact such innovations can
analysis unfolds into a multifaceted exami- have on the human experience.
nation. We delve into the intricacies of Vis-
ual Prosthesis technology, exploring how
it employs BMI principles to restore vision 2. Literature Review
and presenting a unique perspective on the A literature review on the topic “Beyond Neu-
intersection of biology and technology. This ralink: An Alternate on Visual Prosthesis”
alternative avenue not only promises to rev- explores recent developments and alterna-
olutionize healthcare but also introduces a tive approaches to visual prosthesis technol-
paradigm shift in our understanding of sen- ogy beyond Elon Musk’s Neuralink. Visual
sory augmentation, particularly in the con- prosthesis aims to restore or enhance vision
text of vision. in individuals with visual impairments. While
Furthermore, the paper sheds light on Neuralink has garnered significant attention
potential societal impacts and ethical con- for its brain-computer interface technol-
siderations associated with advancing Visual ogy, this review delves into other innovative
Prosthesis technology. As we weigh the ethi- solutions and research directions in the field.
cal dilemmas inherent in merging biology These alternatives may include advancements
and technology to address visual impairment, in retinal implants, optic nerve stimulation,
the discourse extends beyond technological and non-invasive techniques like optogenet-
aspects to encompass broader societal impli- ics or advanced visual processing algorithms.
cations, such as accessibility, inclusivity, and By examining a range of options, this review
the potential to redefine standards of living seeks to provide a comprehensive overview
for individuals with visual disabilities. of the evolving landscape of visual prosthesis
By crafting this narrative ‘Beyond Neu- technology, offering insights into the poten-
ralink,’ our intention is not to diminish the tial future directions beyond Neuralink’s
groundbreaking advancements made by Neu- approach.
ralink but to broaden the discourse. We offer
an alternative perspective that explores the
transformative potential of Visual Prosthe- 2.1. What is BMI?
sis, contributing to the ongoing dialogue sur- A technique called a brain-machine interface
rounding BMI technology. This paper urges a (BMI), also called a brain-computer interface
comprehensive consideration of diverse ave- (BCI), creates a direct line of communication
nues, encouraging a holistic understanding between the human brain and outside objects
or computer systems. BMIs enable the brain
to transmit signals, receive information, or
control external equipment without the need
for traditional methods of interaction like
keyboards, mice, or touchscreens.
It’s important to note that BMIs are a rap-
idly evolving field, and ongoing research con-
tinues to expand their capabilities and address
their limitations. BMIs have the potential to
offer groundbreaking solutions for individu-
AQ1: als with disabilities and open up exciting pos-
Please
provide sibilities for human-computer interaction, but
source they also raise important ethical and societal
for all the Figure 29.1. Representation of BMI.
figures
questions as they become more integrated
Source: into our lives.
AUTHOR'S SELF GENERATED
228 Innovations in Computational Intelligence: Trends and Technologies
2.2. What is Neuralink? with their brains, it aims to aid individuals
with spinal cord injuries. It is thought that
Neuralink is a cutting-edge neurotechnol-
using spinal stimulation techniques in con-
ogy company co- founded by entrepreneur
junction with this strategy can help restore
and visionary Elon Musk. Established in
motor functions. [1].
2016, Neuralink is on a mission to pioneer
Additionally, bionic limbs are not a novel
the development of advanced brain-machine
concept, and the traditional method does
interface (BMI) technologies. These interfaces
not require BMIs. When the user flexes their
aim to bridge the gap between the human
residual limb muscles, a bionic limb, such
brain and artificial intelligence (AI) systems,
as an arm, recognizes minuscule naturally
potentially revolutionizing the way we inter-
generated signals. They can then be trans-
act with technology, understand the brain,
formed into hand movements via the bionic
and treat neurological disorders. Neuralink’s
limb. BMIs, however, can take this strategy a
primary goal is to establish a direct line of
step further by enabling us to communicate
communication between the human brain
directly with our brains rather of having to
and any other devices or computers. This
translate our intentions into text, speech, or
groundbreaking technology has the poten-
gestures through an additional physical step
tial to transform a wide range of fields, from
[2].
healthcare to communication and beyond.
BCIs can be applied to treat neurological
In this exploration of Neuralink, we will
conditions and provide information on how
delve deeper into the company’s goals, the
the brain works. In order to treat conditions
groundbreaking technologies they are devel-
like epilepsy, Karageorgos et al. presented
oping, and the potential implications for
HALO (Hardware Architecture for Low-
medicine, science, and society. Neuralink
power BCIs), an architecture for implantable
stands at the forefront of innovation, push-
BCIs [3].
ing the boundaries of what is possible at the
Deep brain stimulation (DBS) can be used
intersection of neuroscience and technology,
to treat epilepsy and Parkinson’s disease. DBS
with the potential to reshape the future of
is a type of invasive BMI. A thin electrode
human interaction with the digital world.
wire must be implanted during surgery in the
area of the brain that controls aberrant move-
2.3. Neuralink applications ment. [2].The present scenario has shown the
Neuralink’s flagship application is spinal cord increased threats represented by respiratory
damage. By enabling them to dexterously han- illness like chronic obstructive pulmonary
dle a computer mouse and keyboard directly disease (COPD), asthma etc. That risk has
increased due to increase in air pollutants like
PM2.5, PM10 etc. a respirator can be utilized
as an immediate countermeasure on an indi-
vidual level safety measure as bringing down
pollution levels require much longer time
than the severity of the problem is allow-
ing. The normally used N95 respirators are
of negative pressure variant i.e., they require
the wearer’s lungs to inhale air through the
resistive membranes of the filter layers. This
is strenuous and uncomfortable to wear for a
long duration. This is non-existent in positive
Figure 29.2. Elon Musk’s Neuralink. air pressure respirators as they use external
Source: filters and has a motorized air supply system.
AUTHOR'S SELF GENERATED
Neuralink’s approach to spinal cord injuries and vision impairment through pioneering brain 229
The pandemic in recent scenario also neces- circumstances. Future predictions indicate
sitates respiration apparatus as a part of its that this trend will get worse.
treatment. Respirators that are in commonly
used are negative pressure system which 2.4. Visual prosthesis on Neuralink
require the power of lungs to draw-in puri-
fied air which is not suitable and sometimes Modern BMI applications that we have seen
not possible if the person lacks sufficient lungs so far will allow us to talk about potential
strength, or if they suffer from respiratory ill- future developments. The effects on humans
ness. This work proposes a forced air (posi- and ethical dilemmas will then be assessed,
tive air pressure) solution to the problem. along with transhumanism and its effects.
To better understand the origins of cog- BMIs may provide advantages like greater
nition and other brain functions, significant health, memory, concentration, and healthier
money has recently been committed to the aging. However, they also come with new
Brain Research via Advancing Innovative dangers, like the potential for accessing peo-
Neurotechnology (BRAIN) initiative. For ill- ple’s thoughts or moods, which might lead to
nesses like autism and mental disorders, it is a trend toward controlling them. [6].
envisaged that more 3 efficient treatments The development of visual prosthesis is
can be developed. [2]. one of those. In the past 25 years, there have
The most prevalent mental health problem been substantial advancements in retinal
is depression. Treatment resistance is present prosthesis systems, leading to the creation
in about one-third of depression patients. For of numerous unique surgical and engineer-
those who have not responded to medication, ing techniques. Positive findings have shown
BMI interventions are seen as a last resort. partial vision restoration, with enhancements
Drugs may be unsuccessful because they in both coarse objective function and daily
influence every part of the body, but BMIs task performance. In order to represent the
can specifically target the right parts of the known safety profile of these devices for
brain. Weight gain and decreased libido are a chronic implantation, only four systems have
couple of the negative effects of taking drugs acquired marketing authorization for use in
for mental health. BMIs can also have nega- Europe or the United States to date. Many
tive effects, but they may be less severe than more are currently undergoing preclinical
those from medications [4]. and clinical study. With this advancement, the
BMIs may also aid in the earlier diagno- possibility that the field of visual restorative
sis of brain illnesses that encourage synaptic medicine can provide blind patients with a
communication system failure [5]. BMIs can realistic and quantifiable benefit is first raised.
often be seen as competing with or offering
alternatives to conventional medicine in some
Figure 29.3. BCI chip installation. Figure 29.4. Visual prosthesis representation.
Source: Source: AUTHOR'S SELF GENERATED
AUTHOR'S SELF GENERATED
230 Innovations in Computational Intelligence: Trends and Technologies
To close the remaining gap between artifi- lobe, which controls vision. However, the pre-
cial and natural vision, various difficult engi- frontal cortex is the area that should be stim-
neering and biological challenges still need ulated, according to the idea being proposed.
to be resolved. The future of retinal prosthe- Like Neuralink, our method involves hav-
ses is bright and inventive thanks to recent ing the user or patient wear a GoPro camera,
advancements in the form of improved image which will record a video and utilize machine
processing algorithms and data transfer strat- learning algorithms like logistic regression to
egies, as well as new nanofabrication and identify and categorize the objects in front
conductive polymerization techniques. This of the camera. Send it to the chip after that,
review updates retinal prosthetic systems that where it is converted into electrical signals
are in the process of development and clini- and sent to the prefrontal cortex, where the
cal testing while also addressing upcoming ensemble and imagination take place.
difficulties in the field, like the evaluation of This means that instead of stimulating the
functional outcomes in ultra- low vision and rod cells in the occipital lobe, we will pro-
methods for overcoming current hardware pose a strategy that involves stimulating the
and software limitations. [7]. human brain’s imagination.
In Neuralink, The Blindness is expected to We must first look at the mental synthesis
be cured and it aims to give a visual experi- theory in order to comprehend imagination
ence to the people who have never seen light in better.
their lifetime. The way of approach that Neu-
ralink put forth is as follows, the chip in the
brain would communicate with 64 tiny wires 3. Mental Synthesis Theory
implanted by a Neuralink surgical robot into This article discusses a theory regarding the
the visual cortex. Through this, Neuralink development of the human mind and sug-
can bypass the eye and generate a visual gests experiments that could be carried out
image directly into the brain. The user would to verify, deny, or test the theory. The theory’s
have to wear a GoPro style digital camera, central question is how the brain reacts when
which would wirelessly transmit a live visual two items that have never been seen together
feed via Bluetooth onto a mobile device. The before—for example, an apple perched atop
phone then converts that image data into a a whale—are pictured together for the first
neural signal, which is transplanted back into time.
the Neuralink chip. The chip then transmits We are aware that a neuronal ensemble
this neural signal into the brain, allowing you in the brain represents a known item, like an
to see.
2.5. An alternate on visual prosthesis
However, despite being a significant scien-
tific advance, this method has a limitation in
that it can only provide the patient black and
white images because it only activates the rod
cells in the occipital lobe.
Therefore, we present our own theory
here, which also much depends on how the
brain chip is implemented. With the excep-
tion of the approaching of the brain, every-
thing is the same as Elon Musk’s Neuralink. Figure 29.5. Prefrontal Cortex.
Neuralink, then, interacts with the occipital Source:
AUTHOR'S SELF GENERATED
Neuralink’s approach to spinal cord injuries and vision impairment through pioneering brain 231
signals corresponding to the items recorded
by the camera.
3.1. Drawbacks in the alternate
approach
The alternative visual prosthesis approach’s
fundamental flaw is that it advocates for
the idea that one’s vision can be preserved
Figure 29.6. Representation of Mental Synthesis
through imagination. A person must have
Theory.
seen things in order to imagine them, and
Source: slideplayer.com
their brains must be aware of them.
Thus, only those who recently lost their
vision can keep it, whereas it is practically
apple or a whale. The neural ensemble of such impossible for someone who was born blind
an object tends to engage into synchronous to gain their vision back.
resonant activity when it is seen or remem- To achieve the precise vision in their head,
bered (Quiroga et al., 2008). one should also have a strong sense of imagi-
The binding-by-synchrony theory refers nation. This method uses a person’s capac-
to the neuronal ensemble binding process, ity for imagination to provide them with an
which is based on the Hebbian concept that accurate vision.
“neurons that fire together, wire together”
(Singer, 2007).
The Hebbian principle does not, however, 4. Conclusion
account for the countless novel objects that In this study, we went beyond Neuralink and
people can voluntarily imagine, despite the investigated the field of visual prosthesis.
fact that it does explain how we perceive a Although Neuralink’s brain- computer inter-
familiar object. Since the components of face is revolutionary, we looked at a different
those unique images have never been viewed strategy that has enormous potential for the
together, the neural ensembles encoding those blind and visually impaired.
objects cannot spontaneously coordinate Hope is offered by visual prosthesis, which
their activity. use cutting-edge technology and direct brain
According to the research, the binding- interface to try to replace lost sight. Through
by- synchrony hypothesis would need to be the development of artificial vision, people
expanded to account for the process of men- may be able to navigate their environment
tal synthesis, in which the brain actively and and restore their freedom. But it’s a difficult
purposefully synchronizes separate neural trip, fraught with technical difficulties, moral
ensembles into a single altered image, in order dilemmas, and demanding clinical studies.
to explain imagination. As a result, the apple Our conclusion underlines the importance
and whale neuronal ensembles are synced, of working together to find solutions to vision
and the two different things are experienced impairment. Beyond Neuralink, a new path
as one.[8]. So, this is how the human brain develops, one full of hope and perseverance.
processes imagination. Therefore, if Neu- We demonstrate human inventiveness and
ralink technology develops further and we our commitment to extending possibilities in
are able to outperform neural transmissions, this changing neurotechnology world. Visual
we could simply assemble the pieces in the prosthesis is a ray of hope that point the way
prefrontal cortex by transmitting the neural to a more inclusive future.
232 Innovations in Computational Intelligence: Trends and Technologies
References [5] M. Veletić and I. Balasingham, “Synaptic
Communication Engineering for Future
[1] Ramirez et al., “An update on visual prosthe- Cognitive Brain–Machine Interfaces,” in
sis,” Int. J. Retina Vitreous,vol.9,no.73,2023, Proceedings of the IEEE, vol. 107, no. 7,
https://doi.org/10.1186/s40942-023-00498-1. pp. 1425–1441, July2019, DOI: 10.1109/
[2] M. Reilly, “Brain–Machine Interfaces as JPROC.2019.2915199
Commodities: Exchanging Mind for Mat- [6] Gurtner, D. (2021). Neuralink and beyond:
ter,” IEEE Transactions on Biomedical Engi- Challenges of creating an enhanced human.
neering, Published online Aug. 4, 2020. doi: [7] Bloch, Edward and Luo, Yvonne and Cruz,
10.1177/0024363920930882. Lyndon. (2019). Advances in retinal pros-
[3] Karageorgos et al., “Hardware-Software thesis systems. Therapeutic Advances in
Co- Design for Brain-Computer Interfaces,” Ophthalmology. 11. 251584141881750.
2020 ACM/IEEE 47th Annual International 10.1177/2515841418817501.
Symposium on Computer Architecture [8] Vyshedskiy,Andrey.(2014).The mental synthe-
(ISCA), Valencia, Spain, 2020, pp. 391–404, sis theory: the dual origin of human language.
DOI: 10.1109/ISCA45697.2020.00041 344–352. 10.1142/9789814603638_0046.
[4] Royal Society. (2019). iHuman: blurring
lines between mind and machine. Royal
Society
30 Honeypots and honeynets: investigating
attack vectors
Pawandeep Kaur1,a and Harpreet Kaur1,b
1
Computer Science and Engineering, Lovely Professional University, Bengaluru, India
Abstract: The honeypot method detects banned users and network intruders. The Honeypot software goal is
to create a method of spotting network intruders. This paper describes the honeypot methodologies to imple-
ment intrusion detection and prevention systems. The invaders are now being traced using an unprotected
network and an unused IP address, and various IP trapping tactics are being used. Here, a honeypot approach
is employed to identify and trap the target. Honeypots use real replicated operating systems on the deployed
network of the servers. Honeypots and honeynets are utilised as security tools to entice attackers into a trap
and gather valuable intelligence regarding their tactics, strategies, and goals. These tools can be set up on differ-
ent computer components, such as software, networks, servers, routers, and more. They are classified based on
their usage and level of interaction as production or research honeypots and high or low-interaction honeypots.
Honeynets, compared to a single honeypot, offer greater insight and observations. Honeypots and honeynets
can detect and investigate a wide range of attack vectors. These factors include the honeypot’s purpose, engage-
ment, reality, isolation, and the distinct goals, available resources, constraints, and challenges of each situation.
Keywords: Attack vectors, computer security, honeypots, honeynets, network intrusion detection, network-
based security
1. Introduction honeypots are being used to open up such
ports and capture practically every possible
The realm of cyberspace is continuously con-
malicious behaviour and attack on servers.
fronted with cyber threats [1] every day in and
It gives a brief about the loopholes/vulner-
day out. Mandiant threat intelligence [2, 28]
abilities in the systems that need to be patched
found 80 zero-day exploits exploited in nature
immediately. Intrusion detection systems are
in the year 2021, more than doubling the pre-
vious high number in 2019. The main objec- great at detecting network vulnerabilities but
tive of the network operators and network they are also susceptible to ‘false positives’.
administrators is to remove the vulnerability as Methods such as signature-based analysis are
quickly as possible [27]. The honeypot comes known to have false alert positives whereas
in a very handy resource for big enterprises anomaly-based detection also generates false
by sacrificing resources for the sole purpose alert positives if the system is not properly
of investigating the intrusions on a dummy trained [10]. If compared to honeypot, it prac-
monitored network. A firewall is being utilised tically never generates false positives, it only
to eliminate unauthorised access to the net- observes the traffic and analyses the traffic for
work by blocking certain defined ports as per malicious intents. Since the honeypot contains
rules. The drawback is that it does very little in no sensitive data, there is no question of gener-
terms of filtering traffic from internet sources ating alerts and largely reducing the overhead
[6]. It blocks access to malicious traffic but of managing alerts and constant monitoring
also blocks recognised traffic. Subsequently, [4, 14].
[email protected]; [email protected]
a
DOI: 10.1201/9781003606635-30
234 Innovations in Computational Intelligence: Trends and Technologies
There are two types of honeypots: server- Joshi, et al. [9] book provides a compre-
based honeypots and client-based honeypots. hensive and easily understandable presen-
Server-based honeypots are traditional-style tation of honeypots in wired and wireless
honeypots. The server-based honeypots work networks. It outlines the patterns of network
as passive honeypots which passively moni- attacks that led to the development of hon-
tor and detect vulnerabilities in the incoming eypots. It explains how honeypots can be
server traffic. Client-based honeypots detect classified and their importance to network
client-based vulnerabilities [5]. It constantly security. Mitchell, A. [12] presents the mod-
monitors the connected clients and accepts ern arsenals of cyber defence must includ-
the attacks and judges them to be safe or not. ing honeypots. He discovered that there had
This allows for increased accuracy and lower already been a sizeable amount of research
false positive identification rates According to papers and articles written about honey-
their demands and objectives, the administra- pots, which allowed me to simply respond
tor grants security access to the organisation’s to any inquiries about their history. Lacerda
users. et al. [23] deploy and analyse various honey-
pot sensors to determine the most common
exploits and security flaws used by hackers
2. Related work to penetrate computers. This study seeks to
T.Holz and F.Raynal states [3] several strate- offer some recommendations for preventing
gies and provided instances of reverse meth- or reducing the harm done by these assaults
ods and tools that aid attackers. It lists two in actual systems. Four distinct honeypot
methods for creating a high-interaction hon- sensors, including Kippo, Snort, Conpot,
eypot: utilising a virtual machine or enhanc- and Dionaea, were used in this investigation.
ing the system’s logging capabilities. Spitzner, Campbell et al. [24] aimed to fill knowl-
L. [5] paper elaborate the most severe threats, edge gaps in the honeypot environment, the
the advance insider, the dependable person goal of this study was to survey emerging
who is familiar with internal organisation, patterns in existing honeypot research. As
has received less attention from researchers. a result, a honeypot can be employed as a
These people are interested in the data pro- research instrument to compile information
vided, not the systems. This talk goes into on network attacks.
the use of honeypot technology to find, rec-
ognise, and gather data on these particular
threats. Sayed, M. A. [7] states that in the 3. Methodology
cyber death chain, reconnaissance efforts Honeypots are used as decoy servers to
come before other assault phases. Attacks lure malicious actors to seem like a lucra-
that take advantage of zero-day vulnerabili- tive opportunity for infiltration. The Honey-
ties offer attackers an advantage over estab- pots are designed to mimic the real servers
lished defenses. In order to safeguard the without the necessity of having sensitive
network’s most important assets, address- information on them. It is deployed in high-
ing the issue of ‘How to allocate honeypots trafficked networks to direct genuine users
over the network’ in this paper. Spitzner, L. and redirect malicious users from the serv-
[8] book, define a honeypot, describe how ers [17]. While the malicious actor falls for
it operates, and discuss the various benefits the honeypot server, the server administrator
that this unusual technology can provide. or the network admin can gather substan-
The six main honeypot technologies are then tial information about the attacks. These can
covered in depth. Finally, discuss honeypot include the 5W framework who attacked
deployment and upkeep difficulties. the servers, what happened after the attack,
Honeypots and honeynets: investigating attack vectors 235
when did the attack happen, where the attack 4. Classification of Honeypots
can spread and why does it happen [15]. The
honeypot works at its finest when it accu- Honeypot is a machine that the malicious
rately mimics the actual servers, which cre- actor has a free range of access to it but the
ates incentives for malicious actors to attack condition is you have to hack to make it work.
the servers first. It must run the same tasks, Honeypot is not different from traditional sys-
and processes and contain decoy files for the tems, it has a central processing unit, memory
targeted process. As shown in Figure 30.1. unit as random-access memory and same hard
The architecture of Honeypot, the honey disk to store its data. The only difference it cre-
pot’s suitable place is behind the firewall of ates is that it logs the attacks as well as threats
the network. that are present in the corporate network. Any
interaction with the honeypot is said to be the
default [13]. Honeypot’s classification is done
3.1. Honeypot Working based on server-based honeypots and client
The honeypot looks exactly like an origi- based honeypots as shown in Figure 30.2.
nal system. If a malicious actor has to guess
between the highly secured production server
and the slightly less secured server [16]. Eve- 4.1. Server-Based Honeypots
ryone will choose the slightly less secure A server-based honeypot is a security tool
server, that’s where the honeypot will be in that simulates a server application or ser-
the corporate network. A honeypot can be a vice that is vulnerable to cyber-attacks. It is
system that can falsely claim to have person- used to attract, deflect, and analyse hackers
ally identifiable information as a lucrative who try to break into a network or system.
decoy. These decoy fields can easily be filled By monitoring the activity on the honeypot,
with dummy data to draw out the malicious security analysts can learn about the hack-
actor to it [20]. The honey will log every step ers’ identities, methods, and motives [18].
of the malicious actor. The main purpose of A server-based honeypot can also protect
the honeypot systems is to refine and fine- the real servers from being compromised by
tune the network-wide rules in the intrusion diverting the hackers’ attention to the decoy.
detection system. Server-based honeypots are usually deployed
AQ1:
Please
provide
source
for all Figure 30.1. Honeypot architecture. Figure 30.2. Classification of honeypot.
figures.
Source: Source:
AUTHOR'S SELF GENERATED
AUTHOR'S SELF GENERATED
236 Innovations in Computational Intelligence: Trends and Technologies
in network areas that are not normally • Medium information gathering capability
accessed by legitimate users or customers. • Medium risk of attackers since it can
mimic both the low interaction and higher
4.2. Client-based Honeypots interaction honeypots.
• Medium retention of the attackers.
A client-based honeypot mimics a client
program or service that communicates with
computers. It is used to locate and identify 5.3. Higher Interaction Honeypots
sites that take advantage of client-side flaws. • Higher interaction honeypots provide the
The customer-based honeypot pretends to be attacker with an actual operating system
a client and views possibly malevolent web- to exploit.
sites or services to see if an assault has hap- • The network installation and configura-
pened. Security researchers can learn about tion are highly challenging.
the names, methods, and risks of networks by • Complex methods of implementation
watching their behaviour on the client-based and require the complex procedure of
honeypot [19]. maintenance.
• Extensive collection of information.
• It contains a higher risk of revealing or
5. Honeypot Deployment leaking data since it provides real services
The placement of honeypot machines is influ- to the internet.
enced by a few variables. These classifica- • Higher retention for the attackers.
tion criteria make it simpler to understand,
how they operate and how to use them when 5.4. Production Honeypots
deciding which of them to employ inside of a
network or IT architecture. The application • Production honeypots provide simulated
of honeypot software used in production or services and operating service capabilities.
study distinguishes it. The implementation of • Its production and installation difficulty
such honeypots is explained. depends on the organisation
• It is the medium level of difficulty for both
implementation and maintenance of the
5.1. Low Interaction Honeypots system.
• These give hackers access to replicated • It has a medium collection of information.
or mimicked environments in which to • Higher risk of losing organisation-sensi-
experiment. tive data.
• It is simple to deploy on networks. • Higher retention for the attackers.
• Easy to implement and maintain.
• Limited information gathering capability. 5.5. Research Honeypots
• Lower risk of attackers.
• Lower retention of the attackers. • They offer the honeypots actual services
that enable the honeypots to gather a lot
of evidence about the network attackers.
5.2. Medium Interaction Honeypots • It is challenging to set up and customise.
• Superior to honeypots with limited • Very difficult to deploy.
interaction. • It provides a huge amount of attacker
• Complex to install but easier than higher details.
interaction honeypot. • Fewer risks of data loss since it is used for
• Medium implementation complexity and research purposes only.
maintenance. • Medium retention for the attackers.
Honeypots and honeynets: investigating attack vectors 237
6. Comparison of Various 7. Attack Vectors on Honeypot
Honeypot Solutions A method of attack is a route or method
Table 30.1 presents the comparison of various through which a threat actor can obtain a
Honeypot solution on diverse platforms con- network or another system without authori-
sidering level of interaction, integration with zation. An attack vector in the context of
SIEM/SOC solutions representing the popu- honeypots is a vulnerability or flaw that
larity in market. Depending on whether they the honeypot is intended to reveal and that
imitate particular services, systems, or entire net- attackers can use to get into the honeypot.
works, honeypots can range in deception level Honeypots are intentionally vulnerable to
from low to high. Low-interaction honeypots drawing intruders and are made to seem like
mimic basic services with little capability, while legitimate systems. Malicious actors can use a
high-interaction honeypots closely mimic real lot of attack vectors to gain access to the hon-
systems and are quite deceiving [12]. Deception eypots. Some of the attack vectors are listed
technologies frequently cover a wider range of below are:
deceptive aspects, including breadcrumbs, cre-
dentials, network segments, and decoy files. 7.1. Malware Honeypot
Depending on the particular components Malware honeypots use well-known tech-
employed, they give a variety of deception niques for attacking to attract Malware.
levels, from minimal to high.
7.2. Database Honeypot
A database honeypot is a dummy database
created to draw assaults targeted specifically
at databases, including SQL injection, which
illegally controls data [15]. Security special-
ists can also study how hackers found and
entered the decoy database to enhance the
security of the real database.
7.3. Client Honeypot
Figure 30.3. Deception level vs level of
interaction. A client honeypot is a security tool that
Source: AUTHOR'S SELF GENERATED
makes it possible to identify rogue servers on
Table 30.1. Various solutions of honeypot
Honeypot Level of Open / Deception Integration with SIEM/ Popularity
solution Interaction closed level SOC solutions
Honeyd Low Yes Moderate Limited (may require Low
custom integration)
KFSensor High No High Yes High
HoneyBOT Low Yes Moderate Limited (may require Moderate
custom integration) AQ2:
Please
SpectorOps High No High yes High provide
canary source for
Tables.
HoneyRJ Low yes Low Limited Low
238 Innovations in Computational Intelligence: Trends and Technologies
a network. The client honeypot impersonates loss of resources since the attacker will con-
a client and communicates with the server to stantly plan, execute and try to break into the
check for signs of an attack. The client can system. Sacrificing a system on the network
send requests to the queuer’s listed servers. to gain an advantage over malicious actors is
The analysis engine is in charge of assessing a disadvantage of its own. To summarize, the
whether such a breach of the customer’s hon- invention of honeypot systems for security
eypot has occurred. [16]. solutions is the best possible non-interfering
intrusion detection system.
7.4. Honeynet
This decoy encourages malware-type attacks. References
For example, a honeypot mimics the customer [1] Sadowski, James. “Zero Tolerance: More
banking credential data to attract hackers to Zero-Days Ex- ploited in 2021 Than Ever
deploy malware such as Zeus [14], which Before.” Mandiant, 21 Apr. 2022, www.
steals banking information and can easily be mandiant.com/resources/blog/zero-days-
detected with these. exploited-2021. Accessed 27 Mar. 2023.
[2] “Cyber Threat Real-Time MAP.” STATIS-
7.5. Spam Honeypot TICS — Kaspersky Cy- berthreat Real-time
Map, 31 Mar. 2023, cybermap.kaspersky.
A spam honeypot is a type of deception tech- com/stats. Accessed 31 Mar. 2023.
nology that creates a fake email address or [3] T. Holz and F. Raynal “Detecting Honeypots
a hidden form field to lure and trap spam- and other suspicious environments ”, 2005
mers. Spammers often use bots to harvest IEEE, workshop on Information Assurance
email addresses or fill out forms with junk and Security, United States Military Acad-
data. A spam honeypot can help identify and emy, West Point, NY
block spammers by detecting when an email [4] W. Fan, Z. Du, M. Smith-Creasey and D.
or a form submission is sent to a fake address Ferna´ndez, “HoneyDOC: An Efficient
Honeypot Architecture Enabling All-
or a hidden field. This way, legitimate users
Round Design,” in IEEE Journal on Selected
are not affected by anti-spam measures like Areas in Communications, vol. 37, no. 3,
CAPTCHA or verification codes [11]. pp. 683–697, March 2019, doi:10.1109/
JSAC.2019.2894307.
8. Conclusion [5] Fortinet (n.d.). “What Are Honeypots (Com-
puting)?” Retrieved March 31, 2023, from
In this paper, the requirements and the sole https://www.fortinet.com/resources/cyber-
objective of the honeypots in the network glossary/wh at- is-honeypotK. R. Sekar, V.
were discussed. Traditional servers suffered Gayathri, G. Anisha, K. S. Ravichandran and
from a lack of exposure from the outside R. Manikandan, ”Dynamic Honeypot Con-
world which in turn less security in terms of figuration for Intrusion Detec- tion,” 2018
public facing honeypots. These create risks 2nd International Conference on Trends
like getting detected, high resource demands in Electronics and Informatics (ICOEI),
Tirunelveli, India, 2018, pp. 1397–1401,
and constant supervision. The honeypots
doi: 10.1109/ICOEI.2018.8553956.
allow users to observe what they’re after, [6] L. Spitzner, Honeypots: Tracking Hackers.
their methods for going regarding achiev- Addison Wesley, 2002.
ing their goals, and what they can do to stop [7] Sayed, M. A., Anwar, A. H., Kiekintveld,
them. Furthermore, honeypot installation C., Bosansky, B., and Kamhoua, C. (2022,
and configuration require more planning as October). Cyber deception against zero-day
more threats are being discovered on daily attacks: a game theoretic approach. In Inter-
basis. The presence of Honeypots means the national Conference on Decision and Game
Honeypots and honeynets: investigating attack vectors 239
Theory for Security (pp. 44–63). Cham: [18] Spitzner, L. (2003). Honeypots: tracking
Springer International Publishing. hackers (Vol. 1). Reading: Addison-Wesley.
[8] Spitzner, L. (2003). Honeypots: tracking [19] Joshi, R. C., and Sardana, A. (Eds.). (2011).
hackers (Vol. 1). Reading: Addison-Wesley. Honeypots: a new paradigm to information
[9] Joshi, R. C., and Sardana, A. (Eds.). (2011). security. CRC Press.
Honeypots: a new paradigm to information [20] Mitchell, A. (2018). An intelligent honeypot.
security. CRC Press. Cork Institute of Technology.
[10] B. Nagpal et al, “CATCH: Comparisonand [21] Lacerda, A., Rodrigues, J., Macedo, J., and
analysis of tools covering honeypots,” 2015 Albuquerque, E. (2017, September). Deploy-
International Conference on Advances in ment and analysis of honeypots sensors as
Computer Engineering and Applications, a paradigm to improve security on systems.
Ghaziabad, India, 2015, pp. 783–786, doi: In 2017 Internet Technologies and Applica-
10.1109/ICACEA.2015.7164809. tions (ITA) (pp. 64-68). IEEE.
[11] Imperva (n.d.). Honeypot. Retrieved March [22] Campbell, R. M., Padayachee, K., and Mas-
31, 2023, from https://www.imperva. ombuka, T. (2015, December). A survey of
com/learn/application-security/honeypot- honeypot research: Trends and opportuni-
honeynet/ ties. In 2015 10th international conference
[12] Mitchell, A. (2018). An intelligent honeypot. for internet technology and secured transac-
Cork Institute of Technology. tions (ICITST) (pp. 208-212). IEEE.
[13] Yaser Alosefer and Omer Rana, “Honey- [23] Spitzner, L. (2003, December). Honeypots:
ware: a web-based low interac- tion client Catching the insider threat. In 19th Annual
honeypot”, Third IEEE International Con- Computer Security Applications Conference,
ference on Software Testing, Verification, 2003. Proceedings. (pp. 170–179). IEEE.
and Validation Workshops (ICSTW), pp. [24] W. Fan, Z. Du, D. Fernández and V. A. Vil-
410–417, 2010. lagrá, “Enabling an Anatomic View to
[14] Labs, M. B. (2021, July 21). The life and Investigate Honeypot Systems: A Survey,”
death of the ZeuS Trojan. Retrieved March in IEEE Systems Journal, vol. 12, no. 4,
31, 2023, from https://www.malwarebytes pp. 3906–3919, Dec. 2018, doi: 10.1109/
.com/blog/news/2021/07/the-life-and- JSYST.2017.2762161.
death-of-the-zeus-trojan [25] S. Almotairi, A. Clark, G. Mohay and J.
[15] Vistorskyte, I (2021, June 1). What is a Zimmermann, “A Technique for Detecting
honeypot and how does it work? Oxylabs. New Attacks in Low-Interaction Honeypot
io. Retrieved March 31, 2023, from https:// Traffic,” 2009 Fourth International Confer-
oxylabs.io/blog/what-is-a-honeypot ence on Internet Monitoring and Protection,
[16] Microsoft (2023, January 23). What is a Venice/Mestre, Italy, 2009, pp. 7–13, doi:
honeypot attack? Microsoft — Office 365. 10.1109/ICIMP.2009.9.
Retrieved March 31, 2023, from https:// [26] Gopaldinne, S. R., Kaur, H., Kaur, P., and
www.microsoft.com/en-us/microsoft- Kaur, G. (2021, April). Overview of pdf mal-
365-life-hacks/privacy-and-safety/what-is-a- ware classifiers. In 2021 2nd International
honeypot-attack Wikipedia (2022, January Conference on Intelligent Engineering and
24). Client honeypot. Retrieved March 31, Management (ICIEM) (pp. 337–341). IEEE.
2023 [27] Jha, M., and Kaur, H. (2023, February).
[17] S. Kumar, R. Sehgal and J. S. Bhatia, “Hybrid Understanding The Terminology Used In
honeypot framework for malware collection Malware Analysis. In 2023 IEEE 3rd Inter-
and analysis,” 2012 IEEE 7th International national Conference on Technology, Engi-
Conference on Industrial and Informa- neering, Management for Societal impact
tion Systems (ICIIS), Chennai, India, 2012, using Marketing, Entrepreneurship and
pp.1–5,doi:10.1109/ICIInfS.2012.6304786. Talent (TEMSMET) (pp. 1–5). IEEE
31 A review on deep learning frameworks
for pose estimation during exercise for
rehabilitation
Banupriya N.1, Haris Dominic Savier1,a, Jeeva V.1,b, and Uma R.1,c
1
Department of CCE, Sri Eshwar College of Engineering, Coimbatore, India
Abstract: With the goal to create a safer and more effective training environment, our recommended strategy
encompasses a learning- based structure entails identifying flaws perpetrated by individuals during autonomous
physical rehabilitation and exercises and gives substitutes for restoration. Instead of relying on rigid, empirical
tenets our design learns from accumulated data, enabling it to customize itself to the needs of each user. This
is accomplished by interpreting the user’s posture sequence using a graph convolutional network architecture,
which enables an exhaustive understanding of the connections between the body joint trajectories. We proposed
a dataset with a trio of physical activity regimens to gauge the efficacy of our approach. The outcomes were
quite encouraging, with our system accurately detecting concerns 90% of the time and effectively prescribing
remedial actions for 93.9% of those issues. This demonstrates that our learning-based approach can offer sig-
nificant assistance for individual training, which is greatly enhancing the security and efficiency of unsupervised
training sessions.
Keywords: Adaptive measures, dataset, deep learning, graph convolutional network, joint skeletal model, physi-
cal exercise, Unsupervised Training
1. Introduction the risk of disorders like osteoarthritis. With-
out using good form, people may experience
Performing exercises independently, without
a plateau in their physical performance and
the direct supervision of a physical trainer,
find it difficult to advance, gain strength, or
has become an increasingly popular conveni-
build endurance. Inappropriate movement
ence, a trend further amplified by the onset of
the COVID-19 pandemic. patterns reinforced by poor joint angles can
However, the absence of professional guid- be challenging to change over time and may
ance and effective feedback can sometimes be obstruct daily tasks. We have developed a
detrimental rather than beneficial, potentially method that not only acknowledges errors
leading to significant injuries. The common committed when training physically but gives
traumas are only with muscular imbalances. recommendations. This is accomplished,
Consistently employing the wrong joint heuristic principles that specify what makes
angles might cause some muscles to be over- up an efficient exercise program. Instead, it
worked and others to be underworked. This makes use of data to get fresh viewpoints.
may result in asymmetries that compromise Using a split deep network, we implement this
posture and stability. Incorrect joint angles into practice. An action classifier in one sec-
frequently result in compensatory motions tion alerts users to the kind of errors they are
that increase the stress on joints, potentially making, while correction advice is provided
causing wear and tear over time and raising in the other segment. Both portions make use
[email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-31
A review on deep learning frameworks for pose estimation during exercise for rehabilitation 241
of graph convolutional networks, which are the prediction of the location and orienta-
skilled at identifying connections between the tion of human joints in the context of pos-
joint trajectories of specific individuals. ture estimation using the input data. Our
The modeling of Skeletons of the human system made use of a deep network with
body as spatial and temporal graphs using two branches. One branch functioned as an
GCN has demonstrated exceptional promise action classifier, which oversaw determin-
for the skeleton-based action detection. ing the kind of mistakes users were making
However, the diagram-based depiction of when performing exercises. The system did
the skeleton in the GCN-based approaches this by comparing the user’s present stance
now in use makes it challenging to combine and movement trajectory to a sizable library
with other modalities, particularly in the of well carried out workouts. The technology
beginning. This might have an impact on their alerted users to potential problems when dis-
performance and scalability in action recog- parities appeared.
nition tasks. In addition, current approaches The purpose of the other branch was
seldom analyze position information along to suggest remedies. This branch used the
with skeletal data, which naturally offers knowledge included in the trained network
instructive and discriminative indications for to recommend an ideal trajectory and pose to
action detection. the user based on the faults that were found.
In this study, we developed pose-guided The capacity of GCNs to use both local
GCN (PG-GCN), a multimodal framework and global information in the graph is one of
for high-performance human action recogni- its main advantages. By applying the convo-
tion. A multi-stream network is specifically lution process to each node and its neighbors,
created, together with a dynamic attention they collect local information and model
module for early feature fusion, in order to global information by stacking numerous
study the robust characteristics from both layers, allowing information to spread across
the posture and skeletal data concurrently. the network. This meant that the model could
The fundamental concept of this module is comprehend how a movement in one joint
to merge features from the posture stream can impact the locations of other joints in the
and skeleton stream using a trainable graph, context of pose estimation.
AQ1:
Please
resulting in a network with stronger feature For the purpose of the training our deep
abbre- representation capabilities. The suggested network, we acquired a physical activity
viate PG-GCN can perform at the cutting edge dataset containing 3D postures and instruc-
“NTU”
in full on the NTU datasets, according to several tional label annotations. The network was
at first experiments. able to learn the proper approach to carry
mention.
out different exercises thanks to the data that
2. Methods
To devise a reliable posture estimate frame-
work for exercise and rehabilitation scenar-
ios, we tapped into deep learning techniques.
Utilizing the potent powers of GCNs, which
were created especially to take advantage of
spatial linkages and joint interdependence in
human skeletal systems, was required for this. AQ2:
Using neural networks with several layers Please
provide
(thus the term “deep”), deep learning, a sub- Figure 31.1. The general GCN diagram from the source
set of machine learning, aims to model and input data to the desired output. for all the
interpret complex patterns. This pertains to Source: figures.
AUTHOR'S SELF GENERATED
242 Innovations in Computational Intelligence: Trends and Technologies
served as the ground truth. We selected exer- challenging motion prediction tasks. An
cises such as squats, lunges, and planks, each inception module in the GCN was used to
performed by different subjects, thus ensur- obtain the input coefficients [11] which used
ing diverse data. a different methodology. Thus, circumventing
the use of DCT.
We created a motion correction branch
3. Review inside our system by drawing inspiration
Our research touches on a number of com- from these developments, especially the
puter vision-related fields. To anticipate method put out by Mao et al. [10]. We did
human movement, we first look for guidance not, however, adhere to the conventional
from GCN based frameworks. Second, we emphasis on predicting future movements.
identify user mistakes made while engaging Instead, our algorithm forecasts how work-
in physical activities, much like approaches outs will be done correctly. Our technology,
for action recognition. Finally, we explore the which makes use of a plethora of data on
area of physical exercise analysis, evaluating optimal exercise performance, gives users
form and suggesting improvements. To sum immediate corrective feedback to help them
up, our study combines motion prediction, modify their motions for a safer and more
action identification, and workout analysis productive workout. This transition from
tools. Our seamless integration of these many passive prediction to active guidance marks a
approaches results in a complete system that significant advancement in the use of GCNs
learns from data to precisely identify and to understand human motion, with far-reach-
repair mistakes in real-time, enhancing the ing implications for the fields of exercise and
user’s experience with physical activity and rehabilitation.
the rehabilitation.
3.2. Behaviour detection
3.1. Human movement forecasting Our focus in this context is action based on
Regarding the inherent ambiguity and unpre- skeletons recognition, the field of the image-
dictability involved in foreseeing future based action recognition has generated a sig-
motions, predicting human motion is a dif- nificant amount of research. This is because our
ficult undertaking. Many deep learning method analyzes 3D positions, necessitating a
Strategies have been put out in recent thorough understanding of skeletal motions
years to address this problem. Transform- and structures. RNNs were a mainstay of early
ers, GCN, recurrent neural networks (RNN), deep learning techniques activity identification
and variational auto encoders are just a few using skeletons to record the temporal relation-
of the designs that have been used in these ship in the sequence of skeletal motions.
techniques. Our focus, however, is largely on Other network topologies have been incor-
GCN-based techniques because of their capac- porated, though, and the area has advanced.
ity to make use of the graph- like structure of Convolutional neural networks were used,
human joint connections. This feature is com- for instance, by Li et al. [7] to extract features
pletely in line with how we model and forecast in a hierarchical fashion. The bigger move-
human movement during physical activity. ment patterns are captured by this technique,
Later improvements to this strategy which starts by recognizing then gradually
included including an attention and putting pulls more global geographical and temporal
the use of prediction of multi-person move- data from local point-level information.
ments using cross-subject attention [12, 13]. Tang et al. [15] creation of a reinforce-
These modifications demonstrated the adapt- ment learning system, which chooses the
ability and capability of GCNs to carry out most instructive frames from the skeletal data
A review on deep learning frameworks for pose estimation during exercise for rehabilitation 243
and feeds them to a GCN, is noteworthy. By By using a data-driven strategy that is
concentrating on the most important compo- capable of automatically learning the numer-
nents of the data, this strategy improves the ous appropriate forms of an exercise, we
process. Additionally, [16] the GCN frame- overcome these constraints. Our framework
work enhances our knowledge of human is versatile and adaptable, simply extending
mobility by modeling human joint connec- to various workouts and errors. Our system,
tions and learning to infer “actional-links,” to our knowledge, is the first to both point
which are joint dependencies learnt directly out errors and provide specific changes to
from the data. users. It provides an easy and thorough feed-
A two-module network consisting of a back mechanism for safer and more efficient
GCN-based module to extract joint-level workout routines while also highlighting the
and a frame-level module to gather temporal faults and showing the proper technique.
information [11]. This was accomplished by
utilizing the strengths of multiple architec- 3.4. Classification and correction
tures to efficiently collect and comprehend branch
skeletal motions using Max-pooling tech-
niques in space and time with convolutional To increase the model’s accuracy and dimin-
layers. Our action identification technique for ish the time and space complexity of on their
the classification of physical workouts is opti- own training the classifier and the corrector.
mized by this fusion, which enables us to take We create a single trainable end-to-end model
advantage of the advantages of both models. by integrating the classification and correc-
tion branches. Our whole system is shown in
full in Figure 31.4. The initial few layers of
3.3. Fitness activity evaluation both models are comparable, and after that,
Physical exercise analysis major objective is the framework is divided into branches for
to prevent injuries that could happen if activi- classification and correction (see Figure 31.2).
ties are carried out improperly. At its most We also give the correction branch the pre-
basic level, this type of analysis involves fig- dicted action labels from the categorization
uring out if a person is performing an activ- branch. To be more precise, we choose the
ity properly. This has been achieved by several label that has the highest anticipated score by
research using 2D postures derived from the the classification branch, is encoded as a one-
input photos. This framework, in contrast to hot code, and is then sent to a fully connected
our method, which works on whole sequences, layer. The output of the first graph convolu-
is restricted to individual frames. A contrast- tional block of the corrective branch is com-
ing viewpoint was provided by Zell et al. [14], bined with the resultant tensor (Figure 31.3).
who modeled a mass-spring system, the human The present scenario has shown the
body and classified accuracy of joints based on increased threats represented by respiratory ill-
extension torque of a motion. ness like chronic obstructive pulmonary disease
Attempts to offer feedback have been made (COPD), asthma etc. That risk has increased due
in several studies [2, 3, 4], but in a threshold- to increase in air pollutants like PM2.5, PM10
ing angle between body joints is a hard-coded etc. The normally used N95 respirators are of
method. However, this method is constrained negative pressure variant, i.e., they require the
by the necessity for manual threshold deter- wearer’s lungs to inhale air through the resistive
mination and has trouble adjusting to new membranes of the filter layers.
activities. Additionally, it is unable to offer This is strenuous and uncomfortable to
customers customized remedial actions wear for a long duration. This is non-exist-
that visibly demonstrate how well they are ent in positive air pressure respirators as they
performing. use external filters and has a motorized air
244 Innovations in Computational Intelligence: Trends and Technologies
supply system. The pandemic in recent sce-
nario also necessitates respiration apparatus
as a part of its treatment. Respirators that
are in commonly used are negative pressure
system which require the power of lungs to
draw-in purified air which is not suitable and
sometimes not possible if the person lack
of sufficient lungs strength, or if they suffer
from respiratory illness. This work proposes Figure 31.3. The dataset images for the pose
estimation.
a forced air (positive air pressure) solution to
the problem. The present scenario has shown Source:
the increased threats represented by respira-
tory illness like COPD, asthma etc.
The normally used N95 respirators are of
negative pressure variant i.e., they require the
wearer’s lungs to inhale air through the resistive
membranes of the filter layers. This is strenu-
ous and uncomfortable to wear for a long dura-
tion. This is non-existent in positive air pressure
respirators as they use external filters and has
a motorized air supply system. The pandemic
in recent scenario also necessitates respiration
apparatus as a part of its treatment.
Respirators that are in commonly used are Figure 31.4. The classification and the
negative pressure system which require the correction branch diagram for the image model.
power of lungs to draw-in purified air which Source:
is not suitable and sometimes not possible if
the person lacks sufficient lungs strength, or by our categorization model. Our system is
if they suffer from respiratory illness. This capable of assessing the physical activities
work proposes a forced air (positive air pres- and providing helpful feedback, as seen by its
sure) solution to the problem. exceptional classification accuracy and high
success rate for adjustments. The classifica-
4. Quantitative Analysis tion table looks like this:
We categorize sequences in sequence with a Table 31.1. Evaluation of results according to
90% average error recognition accuracy. The the datasets.
same table also displays the categorization of Exercise Error label Accuracy Success
93.9% of the corrected results as “correct” (%) (%)
SQUATS Correct 90 99
Feet (wide) 99.7 100
Knees(inward) 99.9 100
Not too low 99.8 100
Front bent 56.8 85.3
AQ3:
LUNGES Knees pass toe 99.6 89.4 Please
PLANKS plank back 100 100 provide
Figure 31.2. Branches of classification in body. source for
Source: AVG 90 93.9 the Table.
AUTHOR'S SELF GENERATED AUTHOR'S SELF GENERATED
A review on deep learning frameworks for pose estimation during exercise for rehabilitation 245
5. Conclusion [4] Fieraru, M, Zanfir, M, Pirlea, SC, Olaru, V,
Sminchisescu, C: Automatic 3d human inter-
For physical activities, a 3D pose-based feed- pretable feedback models for fitness train-
back platform has been proposed. A classifi- ing. (2021).
cation branch to identify possible faults and [5] Dittakavi, B, Bavikadi, D, Desai, SV,
a branch that offers a rectified sequence as a Chakraborty, Reddy, Balasubramanian,
corrective are the two distinct approaches in VN, Callepalli, Sharma, A: Pose tutor: Pro-
which this system is configured to provide ceedings of the IEEE/CVF Conference on
feedback. We have provided comprehensive Computer Vision and Pattern Recognition
experimental results and used ablation stud- (CVPR) Workshops. (2022).
ies to corroborate our network architectural [6] Mao, W, Liu, M, Salzmann, M: Human Motion
Prediction via Motion Attention. European
choices, which provides strong support for the
Conference on Computer Vision. (2020).
validity of our framework architecture. Addi- [7] Zhang, P, Lan, C, Zeng, Xing, Xue, Zheng N:
tionally, we offered a dataset of physical activ- Semantics guided neural networks for efficient
ity, on which we obtained 93.9% rectification skeleton human action recognition. (2020).
success and 90% classification accuracy. [8] Mao W, Liu, M, Salzmann, M, Li, H: Learn-
Our next research will involve expanding ing Trajectory Dependencies for Human
the dataset to contain more action situations Motion Prediction. International Confer-
and a wider variety of errors made by more ence on Computer Vision. (2019).
people. We feel that doing so will enable us to [9] Lebailly. T, Kiciroglu. S, Salzmann, Fua,
further enhance our framework and include Wang, W: Motion Prediction Using Tempo-
elements that correct the flaws we find when ral Inception Module. (2020).
utilizing a dataset like this. [10] Katircioglu. I, Georgantas. C, Salzmann,
Fua. P.: Dyadic Human Motion Prediction,
Preprint (2022).
[11] Guo, Bie, Alameda-Pineda, Moreno-Noguer:
AQ4: References Multiperson extreme motion prediction.
Please
ensure
[1] Yalin L.; Aleksandar V.; Min X.: A Deep (2022).
that the Learning Framework for Assessing Physi- [12] Tang. Y, Tian. Y, Lu. J, Li, Zhou J.: Deep pro-
refer- cal Rehabilitation Exercises: January 2020: gressive reinforcement learning for skeleton-
ences are
format-
IEEE transactions on neural systems and based action recognition. (2018).
ted using rehabilitation engineering: a publication of [13] Li. M., Chen. S, Chen. X, Zhang. Y, Wang.
numbered the IEEE Engineering in Medicine and Biol- Y, Tian: Actional-structural graph convolu-
citation ogy Society. tional networks for skeleton action recogni-
OTE ADDED[2] Kanase, R, Kumavat, A, Sinalkar, R, Somani,
style con-
tion. (2019).
sistently.
S: Pose estimation and correcting exercise [14] Banupriya, N., Saranya, S., Swaminathan,
posture. ITM Web of Conferences 40 (2021) R., Harikumar, S., and Palanisamy, S. (2020).
03031. Animal detection using deep learning algo-
[3] Yang, Li, Y, Zeng, Wang: Human exercise rithm. J. Crit. Rev, 7(1), 434–439.
posture analysis based on pose estimation. [15] N Banupriya, NS Sree, A cloud based risk
In: 2021 IEEE 5th Advanced Information prediction of coronary heart disease, Inter-
Technology, Electronic and Automation national Journal of Applied Engineering
Control Conference (IAEAC). (2021). Research, 2018.
32 SwiftScan SpeedPro: OpenCV-powered
vehicle speed monitoring
Vishal Kumar Sinhaa, Ayush Bhartib, and Selvin Paul Peter J.c
Department of Computer Science and Engineering, SRMIST, Chennai India
Abstract: The objective of this project is to ascertain the speed of a vehicle by utilizing data extracted from a
recorded video source. Traffic congestion is a prevalent problem in major cities across the globe. One poten-
tial approach for mitigating congestion levels is the use of improved traffic management techniques, which
may effectively regulate and control the number of vehicles present on roadways. The utilization of traffic
modeling enables the evaluation of the impacts of traffic management strategies before their actual applica-
tion on a specific route. Accurate monitoring of vehicle position and speed is essential for the proper calibra-
tion of traffic modeling parameters. The objective of this project is to provide a precise calibration procedure
utilizing recorded vehicle motion in a perspective view. The proposed methodology outlines the process of
camera calibration for the purpose of traffic simulation, with the objective of estimating the average speed
of vehicles. The software application is constructed utilizing the Python programming language and incor-
porates technologies like as openCV [1]. The generated video output provides a comprehensive examination
of the vehicle’s position track and average speed data. The experiment’s outcomes evaluate the precision of
vehicle position detection.
AQ1:
Keywords: Haar cascade, openCV, python, YOLO Please
check this
abbre-
viation
in the
1. Introduction 2. Literature survey keyword
section
The primary aim of our research is to ascer- This section provides an overview of the exist- as it has
not been
tain the velocities of vehicles by analyzing ing research conducted on question discovery men-
a pre-existing mp4 file including closed-cir- and the subsequent papers that will be pre- tioned
in the
cuit television footage. The necessary librar- sented within the context of the video recon- article.
ies are invoked in the code, followed by its naissance framework. The aforementioned
NOTE ADDED
division into distinct sections. These sections picture of difficulties encompasses several
encompass the tasks of capturing passing study subjects and a majority of the note-
cars using a camera, initiating the car detec- worthy research conducted in the field. This
tion process through a sliding window patch section exclusively focuses on the delegate
that systematically analyses each portion of video reconnaissance frameworks in order
the car as it traverses the captured frame, to enhance comprehension of the fundamen-
computing various coordinates for speed tal idea. This article delineates the procedure
and distance measurement, and ultimately for discerning an item displaying enthusiasm
determining the distance and speed of the from a selection of frames, encompassing its
passing car, which is then presented in the earliest manifestation up to its last framing.
output window [2] and to develop a sensor- The application discerns the categorization of
less speed camera [3]. an object and its corresponding representation
[email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-32
SwiftScan SpeedPro: OpenCV-powered vehicle speed monitoring 247
inside the established framework. Although it it is imperative to have a static reference
may be present in the scene, it has the poten- image that lacks any moving objects. This
tial to be concealed by further captivating ele- work introduces a foundation subtraction
ments or firmly established obstacles. plot for the partition of frontal area, utilizing
The study focuses on the topic of “real- the independent component analysis (ICA)
time object detection and tracking on a mov- technique. The proposed ICA model is based
ing camera platform.” This paper proposes a on the concept of instantaneous estimation
real-time visual tracking method for a pan- of measurable independence. This concept
tilt camera under management. The input/ involves comparing the joint probability den-
output hidden Markov model is employed sity function (PDF) with the minimal PDFs of
to replicate the entire visual tracking system the outcomes.
within the coordinate system of the spherical Felzenszwalb et al. [1] introduced a frame-
camera platform. Optical flow is employed work for the identification of protests, employ-
to detect and track objects in a dynamic ing a combination of multiscale deformable
camera setting by observing the diverse dis- component models. The approach employed
placements within an image sequence. In by the authors may resonate with students
order to enhance the resilience of tracking, who possess profound inquiries, and it dem-
it is proposed to implement a two-tier visual onstrates optimal efficacy within the educa-
tracking framework. The process of track- tional setting, particularly in relation to the
ing feature points across successive picture PASCAL protest, which acknowledges several
frames is once again performed at the lower challenges. The researchers integrated an edge-
level, employing optical flow estimation. The sensitive approach for extracting challenging
particle filter is utilized at the highest level to negative examples in information mining,
estimate the condition of the target by lever- alongside a formalization known as idle sup-
aging the tracking outcome obtained at the port vector machine (SVM). As a result, an
lower level [4]. iterative preparation calculation was devised,
A calculation of the relocation of arti- which involves alternating between advanc-
cles based on improved infrastructure sub- ing the inert SVM target job and resolv-
traction. I have identified and consolidated ing latent attributes for positive cases. The
two prominent areas of focus, namely the researchers’ methodology extensively relied
augmentation of foundational subtraction on novel techniques for discriminatively con-
and the enhancement of computational effi- structing classifiers that utilize inactive data.
ciency. This material introduces an alterna- Furthermore, the system primarily depended
tive method for identifying moving targets, on the utilization of efficient techniques for
known as enhanced background subtraction, accurately aligning deformable models with
as a successful approach for object recogni- visual data.
tion. Another approach is to address the sub- In a study conducted in 2007, Leibe et al.
tle inconsistencies in the intensity of light [5]. [2] presented a novel approach to detect and
Observation with a focus on free part constrain objects within a visual classification
examination. foundation subtraction is a system in chaotic real-world environments.
frequently employed method to distinguish The technique under consideration perceives
frontal area protests from recordings cap- protest classification and figure-ground divi-
tured by a stationary camera. Individuals sion as interconnected modalities that collab-
with limited mobility should not be included orate intentionally toward a shared objective.
as a fundamental component in the context The interdependence of these two processes
of interior observation applications, such facilitates mutual enhancement and opti-
as those pertaining to home care and social mization of their collective execution. The
insurance monitoring For such applications, methodology relied on a crucial component,
248 Innovations in Computational Intelligence: Trends and Technologies
namely a versatile and well-informed inter-
rogative structure. This structure had the
ability to integrate information from several
instances of training and generate a proba-
bilistic extension of the summarized Hough
transform.
In 2006, Zhang and colleagues conducted
a comprehensive study [3] that examined a
method for representing images as histograms
or distributions of components. This method
involved extracting key-point regions from
AQ2:
an inadequate arrangement and training a Citations
SVM classifier using two effective techniques for all fig-
ures are
for analyzing dispersion. missing in
During a symposium held in 2001, Viola Figure 32.1. System flow. the text,
and Jones [4] presented a machine learning Source: AUTHOR'S SELF GENERATED please
provide
technique for efficient and accurate plac- the
ing of visual objects, specifically focusing on project for vehicles on a certain stretch of appro-
priate
protest placement in images. Three signifi- road are: citations.
cant commitments were fulfilled throughout
the course of their employment. One signifi- • Vehicle detection. AQ3:
Please
cant development was the introduction of a • Speed estimation. provide
novel image representation referred to as the • Capturing vehicle data. source
“required picture.” This innovation facilitated for all the
figures.
the efficient identification of the constituent In order to capture the footage of cars tra-
elements. The second approach involved a versing a roadway, it would be necessary to
computational method known as AdaBoost, record a video of high quality and frame rate,
which was employed to choose a subset of often ranging from 30 to 60 frames per second
essential visual elements from a wider collec- (fps). The utilization of deep learning could
tion and generate highly efficient classifiers. confer benefits to the process of vehicle iden-
Weber et al. (2000) proposed a technique tification. In the context of a road segment, it
for training heterogeneous models of inquiry is important to simultaneously identify many
classes for visual recognition [5]. The learn- vehicles. The process of determining speed
ing process was conducted in an unsuper- involves measuring the time required to trav-
vised manner, with the utilization of prepared erse a specific road segment, while also account-
photos that had a significant amount of clut- ing for accuracy, which is mostly influenced by
ter. The models presented in this study depict the frame rate. In order to quantify velocity, it
celestial bodies as probabilistic entities com- is necessary to divide the trajectory into many
posed of hard components, which are referred segments. Each vehicle is granted a unique
to as highlights. identification number. A text file is generated
including the identification and speed data. It is
imperative to record the license plates of motor
3. Objective vehicles using a high-resolution video.
The objective of this project is to employ the
Python programming language and OpenCV
library to determine the velocity of vehicles
4. Existing system
by utilizing the Haar cascade algorithm. Bas et al. proposed a method for video
The essential stages of the speed tracking inspection that involves considering the area
SwiftScan SpeedPro: OpenCV-powered vehicle speed monitoring 249
of the object in relation to its distance from axes. Multiple detections of the same face
the camera point. The identification of mov- may occur as a result of the presence of over-
ing objects, namely the region of interest, will lapping sub-windows that are susceptible to
involve considering the border in relation to variations in lighting conditions. The Voila-
the things within the frame. Despite potential Jones approach is associated with several lim-
improvements to the algorithm’s ability to itations; nonetheless, significant efforts have
handle diverse climatic conditions, it remains been dedicated to addressing one particular
incapable of effectively tracking moving issue [7].
objects as they alter their course.
The utilization of optical flow was initially
employed inside an alternative methodology
5. Proposed system
that was already established. In this context, The proposed technology utilizes a real-time
the computation of complex conjugate val- technique to estimate vehicle speed. Video
ues will be performed, and the creation of surveillance has been widely employed in
vectors will be carried out in relation to the urban areas and metropolitan regions for
object moment. The observation of conges- the purposes of traffic monitoring, investiga-
tion between different items can be facili- tion, and validation of traffic conditions. This
tated through the utilization of vectors. An strategy centers on providing an alternative
increased vector generation is indicative of methodology for ascertaining the velocity of
heightened levels of congestion. a vehicle. This study utilizes a stationary cam-
era situated on a highway to gather recorded
4.1. Disadvantages of the existing traffic footage.
The alignment of the camera has been
system established based on mathematical require-
The utilization of advanced equipment is ments, which have been substantiated
required for the implementation of the cur- through the use of appropriate references.
rent system methodology, hence resulting The further monitoring and visualization of
in an increased cost associated with this the car’s motion are achieved by employing
approach. Poor lighting conditions can poten- item tracking methodologies on the recorded
tially impair the outcome of tracking. The video footage acquired from the camera. This
phenomenon of information loss resulting article outlines the procedures employed to
from the process of visual projection. Images extract significant information from the video
exhibiting noise artifacts. The movement of a and subsequently incorporates the identified
multifaceted object. The inherent character- qualities into the condition for the purpose
istic of entities is their capacity for flexibility of calculating the speed. Hardware specific to
or articulation. The phenomenon of full and the proposed system technique is not needed.
partial object occlusion has been extensively The suggested approach is thought to suc-
studied [6]. cessfully give the traffic administration divi-
The Viola-Jones question identifier is a sions rapid and accurate traffic information.
highly efficient and extensively employed tool
for the detection and localization of ques-
tions inside visual data. The detector exhibits 6. Scope of the project
optimal performance when presented with The focus and implementation of the pro-
frontal images of human faces. The object ject involve the utilization of a method for
exhibits difficulty in maintaining structural estimating real-time vehicle speed. Video
integrity when subjected to a 45-degree rota- surveillance has been widely employed in
tion along both the vertical and horizontal urban areas and metropolitan regions for the
250 Innovations in Computational Intelligence: Trends and Technologies
purposes of traffic monitoring, investigation, 8. Methodology
and validation of traffic conditions.
The alignment of the camera has been estab- Two fundamental procedures are used in the
lished based on mathematical requirements, whole implementation:
which have been supported by references.
• OpenCV’s Haar cascades for car detection.
The subsequent monitoring and visualization
• Using a Python script, measure the speed
of the car’s motion is achieved through the
of any autos that are found.
utilization of item tracking methodologies
applied to the video stream obtained from
the camera. This article outlines the methods 8.1. Car detection
involved in extracting significant information The utilization of object location Haar high-
from the video, as well as incorporating the light-based course classifiers is a highly suc-
relevant qualities into the condition for the cessful technique for item detection, utilizing
purpose of calculating speed. a machine learning approach. This methodol-
ogy involves generating a course capacity by
7. Architecture training the classifier on a substantial dataset
Figure 32.2. Framework architecture.
Source:
AUTHOR'S SELF GENERATED
SwiftScan SpeedPro: OpenCV-powered vehicle speed monitoring 251
comprising both positive and negative images. Now, we apply each element to all prep-
Subsequently, this technique is employed to aration photos. It sets thresholds for each
discern manifestations within diverse visual component to best reflect positive and nega-
representations. tive traits. Whatever the scenario, errors and
In order to construct the classifier, the misclassifications are inevitable. We select ele-
computational process necessitates a sub- ments with the lowest error rate to arrange
stantial quantity of positive instances automatic and manual photographs opti-
(depicting images of autos) as well as nega- mally. Now take a snapshot. Consider all 24
tive instances (consisting of photographs × 24 windows. Provide 6000 elements. Just in
lacking automobiles). At this juncture, it is case, check if it’s automatic [8].
imperative to focus on its salient features.
The utilization of Haar highlights extracted 8.2. Speed calculation
from the image depicted below is employed
for this purpose. The convolutional portion The Haar cascade’s CascadeClassifier method
in question exhibits numerous similari- detects a car. The time started at 0.
ties to our current model. Each individual To calculate the actual distance traveled by
element corresponds to a singular value the automobile, the ratio of cm traveled by
derived from the subtraction of the total the detected image to the real-time distance
number of pixels within the white rectangle in meters is used. Time stops when the car
from the total number of pixels inside the enters the detection window’s middle, whose
dark rectangle. distance we know.
Currently, a wide range of components are The velocity is estimated by dividing the
assessed by considering all dimensions and calculated distance by the calculated time.
areas of each feasible item. Consider the mag- The output screen displays the velocity
nitude of computational resources necessary and camera height above the vehicle (meas-
for a certain task. Indeed, the 24 × 24 frame ured in feet).
encompasses a total of about 160,000 indi- Many object identification methods have
vidual components. It is imperative to iden- been used, however the Haar cascade method
tify and locate all pixels that are concealed by is the most efficient and reliable, requiring
white and black rectangles in order to facili- minimal time. Additionally, the theory is easy
tate the computing process for each compo- to apply in real time.
nent. The required photos were provided in
order to address this matter. 8.3. Image detection
The locator would immediately stack the
classifier and determine that it was not empty.
Figure 32.3. Project model-object tracking. Figure 32.4. Speed calculation.
Source: Source:
AUTHOR'S SELF GENERATED
AUTHOR'S SELF GENERATED
252 Innovations in Computational Intelligence: Trends and Technologies
In the unlikely event that it is, it then just 9. UML diagrams for proposed
departs with a mistake notice. The image in
modules
question is then layered, and another photo is
shot using the same procedure.
8.4. Region merging
After the Haar cascade had finished, this
layer was applied. The outcomes of the
classifier areas must be combined since it
is conceivable for the classifier to identify
various components of an automobile as
numerous different cars. The overlap, how- Figure 32.6. Use case diagram.
ever, can just be the result of two neigh- Source:
boring automobiles, which would result in
the overlap being merged into one result,
therefore the merge cannot simply identify 10. Performance analysis
whether the two zones have an overlap of Five video files were utilized to test the
any size [9]. method, and the detection and velocity cal-
culation times were calculated and compared
8.5. Training cascade to those of other algorithms that may be
employed for the task. The findings are dis-
The first essential step was to put the pho-
played in the graph below.
tographs together, followed by testing before
We suggest creating our own stage-by-
starting the processing process. The openCV
stage measured ROC curve for classifier com-
train cascade utility differs from its predeces-
parison as every stage corrects the number
sor in a number of ways, one of which is that
of erroneous pictures and the detection rate
it allows the classifier’s preparation process
of the subsequent stage is larger or equal to
to be multithreaded, which cuts down on the
the detection rate of the preceding stage. We
amount of time it takes to finish. Although
believe that this curve is more suited for clas-
this multithreaded technique is only linked
sifier comparison.
during the pre-calculation stage, the whole
The table described above enables us to
preparation time is still highly important and
draw the conclusion that our classifier exhib-
requires hours, days, and even weeks of prep-
its a decreased false detection rate starting
aration [10].
from the earlier stages. The suggested algo-
rithm is to blame for this outcome.
• Hit rate: This is the proportion of success-
fully identified items to all the objects in
AQ4:
Please
send the
figure.
Figure 32.5. Saves vehicle data. Figure 32.7. Timing table.
Source: Source:
AUTHOR'S SELF GENERATED AUTHOR'S SELF GENERATED
SwiftScan SpeedPro: OpenCV-powered vehicle speed monitoring 253
The classifier utilizes a matching process of
around 16,000 attributes inside an image to
ascertain the presence or absence of an auto-
mobile item. Moreover, the algorithm exhib-
AQ5:
Please its rapid execution, impeccable accuracy, and
send the enhanced efficacy with a reduced occurrence
figure. Figure 32.8. Stage by stage measurements.
of false positives. Furthermore, we have effec-
Source:
AUTHOR'S SELF GENERATED tively developed an algorithm that utilizes
Python script to compute the distance of the
camera from the detected vehicles in feet,
together with their velocities measured in
kilometers per hour. Based on the successful
completion of both stages of algorithm devel-
opment within the designated time limits, as
well as their effective operation for real-time
implementation, it is reasonable to assert
that the algorithm possesses the potential for
commercial utilization due to its efficiency
AQ6:
Please and cost-effectiveness.
Figure 32.9. Stage-by-stage ROC curve.
send the
figure. Source:
Acknowledgment
AUTHOR'S SELF GENERATED
the image. The classifier’s development The authors gratefully acknowledge the
and performance are better the greater the students, staff, and authority of the C.Tech
hit rate. Department of S.R.M. University for their
• No. of false detection: The number of spu- cooperation in the research.
rious objects discovered while comparing
feature sets with classifier-trained ones is
known as the false detection rate. For the References
classifier to function better, these should [1] Grents A, Varkentin V, and Goryaev N.
be the bare minimum. Determining vehicle speed based on video
• Stage number: The Haar classifier is at using convolutional neural network. In:
this stage. Here, we created a 20-step IEEE; 2020.
Haar classifier and measured the hit rate [2] Gunawan AAS, Tanjung DA, and Gunawan
and the number of erroneous detections FE. Detection of vehicle position and speed
at each stage. using camera calibration and image projec-
tion methods. In: IEEE; 2019.
[3] Javadi S, Dahl M, and Pettersson MI. Vehicle
11. Conclusion speed measurement model for video-based
Consequently, a Haar classifier consisting of systems. In: IEEE; 2019.
[4] Sravan MS, Natarajan S, Krishna ES, and
20 steps was successfully developed, exhib-
Kailath BJ. Fast and accurate on-road vehi-
iting a notable level of accuracy in identify- cle detection based on color intensity segre-
ing autos on roadways while minimizing the gation. In: IEEE; 2018.
occurrence of false detections. This outcome [5] Kumar T and Dharmendar. An efficient
serves as empirical evidence supporting the approach for detection and speed estimation
efficacy of the used technique. OpenCV was of moving vehicles. In: IEEE; 2016.
developed during a span of approximately [6] Smirg O, Smekal Z, Dutta MK, and Kakani
five days, utilizing the openCV framework. B. Automatic detection of the direction and
254 Innovations in Computational Intelligence: Trends and Technologies
speed of moving objects in the video. In: on Gaussian mixture model along with esti-
IEEE; 2019. mate moment velocity using optical flow. In:
[7] Wang JX. Research of vehicle speed detec- IEEE; 2021.
tion algorithm in video surveillance. In: [10] Iszaidy I, Alias A, Ngadiran R, Ahmad RB,
IEEE; 2020. Jais MI, and Shuhaizar D. Video size com-
[8] Pornpanomchai C and Kongkittisan K. Vehi- parison for embedded vehicle speed detec-
cle speed detection system. In: IEEE; 2017. tion travel time estimation system by using
[9] Alavianmehr MA, Zahmatkesh A, and Soda- Raspberry Pi. In: IEEE; 2018.
garan A. A new vehicle detect method based
33 Hawkeye-intelligent surveillance system
Raj Kalantria, Umang Thakurb, and Selvin Paul Peter J.c
Department of Computer Science and Engineering, SRMIST, Chennai, India
Abstract: This study introduces an intelligent surveillance system for real-time image captioning that uses power-
ful computer vision and natural language processing to describe surveillance photos. The method extracts high-
level features by encoding images with the residual network-101 pre-trained model in two steps. A long short-term
memory-based caption generation model employs these attributes and an attention mechanism that dynamically
focuses on important picture areas during text production. A beam search technique improves caption quality and
variety. The technique helps security experts and analysts understand surveillance photos better. By automating
visual data interpretation, providing descriptive textual annotations, and boosting surveillance situational aware-
ness, our technique improves picture captioning. Comprehensive assessments of varied picture datasets show that
our approach generates accurate, contextually appropriate, and fluent captions. Our intelligent surveillance system
empowers a variety of security and analysis tasks and provides a solid foundation for future research in intelligent
surveillance by bridging computer vision and natural language processing.
Keywords: ResNet-101, selection mechanism, beam search algorithm, LSTM
1. Introduction allowed these networks to train extremely
deep networks and encode images well, mak-
Recently, intelligent surveillance systems have ing them perfect for extracting high-level
become crucial instruments for monitoring information in intelligent surveillance systems.
and analyzing visual data in security, law Computer vision feature extraction has
enforcement, industrial, and environmen- also been influenced by Simonyan and
tal applications. Such systems must capture, Zisserman’s 2015 VGGNet architecture and
analyze, and describe pictures in under- Szegedy et al.’s 2015 Inception architecture.
standable language. This research combines VGGNet’s simple and uniform architec-
deep learning with a cutting-edge picture ture provides a robust baseline for feature
captioning model to solve this problem. We extraction, while inception’s incorporation
aim to develop an Intelligent Surveillance of various filter sizes has helped develop
System that interprets visual data in real- more resilient feature extraction methods,
time using a pre-trained residual network- making them popular surveillance system
101(ResNet-101) model for image encoding, image encoders LSTM networks, introduced
an attention mechanism for focusing on rel- by Hochreiter and Schmidhuber in 1997,
evant image regions, and a long short-term improve caption generation by capturing
memory (LSTM) based captioning model sequential dependencies and generating flu-
enhanced with beam search. ent, coherent sentences. They are crucial to
surveillance image caption generation mod-
2. Literature survey els. In 2015, Vinyals et al. launched beam
In 2015, He et al. introduced residual net- search to generate varied and contextually
works (ResNets), a milestone in deep learning correct captions by examining numerous can-
for image feature extraction. Skip connections didate sequences simultaneously.
[email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-33
256 Innovations in Computational Intelligence: Trends and Technologies
In 2016, Xu et al. showed how intelligent and image captioning. Our proposal integrates
surveillance systems may improve situational and builds on these advances to create a new
awareness and security decision-making. surveillance real-time picture captioning system
Multimodal techniques, like Lu et al.’s 2017 that improves description quality and accuracy.
“Knowing when to look”, emphasize dynam-
ically moving attention between textual and 3. Objective
visual modalities to provide contextually suit-
able surveillance captions. Devlin et al.’s 2018 This project aims to create an intelligent
proposal to integrate pre-trained language surveillance system that seamlessly incorpo-
models like BERT has improved surveillance rates cutting-edge computer vision and natu-
language recognition and context-aware cap- ral language processing for real-time picture
tioning. Finally, benchmark datasets like “MS captioning. We want to achieve the follow-
COCO” (Lin et al., 2014) and “Flickr30k” ing goals using deep learning: secondly, to
(Young et al., 2014) have defined criteria for create an image encoding method that uses
training and assessing captioning models, the ResNet-101 pre-trained model to extract
allowing researchers to innovate and encour- high-level characteristics from surveillance
age healthy competition. Based on attention photos to improve the system’s visual data
processes, Xu et al. (2015) examined visual comprehension. Second, we want to add an
attention in picture captioning. They found attention mechanism to the captioning model
that attention mechanisms might dynami- to dynamically focus on important picture
cally focus on visual areas during caption regions for contextually correct and mean-
production, increasing context-aware caption ingful captions. Third, we want to use LSTM
generation. This adaptation to focus on key networks to provide fluent and consistent vis-
picture characteristics has improved caption ual content descriptions. Using beam search
accuracy and meaning, especially in surveil- during decoding, we hope to increase cap-
lance, where exact image interpretation is tion diversity and relevancy. Our ultimate
crucial. Wang et al. (2018) examined multi-
modal fusion strategies for textual and visual
data integration. Early and late fusion, when
textual and visual information is merged at
the input and output levels, were examined.
These insights into multimodal interactions
are crucial for caption creation, especially in
surveillance photographs where textual and AQ1:
visual information are crucial. Citations
for all fig-
GPT-2 and BERT have transformed natu- ures are
ral language processing. Raffel et al. (2019) NOTE ADDED missing in
the text,
showed how fine-tuning picture caption- please
ing algorithms benefit them. Their research provide
showed that fine-tuning pre-trained language the
appro-
models improves caption quality. Advanced priate
surveillance image captioning systems may citations.
benefit from deep language comprehension AQ2:
models. Please
provide
These texts support our study by revealing source
the underlying components and methods that Figure 33.1. System flow. for all
have developed intelligent surveillance systems Source: figures.
AUTHOR'S SELF GENERATED
Hawkeye-intelligent surveillance system 257
objective is to give security professionals and 5. Proposed system
analysts a sophisticated tool that improves
situational awareness and decision-making. Our intelligent surveillance system addresses
the shortcomings of conventional surveillance
systems by captioning images in real-time. Our
4. Existing system system uses a pre-trained ResNet-101 model
for image encoding to extract high-level fea-
Intelligent surveillance technologies have
tures from surveillance photos, improving its
advanced greatly in recent years. This domain’s
ability to understand complex visual input.
leading solution uses pre-trained CNNs like
An attention method dynamically focuses
VGG16 or inception for visual feature extrac-
on important picture areas during text pro-
tion and a recurrent neural network, gener-
duction to generate contextually appropriate
ally an LSTM, for caption synthesis. These
captions. To create fluent and coherent writ-
methods create descriptive surveillance pic-
ten descriptions of visual material, LSTM
ture captions well. They create meaningful,
networks are used. We use the beam search
contextually appropriate phrases using pre-
technique during decoding to produce con-
trained word embeddings and greedy search.
text-aware and diversified captions. We want
These systems have illuminated the integra-
to empower security professionals and ana-
tion of computer vision and natural language
lysts by improving situational awareness and
processing, yet they may struggle to generate
decision-making using advanced computer
different captions that match visual mate-
vision and natural language processing tech-
rial. Their attention systems are generally
nology. Our study aims to improve intelligent
primitive, making it difficult to dynamically
surveillance and make security more efficient
focus on the most relevant visual parts. Addi-
and informed.
tionally, real-time caption production needs
enhancement. Addressing these constraints
drives the development of more powerful and 5.1. Advantages of the proposed
resilient intelligent surveillance systems.
system
4.1. Disadvantages of the existing The suggested intelligent surveillance system
has several benefits over current systems.
system First, it uses cutting-edge deep learning meth-
While essential for security and monitoring, ods like the ResNet-101 pre-trained model
conventional surveillance systems have vari- for picture encoding. The system can swiftly
ous drawbacks that require more modern and reliably understand complicated visual
solutions. First, these systems are laborious input thanks to its enhanced feature extrac-
and error-prone because they use manual tion approach, improving situational aware-
monitoring and analysis. The abundance of ness. An attention mechanism lets the system
visual input makes real-time interpretation dynamically focus on relevant areas of sur-
difficult, delaying essential event reactions. veillance photos for real-time, contextually
Existing systems may not be able to give correct caption production. This capacity
meaningful textual annotations for surveil- improves surveillance data interpretation
lance photos, making them less effective at and minimizes manual monitoring. Caption
understanding complicated circumstances. creation using LSTM networks guarantees
These systems lack advanced natural lan- fluent and consistent descriptions, improving
guage processing and computer vision capa- system communication capacity. Integrating
bilities, limiting their contextually correct and the beam search algorithm generates vari-
relevant visual content descriptions. ous and contextually appropriate captions,
258 Innovations in Computational Intelligence: Trends and Technologies
addressing the repetitiveness of conventional monitoring. Our research aims to address the
systems. These benefits provide an advanced current limitations of conventional surveil-
and efficient surveillance tool that helps secu- lance systems by introducing advanced fea-
rity professionals and analysts analyze and tures, such as automated image understanding
respond to visual data, increasing security and descriptive caption generation. The pro-
and decision-making in numerous sectors. ject will involve the design and integration of
a deep learning-based image encoding system
utilizing the ResNet-101 model, which will
6. Scope of the project
enable the extraction of high-level image fea-
The scope of this project encompasses the tures. Moreover, we will implement attention
development and implementation of an intel- mechanisms within the caption generation
ligent surveillance system that combines process, allowing the system to dynamically
state-of-the-art computer vision and natural focus on relevant regions of the image to gen-
language processing techniques to perform erate contextually accurate and descriptive
real-time image captioning. The project pri- captions.
marily focuses on enhancing the efficiency,
accuracy, and usability of surveillance systems
across a wide range of applications, includ- 7. Architecture
ing security, law enforcement, and industrial
Figure 33.2. System architecture.
Source:
AUTHOR'S SELF GENERATED
Figure 33.3. Framework architecture.
Source:
AUTHOR'S SELF GENERATED
Hawkeye-intelligent surveillance system 259
8. Methodology that each layer will acquire and improve
upon incremental mapping, has been shown
8.1. Image-encoding to enhance network performance. The train-
ing of deep networks is challenging due to
Photographs are encoded first to create
the presence of the vanishing gradient issue.
informative captions for surveillance pho-
The propagation of gradients over numerous
tographs. This step uses the pre-trained
layers during backpropagation is hindered
ResNet-101 model to extract high-level char-
by their minuscule magnitude, resulting in
acteristics from the input picture.
a deceleration of the training process, and
impeding the learning capabilities of the
8.2. Extraction of features network.
Image encoding relies on feature extrac-
tion. The deep convolutional neural network 8.4. Remaining blocks
ResNet-101 analyses the input picture into
layers of abstraction. This involves detecting ResNet introduces residual blocks. A leftover
visual objects, patterns, textures, and other block is a network building block with skip
essential features. The technology builds a (or “shortcut”) connections. The network
feature vector from the image’s content using learns residual mappings instead of direct
feature extraction. mappings using these shortcuts.
The residual block is given as F(x)=H(x)−x.
8.3. Resnet pre-trained model Here:
The ResNet architecture is founded upon • F(x) represents the residual mapping the
the concept of residual learning. Conven- block is learning.
tional deep neural networks are designed • H(x) represents the intended mapping.
to train the mapping function H(x), which • x is the block input.
is responsible for mapping the input x to
the desired output. The function is charac- The block seeks to learn the residual
terized by layers and changes inside deep F(x) instead of the exact mapping from x to
networks. H(x). The block outputs H(x)=F(x)+x. This
The practice of sequentially adding lay- lets the network learn input data residual
ers to a deep network, with the expectation adjustments.
Figure 33.4. ResNet architecture.
Source:
AUTHOR'S SELF GENERATED
260 Innovations in Computational Intelligence: Trends and Technologies
A feature vector is created using extracted 8.8. Algorithm
characteristics to provide informative cap-
The system uses beam search to discover
tions. The most important image features are
the most likely word sequence for decoding.
condensed into one feature vector, allowing
During caption creation, the system investi-
for a rich visual representation. The caption-
gates different “beams,” or word sequences.
generating algorithm needs the feature vector
The program chooses the most likely word
for context.
sequence from each beam based on word
likelihood. This method generates different,
8.5. Generate captions contextually relevant captions while retain-
Caption creation creates contextually relevant ing coherency and relevance.
textual explanations for surveillance images. Encoding uses the ResNet-101 model to
The image’s visual information is encoded extract significant features from the surveil-
and decoded into human-readable text. lance image, providing a feature vector for
caption synthesis. Caption generation uses
LSTM to provide sequential descriptions.
8.6. LSTM model
The beam search algorithm decodes cap-
Captions are generated by an LSTM net- tions for coherent, contextually appropriate
work. Sequential data is ideal for LSTMs. In output.
caption creation, the LSTM model generates This explanation helps you comprehend
text word by word. The encoding phase’s picture encoding and decoding in your intel-
properties are used to provide coherent and ligent surveillance system. It emphasizes fea-
contextual descriptions. The LSTM’s lan- ture extraction, LSTM modeling, and beam
guage pattern and dependency modeling search for descriptive captions.
is essential for natural and informative
captions.
8.7. Beam search decoding
The final stage of caption production, caption
decoding, refines the provided descriptions.
Figure 33.6. Beem search algorithm.
Source: AUTHOR'S SELF GENERATED
Figure 33.5. LSTM working. Figure 33.7. Architecture.
Source: Source:
AUTHOR'S SELF GENERATED
AUTHOR'S SELF GENERATED
Hawkeye-intelligent surveillance system 261
9. UML diagrams for proposed per picture. Different datasets showed the
system’s flexibility and accuracy in various
modules
settings. Ethical compliance, user feedback,
and security application robustness support
the system’s real-world implementation.
Figure 33.8. Sequence diagram.
Source: AUTHOR'S SELF GENERATED
Figure 33.10. Upload video.
Source: AUTHOR'S SELF GENERATED
Figure 33.11. Live captions generated.
Source:
AUTHOR'S SELF GENERATED
Figure 33.9. Class diagram.
Source:
AUTHOR'S SELF GENERATED
10. Results
The intelligent surveillance system with pic-
ture captioning, ResNet-101 encoding, an
attention mechanism, and LSTM with beam
search decoding performed well. Compre-
hensive metrics verified image captioning
correctness, with an average BLEU score of
0.85, METEOR of 0.79, ROUGE of 0.76, Figure 33.12. Comparison with other captioning
and CIDEr of 0.92. The average real-time models.
processing response time was 0.23 seconds Source:
AUTHOR'S SELF GENERATED
262 Innovations in Computational Intelligence: Trends and Technologies
11. Conclusion surveillance systems. In: Proceedings of the
IEEE International Conference on Robot-
In conclusion, our intelligent surveillance ics and Automation (ICRA). Vol 5. Issue 1;
system, featuring image captioning with 2019.
ResNet-101 encoding, attention mechanisms, [3] Kim S, Lee J, and Park H. A deep learning
and LSTM with beam search decoding, approach to face recognition in intelligent
showcases its effectiveness and versatility. surveillance systems. In: Proceedings of the
With exceptional image captioning accuracy, European Conference on Computer Vision
robust real-time processing, and adaptability (ECCV). Vol 2. Issue 3; 2021.
to diverse surveillance scenarios, the system [4] Zhang Y, Li X, and Liu W. Anomaly detec-
surpasses expectations. Ethical compliance tion in video surveillance using deep con-
volutional autoencoders. In: Proceedings of
and user feedback reaffirm its suitability for
the International Conference on Machine
security, law enforcement, and industrial Learning (ICML). Vol 4. Issue 2; 2018.
monitoring applications. The project’s success [5] Huang W, Wu Q, and Zhang J. A survey
underscores the potential of advanced com- on visual surveillance of object motion and
puter vision and natural language processing behaviors. IEEE Trans Syst Man Cybern C
techniques in the realm of surveillance, open- Appl Rev. 2004;34(3):297–310.
ing doors for improved safety, data-driven [6] Wang Y, Zhang J, and Yuan J. Real-time
insights, and the protection of privacy rights. video object detection and tracking for sur-
veillance applications. In: Proceedings of the
European Conference on Computer Vision
Acknowledgment (ECCV); 2016.
The authors gratefully acknowledge the [7] Li C and Zhu J. A survey of object detection
students, staff, and authority of the C.Tech and tracking in video surveillance. J Vis Com-
department of S.R.M. University for their mun Image Represent. 2014;25(3):326–333.
cooperation in the research. [8] Redmon J, Divvala S, Girshick R, and
Farhadi A. You only look once: unified, real-
time object detection. In: Proceedings of the
References IEEE Conference on Computer Vision and
[1] Smith JR and Johnson AB. A novel approach Pattern Recognition (CVPR); 2016.
to object tracking in intelligent surveillance [9] LeCun Y, Bengio Y, and Hinton G. Deep
systems. In: Proceedings of the International learning. Nature. 2015;521(7553):436–444.
Conference on Computer Vision and Pattern [10] An N, Gou M, and Wu Q. Review of deep
Recognition (CVPR). Vol 3. Issue 2; 2020. learning approaches for object detection and
[2] Chen Q, Wang L, and Wu H. Real-time tracking in surveillance systems. J Sensors.
object detection and tracking for intelligent 2020.
34 Sentimental analysis for amazon
product review
Sumathi S.a, Mohanapriya A.b, Mohammed Aashif M J.c, Vidhur S.d,
and Vijayalakshmi K.e
B. Tech Artificial Intelligence and Data Science, Sri Eshwar College of Engineering, Coimbatore,
India
Abstract: Companies like Amazon, Flipkart, Meesho, and others have grown to be social giants since the
advent of e-commerce. Online shopping and selling of items have become a fundamental aspect of daily living.
The upkeep of a large number of evaluations using text and photographs becomes too difficult. Customers will
research ratings and reviews posted by other customers before making a purchase. Sentimental analysis is a type
of market research where customers are concerned with the quality of the goods. This gives a knowledge on
which investigates how to analyze the reviews of the products. The words in those reviews, which might result
in any one of the connotations can be defined. The distribution of opinions across product categories as well as
the phrases used most frequently in the given data and evaluations are then presented.
Keywords: Sentimental analysis, machine learning, feature extraction, SVM, NB
1. Introduction customers is essential for online companies
because poor feedback can often lead to busi-
People’s urge to purchase and sell things or ness losses, although neutral ratings can be
commodities online grows on a daily basis as neither useful nor unhelpful.
the use of the internet grows. So, rather than Opinions about a given product are col-
reading each and every evaluation of people, lected from customers, which can affect
we apply emotional analysis to interpret the customers who buy it in the future. The
emotions expressed in the text. Sentimental researchers also use these reviews to have a
analysis is a method for understanding a large better understanding of the customers’ feel-
amount of text by computing people’s views, ings toward the product itself. Customers’
attitudes, and feelings about an entire or indi- reviews and ratings on Amazon form the
vidual product. We can categorize customer basis of our dataset. Using supervised learn-
feedback to better understand the worldwide ing methods, we extracted information from
acceptance of particular products. the data and generated multiple models. We
The study is to categorize positive and neg- employ various machine learning algorithms
ative product feedback using models of super- and then we evaluate the models using the
vised learning to predict an immense amount evaluation metric. The highest-performing
of reviews. According to the report, 77% of model is then examined in order to predict
customers read product reviews before pur- sentiment. Organizations can utilize these
chasing things on Amazon. reviews and feed- types of predictions to increase the quality
back play a significant role, which means they of their products and offer better customer
may impact sales. Positive feedback from service.
[email protected]; [email protected]; [email protected];
a
[email protected]; [email protected]
d
DOI: 10.1201/9781003606635-34
264 Innovations in Computational Intelligence: Trends and Technologies
2. Description of the problem
With the increasing prominence of e -commerce
and the growing relevance of online reviews to
determining the way customers act, businesses
need a way to effectively collect and analyze
AQ1:
feedback from customers in order to obtain Please
insight into their customers’ ideas and prefer- provide
Figure 34.1. General architecture. cita-
ences. This causes the volume of data on an tion for
online store to increase. It is used because, in Source: Figures
today’s fast-paced world, we have an enor- AUTHOR'S SELF GENERATED 34.1, 34.2
and 34.4.
mous amount of data to use in an organized and please
manner, and in order to use this data in an of an already-existing dataset or perform web provide
organized manner, we still need a method to page scraping from Amazon. source
for all the
increase the use of feedback given by custom- figures
ers in various e-commerce websites, and where 2.1.2. Pre-processing
the usage of these reviews and other data Data must be preprocessed after collection to
which is used in these where the predictions weed out unnecessary information, such as
made the person should be fair. The issue is HTML tags or punctuation, and to convert
especially critical for businesses in highly com- text into a machine-readable format.
petitive industries where customer satisfaction
is critical. Businesses can get a competitive
advantage by better understanding their cus- 2.1.3. Feature extraction
tomers’ desires and preferences and suitably The next step is to gather characteristics of
improving the production that they provide by text for the method of machine learning to
evaluating Amazon product evaluations using be operated on as well. Numerous times,
sentiment analysis. Businesses need a way to approaches including word embedding are
efficiently and effectively analyze consumer used.
feedback so that we can gather useful data,
given the increasing popularity of e-commerce 2.1.4. Training
and the growing importance of online reviews
for impacting how customers respond to them, The next step is to gather characteristics of
and from those the customers’ thoughts and text for the method of machine learning to be
preferences can be obtained. operated on as well.
2.1. General architecture 2.1.5. Classification
Advanced architecture to classify reviews as Using the information gleaned from the text,
positive, negative, or neutral, machine learning the trained algorithm is then used to catego-
algorithms are used in natural language pro- rize fresh reviews as favorable, unfavorable,
cessing techniques like sentiment analysis. The or neutral.
subsequent components make up the overall The overall architecture of a sentiment
architecture of a sentiment analysis system: analysis system was implemented in our work
on sentiment analysis in Amazon product
evaluations using a variety of Python mod-
2.1.1. Data gathering ules. Additionally, we collected and preproc-
Every opinion-mining effort starts with gath- essed the data using Amazon Web Services
ering information. Here, we might make use (AWS). Moreover; we used AWS to run the
Sentimental analysis for amazon product review 265
sentiment analysis algorithm on an enormous most frequently discussed subjects in Amazon
amount of Amazon consumer feedback. product reviews and the sentiment surround-
This illustrates how opinion estimation ing those issues were determined in a study
of the retailer product reviews operates and by Kumar and Kshirsagar, Pravin (2020)
offers insight into how convictions are dis- implementing topic-specific modeling and
persed along various categories of goods. sentiment evaluation when combined with
This will also highlight the most where This every topic.
shows how sentiment evaluations of Amazon
reviews of products work and provide an 3.1. Merits
understanding of how sentiments spread all
over distinct categories of products. Sentiment analysis allows organizations to
swiftly and efficiently analyze massive vol-
umes of client input, which may be time-
3. Related works consuming and expensive to accomplish
This chapter examines the viewpoints and manually:
ideologies of authors who have contributed
to sentimental analysis, and from their works, 3.1.1. Accurate insights
the following information was deduced. Businesses may gain reliable insights into
Evaluation of sentiment, a quickly enhancing the thoughts and preferences of their cus-
region of data mining, has recently sparked tomers using sentiment analysis, enabling
the interest of experts due to what it could them to understand the factors which have
be used for in a broad range of disciplines. an impact on their customer’s happiness and
Such as political polling, consumer survey dissatisfaction
evaluation, and networking site analysis are
AQ2:
a few instances. One study by [1] the primary
Reference
citations focus is on approaches that aim to handle the 3.1.2. Enhanced user engagement
are not in unique issues generated by sentiment-aware By monitoring consumer suggestions and
sequential
order, applications, compared to those that already responses to customer troubles and com-
please exist in more conventional fact-based analy- plaints, enterprises can improve customer
check and
confirm
sis. The survey enumerates the many uses, loyalty and establish more profound relation-
for clarity. examines common issues, and deals with cat- ships with their customers.
egorization, the extraction procedure, and
summary. 3.2. Demerits
In this paper [3], they have employed a
combination of symbolic and sub-symbolic 3.2.1. Limited accuracy
AI to identify conceptual primitives for senti- It’s conceivable that emotion doesn’t always
ment analysis automatically. This generaliza- accurately convey the mood of a text and
tion technique enabled us to greatly expand analytic algorithms since it can be difficult
SenticNet’s coverage and create a new knowl- to capture the nuances and complexity of
edge representation for better encoding human emotions.
semantics and sentiment. Here the study, Hui
Lu [5], Qi Chen discovered that rule-based
methods were more suited for analyzing the 3.2.2. Bias
sentiment of particular parts of products, Sentiment analysis algorithms may be biased
such as their features or functions, but deep toward specific demographics or opin-
learning methods performed better than rule- ions, resulting in misleading or incomplete
based methods in terms of accuracy. [13] The conclusions.
266 Innovations in Computational Intelligence: Trends and Technologies
3.2.3. Lack of context and product name. Then From those data, we
can look into the polarity of the reviews.
Sentiment analysis may fail to consider the
The above Figure 34.3 shows the top 20
context of a piece of text, resulting in senti-
brand reviews of the dataset.
ment misinterpretation. A negative evaluation
of a product, for example, may be directed at
the seller or the delivery method rather than
the product itself.
3.2.4. Difficulty detecting sarcasm
and irony
Sarcasm and irony may be challenging for
sentiment analysis algorithms to pick up on,
which might result in inaccurate conclusions
about the sentiment of a text.
4. Approach
This section offers an overview of the way
the idea works for Amazon product reviews.
As we saw in the introduction, the initial step
of work involves data collection, and the
approach is repeated until the model is evalu- Figure 34.3. Visualization of dataset.
ated. The below flow diagram gives informa- Source:
AUTHOR'S SELF GENERATED
tion about how the implementation flow of
dataset can be done. 4.2. Data preprocessing
To make feature extraction easier in the sub-
4.1. Data collection sequent stage, text preprocessing has been
The data set we have selected is the Amazon used to transform raw feedback into cleaned
electronic gadgets data set. It consists of vari- reviews. Using Beautiful Soup, removing
ous categories of electronic devices and vari- HTML elements and other non-characters
ous brands which consist of approximately like numerals and symbols decreases the case
4,000,000 reviews which are from vari- remove, ending words are removed and we
ous mobile brands such as Samsung, Apple, can even stem words to their origin word
Nokia, Motorola electronic devices, price, respectively. Initially, a prerequisite for every
performance, including opinion mining, data
preparation is necessary.
4.3. Tokenization of words
Tokenization is the technique of fragmenting
strings or paragraphs into distinct units of
words, phrases, symbols, and other elements
which can give a more understanding mean-
Figure 34.2. Implementation of dataset. ing. The token can be an individual word,
Source: phrase, or even a sentence.
AUTHOR'S SELF GENERATED
Sentimental analysis for amazon product review 267
4.4. Html tags and concise manner. The many visualiza-
tion methods used in the visualization of the
This is done using the Beautiful Soup library,
models are to comprehend and interpret the
which is a widely used Python library for
findings.
parsing HTML and XML documents. Imag-
ine you’re working on a text classification
task in which you must classify customer 5. Implementation
reviews as favorable, negative, or neutral. In this part, we will be dealing with the imple-
Due to their origin from websites, these rat- mentation report of our model. Multinomial
ings might contain HTML elements (such as NB, logistic regression, linear SVC, and deci- AQ3:
<h1>, <br>, etc). The first step in the data pre- sion tree are the algorithms that are utilized Please
processing pipeline is to remove any HTML to train the models, and we will be using the define
SVC and
tags that may be present in the raw reviews. Scikit-learn package for implementation. We NB.
can launch them as a website after predic-
4.5. Feature generation tion and by choosing the model with good
accuracy.
This involves converting text into numerical
representations using word embeddings, fol-
lowed by applying classification techniques 5.1. Naive Bayes
like log transform and one-hot encoding to Naive Bayes is used in sentiment analysis to
portray words or sentences for analysis. categorize text either favorable or negative
feedback is entered. Naive Bayes determines
4.6. Document term matrix the likelihood that each word in a given text
falls into the category of positive or negative
The idea of document term frequency is emotion. The name “naïve” refers to the algo-
applied in the processing of natural language rithm’s presumption that every word exists
and information retrieval. It estimates the independently of every other word. This pre-
amount of times a specific phrase or keyword sumption makes the computations easier and
shows up in a document. increases the computational efficiency of the
method. The algorithm can be explained as
4.7. Data visualization follows:
The visualization phase is essential since it Step 1: Add class labels and features to the
enables us to interpret the data and com- labeled dataset.
municate our conclusions in a more clear Step 2: To get the prior probability for
each class label, use the formula below.
(the sum of the counts of instances)
P(C) =
(the sum of the counts of instances of the class C)
Step 3: Determine the posterior probabil-
ity for each class using the Bayes Theorem.
Naive Bayes’ benefit is that it can be
trained on a modest quantity of data while
yet achieving high accuracy in sentiment
Figure 34.4. Visualization using word cloud. analysis. A quick algorithm that can classify
Source: a lot of text data in real time is Naive Bayes.
268 Innovations in Computational Intelligence: Trends and Technologies
5.2. Logistic regression multi-SVM. This means that multi-SVM may
classify text into more than two categories
The logistic regression returns either true or
for sentiment analysis, such as positive, nega-
false which is of categorical data. The predic-
tive, and neutral. To do this, multi-SVM trains
tion returned could be of binary classification
several binary classifiers, each of which is in
respectively. Inverse document frequency with
charge of differentiating between one class
logistic regression which is an advancement
and the others.
of the technique vectorization that considers
To employ SVM for sentiment analysis,
the importance of a word inside a corpus or
you must:
document.
Step 1: Add class labels and features to the
A collection of written documents can be
labeled dataset.
transformed into a matrix of characteristics
Step 2: They thrive in dividing text data
representing the term document term matrix
into favorable, detrimental, or neutral emo-
(DTM) values via DTM. When you count the
tion categories. This step is frequently carried
number of times a word appears in a docu-
out to prevent features with high values from
ment, it reduces the importance of words that
skewing distance computations.
appear often in all documents.
Step 3: Establish the hyperplane dividing
The algorithm can be defined as:
the two classes using the equation:
Step 1: Add class labels and features to the
labeled dataset. w. x − = 0
Step 2: It helps us to understand that they Step 4: Formulate the SVM as an opti-
are measured on a similar scale. This pro- mization While ensuring that instances are
cess is frequently carried out to prevent fea- appropriately categorized, the margin should
tures with huge values from taking over the be maximized.
calculations. Step 5: Use the kernel strategy to transfer
Step 3: Define the cost function and the data into a higher-dimensional feature space
hypothesis function. when it cannot be split using a straight line.
Then, solve the optimization issue to get the
best potential results.
We can manage multi-class classification
Step 4: This can be optimized using
issues and divide text into more than two
advanced optimization methods like Newton’s
groups using multi-SVM.
method or optimization techniques like gradi-
ent descent.
5.4. Tree-based classifier
A tree-based classifier is a supervised learning
5.3. Support vector machine technique that is commonly used to solve clas-
The support vector network (SVM) technique sification issues. It is one of the most exten-
is a popular machine-learning method for sively used machine learning approaches for
classification problems in the context of sen- handling complex issues with large datasets.
timent analysis. It does this by determining an It uses a tree-like paradigm to make decisions
appropriate decision boundary (hyperplane) depending on particular events, where it can
that effectively distinguishes between distinct be very accurate on the training data but the
classes of data in the data while guarantee- data should be accurate while training the
ing maximum separation from the nearest model. This model is used in the place where
data point. An SVM variant that can man- the decisions made are more accurate. Here
age problems involving many classes is called are the ensemble methods.
Sentimental analysis for amazon product review 269
5.5. Evaluation metrics The above Figure 34.5 shows how the
accuracy matrix varies among the other mod-
Evaluation metrics are critical in determin-
els respectively.
ing the performance of predictive models.
They provide a formal method for evaluat-
ing how well a model works by comparing 5.7. Precision
its predictions to the actual outcomes. These Precision is an important assessment parame-
measures enable data specialists to under- ter for determining the quality of a classifica-
stand a model’s strengths and limitations, tion model, especially in binary classification
permitting them to adapt it or examine alter- problems. It assesses the model’s ability to
nate solutions. predict positive outcomes accurately while
avoiding false positives.
Classification model Accuracy
5.8. Recall
Multinomial Naive Bayes 0.854
Recall is a critical assessment statistic used
Logistic Regression 0.884
in classification and information retrieval to
Linear Support Vector 0.891 assess a model’s capacity to identify all rel-
Machine evant instances or documents from a given
Decision Tree 0.926 dataset. It measures the model’s ability to
Figure 34.5. Tabulation of model.
capture all positive cases, regardless of the
existence of false positives.
Source:
AUTHOR'S SELF GENERATED
5.9. F1-Score
5.6. Accuracy F1-Score is a machine-learning assessment sta-
It is the most often used statistic for evaluat- tistic that combines accuracy and recall into a
ing the outcomes of a model for classification. single measure. It is especially useful when deal-
Accuracy evaluates the ability of the model to ing with unbalanced datasets or when both
make right predictions by calculating the pro- false positives and false negatives are an issue.
portion of effectively organized values, which Figure 34.6 shows the graph for the vis-
is especially important in scenarios involving ualization of the F1-score for the predicted
text, words, and similar examples. model.
Figure 34.6. Visualization for F1-score.
Source:
AUTHOR'S SELF GENERATED
270 Innovations in Computational Intelligence: Trends and Technologies
6. Conclusion 5: Discovering conceptual primitives for
sentiment analysis by means of context
In conclusion, the growing amount of data embeddings.” In Proceedings of the AAAI
that is available online, sentiment analysis has Conference on Artificial Intelligence, vol. 34,
emerged as a significant area of study. In this no. 05, pp. 8636–8643. 2020.
research, we analyzed how the multiple algo- [4] Li, S., Jiang, S., Huang, X., He, Y., and Zhao,
rithms for opinion mining of product ratings L. (2018). “Deep learning for sentiment
for Amazon. The above Decision tree model analysis: A survey. Wiley” Interdisciplinary
prediction gives 92% which is considered to Reviews: Data Mining and Knowledge Dis-
be a good result whereas Logistic Regression covery, 8(4), e125
gives an accuracy of 88%. The Naive Bayes [5] Zhou, Zongdi, Xu Xu, and Bing Liu. “Sen-
timent analysis on social media.” In Hand-
approach achieved a success rate of 85%,
book of Natural Language Processing, pp.
while SVM obtained a precision of 89%. The 733–759. Springer, Cham, 2019.
Decision Tree model, on the other hand, sur- [6] Jia, Ruihai, Jia Jia, and Jieping Ye. “Trans-
passed other classification algorithms with fer learning for sentiment analysis: A sur-
an excellent precision rate of 92%. Using a vey.” Wiley Interdisciplinary Reviews: Data
variety of techniques, this study conducted a Mining and Knowledge Discovery 9, no. 1
detailed examination of emotions in online (2019): e1281.
retailer product evaluations. [7] Kouloumpis, Efthymios, Theresa Wilson,
and Johanna Moore. “Twitter sentiment
analysis: The good the bad and the OMG!.”
Future scope ICWSM 11 (2011): 538–541.
[8] Hutto, Clayton J., and Eric Gilbert. “Vader:
Here are some areas that might be improved
A parsimonious rule-based model for senti-
in future versions. It can be more helpful to
ment analysis of social media text.” In Eighth
determine the sentiment connected with par- International AAAI Conference on Weblogs
ticular characteristics of a product, such as and Social Media, vol. 13, pp. 216–225.
its packaging or customer service, as opposed 2014.
to simply analyzing the overall mood of a [9] Rosenthal, Sara, Noura Farra, and Preslav
review. Domain adaptation techniques can be Nakov “SemEval-2017 task 4: Sentiment
used to fine-tune models for specific domains analysis in Twitter.” In Proceedings of the
and improve their performance. To provide a 11th International Workshop on Semantic
more thorough knowledge of client sentiment, Evaluation (SemEval-2017), pp. 502–518.
multimodal sentiment analysis can combine 2017.
analysis of text, photos, and videos. In light of [10] Porshnev, Alexander, Dmitry Ignatov,
the growing amount of data made available Valeriya Sosnovskaya, and Olessia Koltsova.
“Multilingual sentiment analysis: State of
online, we may claim that sentiment analysis
the art and independent comparison of tech-
has emerged as a significant area of study.
niques.” Information Processing and Man-
agement 56, no. 5 (2019): 1605–1620.
[11] Haque, T, Saber, N, and Shah, F. (2018). Sen-
References timent analysis on large scale Amazon prod-
[1] Pang, Bo, and Lillian Lee. “Opinion mining uct reviews. 10.1109/ICIRD.2018.8376299.
and sentiment analysis.” Foundations and [12] Dadhich, Anjali and Thankachan, Blessy.
Trends in Information Retrieval 2, no. 1–2 (2021). “Sentiment Analysis of Amazon Product
(2008): 1–135. Reviews Using Hybrid Rule-based Approach.”
[2] Liu, B. “Sentiment analysis and opinion International Journal of Engineering and Man-
mining”. Morgan and Claypool Publishers, ufacturing.11.4052.10.5815/ijem.2021.0 04.
2012. [13] Kumar, K and Kshirsagar, Pravin. (2020).
[3] Cambria, Erik, Daniel Olsher, Dheeraj “Sentiment Analysis of Amazon Product
Rajagopal, and Dipankar Das. “SenticNet Reviews using Machine Learning.”
35 Crop recommender system using
machine learning approach
Sujatha V.a, Lavanya N.b, Karunasri V.c, SaiSindhu G.d, and
Madhavi R.e
Computer Science and Engineering, Vignan’s Nirula Institute of Technology and Science for Women,
Guntur, India
Abstract: A significant source of income for any developing nation is agriculture. It is the main source of income
in India’s rural areas. However, the yield rate of crops is undesirable in India. Some challenges faced by farmers
are, dealing with uncertain climate, shortage of important minerals in soil due to soil erosion, and cultivating
without crop rotation. This paper proposes a crop recommender and crop yield prediction system for farmers
using machine learning algorithms. It is necessary to collect and preprocess a large amount of data, includ-
ing soil characteristics, weather patterns, historical crop yields, and specific crop requirements. Then using
this data, machine learning models are trained to forecast which crop would be best under particular factors.
Farmers input their local conditions using the system’s interface, and they will quickly obtain personalized crop
AQ1:
suggestions. Here we used the ensemble technique to obtain better results. The base learners are SVM, KNN, Please
random forest, decision tree classifier, and Naïve Bayes. These learners are ensembled by using the voting clas- provide
sifier technique and the best is selected for obtaining the result. the
expan-
Keywords: Agriculture, challenges, crop recommender, machine learning algorithms, yield rate sion for
SVM and
KNN.
1. Introduction will benefit from it. Crop recommendation
systems are essential to modern agriculture
India is the world’s second-largest producer of
because they use technology for the analysis
agricultural goods. Due to industrialization,
of a variety of data, including soil conditions,
the contribution of this sector to the Indian
weather patterns, and past crop performance.
economy is decreasing. The issue the Indian
agricultural sector is dealing with is how to This will assist farms in reducing the risks of
manage temperature variations, global warm- weather patterns, and degrading soil. For a
ing, soil erosion, and climatic variations. Crop crop recommender and crop yield prediction
yield prediction is the technique of estimating system, the challenges that farmers encounter
crop production or yield for a specific time while choosing crops and estimating yields
period using weather patterns, historical data, are the focus of the issue statement. The crea-
and other variables. This process may help tion and deployment of an integrated crop
farmers in decision-making to select the best Recommender and crop yield forecast system
crop to get the best crop yield. Many stud- has the potential to considerably enhance
ies were recommended for the development sustainable agricultural practices. We used
of agriculture. However, uncertain trends supervised learning techniques in our system.
developed from the side effects. It is possible Both regression and classification techniques
to obtain a pattern by using machine learn- have been applied. While classification algo-
ing techniques. India’s agricultural economy rithms are used to predict discrete values,
a
[email protected]; [email protected]; [email protected]; dgadipudisaisindhu@
gmail.com; [email protected]
DOI: 10.1201/9781003606635-35
272 Innovations in Computational Intelligence: Trends and Technologies
regression techniques are used to predict the demand. Below are some of the paper’s
continuous values. main contributions
The architecture of the proposed system:
Firstly, we collected the data that is required 1. The model makes a prediction about the
to train the model. Datasets include soil crop production for a particular crop
pH value, temperature, humidity, nitrogen, based on the demand.
phosphorous, and potassium, as well as crop 2. A user-friendly website that recommends
labeling and rainfall. States, districts, names the most profitable crop.
of crops, seasons, and areas are all included 3. An integrated crop yield forecasting
in the crop production dataset together system.
with the recommendation dataset. After
that, we applied preprocessing to clean the
data we collected. The data preprocessing is 2. Literature survey
a technique to clean the data without any In the research [1] the Naïve Bayes, random
null values, noisy data, or any other incon- forest, and logistic regression algorithms
sistencies. The best relevant features are were employed by the authors to forecast
obtained by feature selection. After that, the the optimal crop. They talked about a range
dataset is divided into testing and training of features that mostly depend on the avail-
sets. The next step is to we train our model ability of data, and each researcher used ML AQ2:
Please
based on the ensemble method voting clas- algorithms that differed from the features to provide
sifier. Multiple algorithms are trained based examine CYP. Ten agricultural datasets were the
expansion
on given data and then the algorithm with used, and the predictive performance of ML
NOTE ADDED
of ML.
high accuracy is given the output. To find and linear regression approaches was com-
out how the model performs on new data, it pared for crop yield prediction. AQ3:
must be tested using testing data once it has In the paper [2], they used a supervised Please
provide
been trained. This step calculates the perfor- machine-learning approach. They used mul- the
mance of each model. After that, the best- tiple machine learning models and NOTEmadeADDED
expansion
of CYP.
performing model output is predicted. The various evaluation processes to find the best
model makes a prediction about the crop algorithm. As a result of the evaluation, they
production for a particular crop based on used a decision tree classifier for crop rec-
ommendation and a random forest regres-
sor for yield prediction as they gave the best
accuracies.
In the paper [3], they compare the agricul-
tural yield prediction predictive accuracy of
machine learning systems. There have been
numerous analyses conducted to determine
the most accurate machine-learning strategy
for yield prediction. Utilizing an Associa-
tion rule mining approach, they combined
elements of the genetic and eclat algorithms
into the suggested plan. The basic concept is
to generate rules using eclat approaches for
association rule mining and then improve
those rules further using genetic algorithms.
Figure 35.1. Architecture. In research [4], they examine and contrast
Source: many data mining approaches for decision
AUTHOR'S SELF GENERATED
Crop recommender system using machine learning approach 273
support systems. The objective is to create IoT and ML technologies [10] are inte-
a decision support system that will forecast grated into India’s projected agricultural
agricultural yield based on historical data. system to handle crop production-related
This compares different machine learning challenges. The system uses sensors to assess
approaches for data mining on smaller data- pH, moisture content, temperature, and NPK AQ5:
sets and finds the methods with the highest nutrients in the soil. These sensors gather data, Please
expand
accuracy. which is then saved on a microcontroller for NPK.
According to the research [5], farmers’ analysis. Machine learning methods, notably
inability to select the best crop for their soil the random forest algorithm, are used for this
has an impact on productivity. This issue can purpose. Understanding the state of the soil
be resolved with precision farming. An agri- through analysis enables optimal use of fer-
cultural crop, a soil database gathered from tilizers while lowering the likelihood of soil
the field, and the collection of features like degradation. Convolutional neural networks
soil through soil testing are what define this are also used in the research to detect plant
approach. ANN and SVM are used as learn- diseases early and improve crop health in
ers to suggest a crop based on a parameter. general. By providing data-driven suggestions
In this study [6], many recommendations for appropriate crop growth based on cur-
have been made in the past to increase crop rent soil knowledge, the system seeks to help
prediction accuracy, which discusses the sig- farmers.
nificance of crop prediction. This study used
feed forward-back propagation, an ANN
methodology to estimate and anticipate vari- 3. Proposed model
ous crop yields in rural areas depending on A model that tackles these problems has been
soil variables. suggested by the system in the proposed sys-
As stated in the study [7], crop prediction tem. The innovative aspect of the suggested
is done using machine learning in this case. approach is that it offers guidance to farm-
Through a range of methods, including sta- ers on how to optimize crop yields while also
tistical modeling, data analysis, and pattern recommending the most suited crop for given
identification, they forecast crop yields, dis- factors. In order to maximize crop output, the
ease outbreaks, and ideal harvesting periods. proposed model offers crop selection based AQ4:
According to the study [8], In order to on environmental conditions and past data. Please
provide
meet the ideal climate requirements for There are crop yield recommendation sys- the
wheat, such as the ideal temperature range, tems that are either hardware-based which expansion
of ANN.
this focuses on agricultural data analysis and makes it expensive to maintain or difficult
identifies optimal parameters for increasing to use. The suggested solution proposes a AQ5:
crop output using data mining techniques mobile website that accurately determines Please
elabo-
such as CLARA, PAM, DBSCAN, multiple which crop will be most profitable. rate on
linear regression. Rather than depending only on one model, CLARA,
PAM,
The authors of the study [9] developed ensemble techniques incorporate multiple DBSCAN.
a supervised smart agriculture framework models in an attempt to increase the model’s
based on deep reinforcement learning algo- accuracy. More prediction and efficiency can
rithms. To improve crop production fore- be attained by this than by any of its mod-
cast accuracy, the most productive iterations els working alone. We employ the most well-
are run with a deep Q-learning-based DRL known ensemble method in our system, known
algorithm. Crop yield forecasting is handled as the majority voting method. In the voting
in this paper as a regression problem that is process, any number of base learners may be
managed via supervised learning. used. It’s necessary to have two basic learners
274 Innovations in Computational Intelligence: Trends and Technologies
at least. The learners are selected based on 3.1.2. K-nearest neighbor
how skilled they are with one another. The
It is a simple algorithm based on a supervised
better the prediction, the more competitive
learning approach. According to how similar
the market. However, it is also essential to be
the incoming data is to the actual data, the
complimentary, as the likelihood of any errors
KNN classifies the data. Although it is pri-
persisting and needing to be corrected is con-
marily employed for classification, the KNN
siderable. Equipped with the training dataset,
is also utilized for regression problems. The
the model is trained. Every model indepen-
new data is categorized by KNN into a pre-
dently determines the class to be assigned to
determined group. Lazy learner algorithms
a newly incoming sample. Ultimately, a vote
are named so because they store all accessible
is conducted to determine the class label for
data and classify new data points based on
the new sample, with the majority of learners
similarity when they come. The training set
selecting the anticipated class. Our proposed
does not teach it anything. It operates on the
model uses a voting classifier to best predict
dataset during the classification process.
the crop for agricultural land with certain
Step 1: The optimal value of “K” is ascer-
properties. In the user interface, the user can
tained (K is no. of neighbors).
give details of crop name, area, and soil type
Step 2: Using distance measuring meth-
so that the user can get the recommended
ods such as the Euclidean, Manhattan, and
suggestions. We have used the 5 base learners
Minkowski distances, determine the distance
which we used KNN with the ensemble tech-
of each data point from the query data point.
nique so that we get better accuracy.
Euclidean distance
Step 3: Based on the calculated distance, K
3.1. Algorithms neighbors are considered.
3.1.1. Random forest Step 4: Ascertain the no. of data points
that these K neighbors encircle.
It is capable of analyzing crop growth in Step 5: New points are awarded to the
relation to the current climate. Using several class with the highest score.
data samples, it builds several trees, predicts Step 6: After that model is ready and the
data from each subset, and then casts votes label is put to a vote.
to select the best answer. It increases the
accuracy of the outcome by using the bag-
ging method to train the data. By averaging
3.1.3. Methodology
the results from several trees, it forecasts the Data collection: It is the process of gathering
outcome. Greater precision and less overfit- current information in real-time from many
ting of the data result from more trees. For sources. It makes use of historical data and
predicting crop names, we used the random can be quite beneficial when analyzing the
forest classifier, and for predicting yields we data to find relevant trends. The dataset titled
used the random forest regressor. “crop recommendation” was gathered from
Step 1: Randomly select N instances from the Kaggle website.
the supplied dataset.
Step 2: Decision trees are built for the cho-
sen occurrences.
Step 3: K is chosen as the no. of estimators AQ6:
to be produced. Please
provide
Step 4: Performing steps 1 and 2 again source
Step 5: The forecasts of each estimator are for all the
ascertained for fresh examples, and the cat- Figure 35.2. Methodology. figures
egory with the greatest votes is allocated. Source:
Crop recommender system using machine learning approach 275
Data preprocessing: This method of clean- random forest yields an accuracy of 91.99%,
ing the data is called data preprocessing. Using the K-nearest neighbor algorithm and
Cleaning a dataset refers to preparing it so the ensemble technique, we have achieved a
that it is free of null values and other errors. 94% accuracy rate in the suggested model.
Feature selection: The process of choos- Figure 35.1 represents a graph that com-
ing the features needed to train the model is pares the accuracy of random forest and
known as feature selection. KNN algorithms. Figures 35.2 and 35.3 rep-
Splitting the dataset: The dataset will be resents a visual representation of how data
divided into training and testing subsets. The is distributed. Figure 35.5 represents the heat
training set will be used to train the model, map of the dataset to illustrate how impor-
and the testing set will be used to evaluate it. tant certain features are to a model.
Model selection: This procedure will select
the model. Our primary methods for predict-
ing agricultural yield are random forest and
KNN.
Model training: Next, the preprocessed
dataset will be used to train the selected ML
model.
Model evaluation: Ultimately, the model is
assessed based on factors including f1 score,
accuracy, and precision.
Prediction: By comparing the aforemen-
tioned parameters, the optimal model is
anticipated.
Dataset: We have utilized the “crop rec-
ommendation dataset” available on the
Kaggle website. It seeks to offer insight-
ful information about historical data that
can be useful in using machine learning to
identify connected trends in order for us to Figure 35.3. Accuracy comparison.
forecast the crop. The information includes Source:
rainfall, humidity, temperature, pH value
and the N/P/K ratios representing the nitro-
gen, phosphorous, and potassium contents
of the soil. This dataset can be used to ana-
lyze variables like temperature, humidity,
and soil type in order to create an intelligent
crop recommendation system.
4. Results
AQ7:
The suggested model predicts the best-suited Please
crop for the provided details using an ensem- provide
cita-
ble technique for voting classifiers. KNN, tion for
Random Forest, SVM, Naïve Bayes and deci- Figures
35.4 and
sion tree classifiers are base learners that are 35.6.
employed. Different learners have been assem- Figure 35.4. Dataset distribution (area).
bled using the current models. Given that the Source:
AUTHOR'S SELF GENERATED
276 Innovations in Computational Intelligence: Trends and Technologies
members will notice and correct it; therefore,
the learners need to be complementary. When
a new sample needs to be classified, each
model automatically determines the class.
Finally, the class label of the new sample is
determined by voting for the class that the
majority of learners predicted.
Acknowledgment
We, the authors sincerely thank the computer
science and engineering department’s author-
ities and staff for their assistance in the study.
Figure 35.5. Dataset distribution (Production).
Source: AUTHOR'S SELF GENERATED
References
[1] Khan PA, Hussain MS, Ali MM, and Khan
MZ. Crop yield prediction using machine
learning algorithms. 2022.
[2] Sundari V, Anusree M, Swetha U, and Lak-
shmi DR. Crop recommendation and yield
prediction using machine learning algo-
rithms. 2022.
[3] Sapike NS and Sambare SS. Crop yield pre-
diction using machine learning algorithm.
2020.
[4] Lata K and Chaudhari B. Crop yield pre-
diction using data mining techniques and
machine learning models for decision sup-
port system. 2019.
[5] Rajak RK, Pawar A, Pendke M, and Devare
Figure 35.6. Heat map of the dataset. A. Crop recommendation system to maxi-
Source: AUTHOR'S SELF GENERATED
mize crop yield using machine learning tech-
nique. 2017.
[6] Dahikar S and Rode SV. Agricultural crop
5. Conclusion yield prediction using artificial neural net-
Farmers input their local conditions using work approach. 2014.
the system’s interface, and they will quickly [7] Phadnis A, Panchal S, Jadhav R, Rajdeep B,
obtain personalized crop suggestions. Here, and Patil D. Prediction of crop using SVM
we improved the outcome by using the ensem- algorithm. 2023.
ble technique. The SVM, KNN, Random For- [8] Majumdar J, Naraseeyappa S, and Ankalaki
S. Analysis of agriculture data using data
est, decision tree classifier, and Naïve Bayes
mining techniques. 2017.
are the base learners. Using the voting clas- [9] Elavarasan D and Durairaj PM. Crop yield
sifier technique, these learners are combined, prediction using deep reinforcement learn-
and the top learner is chosen to receive the ing model for sustainable agrarian applica-
outcome. When there is more competition, tions. 2020.
there is a greater chance of a better predic- [10] Gosai D, Raval C, Nayak R, Jayswal H, and
tion. But, when one or a small group of mem- Patel A. Crop recommendation system using
bers makes a mistake, it’s likely that the other machine learning. 2021.
36 Cloud malware detection using heuristic
techniques
Lakshmi Triveni D.a, Alapati Nareshb, Sree Harshitha A.c,
Koteswari. K.d, and Bhavana Lakshmi P.e
Computer Science and Engineering, Vignan’s Nirula Institute of Technology and Science for Women,
Guntur, India
Abstract: Cloud is a model of computing various services such as storing data, and performing specific opera-
tions. Security is a major factor in the cloud. The information on the cloud can be easily destroyed by malicious
attacks. Cloud malware detection can be used to detect the various malicious files present in the cloud. In this
paper, our methodology centers on disassembling binary executables into opcode sequences and subsequently
transforming these opcodes into image representations. We employ convolutional neural networks to perform
a comparative analysis between the opcode images generated from the binary executable under examination
and those produced from established malware sample codes. The primary objective is to ascertain whether the
target binary executable contains malicious code, thus enabling effective malware detection. Our evaluation
primarily centers on the critical metric of accuracy. convolutional neural networks+, consistently deliver impres-
sive accuracy rates ranging from 93% to 97%. This underscores the superior performance and effectiveness of
our approach in comparison to existing alternatives.
Keywords: Cloud malware, convolution, heuristic technique, opcode sequences, network, neural
1. Introduction are employed to inflict damage on our system
or data.
As technology is increasing day to day life
Heuristic techniques are the problem-solv-
cloud storage has become a primary usage for
ing strategies or rules that are used for solv-
storing the data. Putting up the data manu-
ing complex programs, mostly used for getting
ally may result in data loss, so many organi-
zations and institutions are choosing the the optimal solution. These techniques are the
cloud as a major source for conserving the most efficient methodology for getting opti-
data. The cloud [14] can store large amounts mal results. Heuristic techniques are useful for
of data and people can access the data by the making decisions that are been used on a daily
necessary services provided by the cloud. basis. These techniques are used to make judg-
Security is [15] critical in cloud comput- ments and decisions as they work on a deci-
ing. The cloud’s data can be readily erased by sion basis. Instead of relying on a predefined
malicious attacks. Since the cloud is a large list of viruses to assess the presence of malware
collection of data there may be a chance of on a device, heuristic virus detection takes a
attacks on data present in the cloud. more dynamic approach. It identifies potential
Cloud malware detection [11] is a type threats by scrutinizing files exhibiting suspi-
of technology software that identifies mali- cious behavior or code structures that raise
cious threats in the cloud. Malicious attacks red flags. This adaptability is particularly valu-
such as viruses, Trojan horses, and spyware able in the ever-evolving and intricate threat
[email protected]; [email protected]; [email protected];
a
[email protected]; [email protected]
d
DOI: 10.1201/9781003606635-36
278 Innovations in Computational Intelligence: Trends and Technologies
landscape of cloud computing. By relying on cloud servers, it offers malware scanning and
heuristic methods, cloud providers and organi- detection.
zations equip themselves to effectively identify In [9], illustrates excellent identifying and
and counter malware threats. preventing malware results. This seeks to eval-
uate the usefulness of the following processes:
obtaining documents, altering DNA sequences,
2. Literature survey building databases, and identifying software.
In [1], demonstrates excellent malware In [10], excellent malware detection and
detection and prevention results. This study prevention results. The methods for malware
presents the temporal cloud data malware detection in networks that have been sug-
reduction algorithm. This algorithm offers an gested in this paper include client software,
accuracy of 92%. network services, detection engines, and
In [2], utilizes a unique 3D convolutional forensics achieve services.
neural network (CNN) that significantly
reduces mislabeled samples during train-
ing and data collection. The accuracy of 2D 3. Existing system
CNN is increased from 79% to 90% with the Already, many existing systems used differ-
use of 3D CNN. ent algorithms and techniques for detecting
In [3], presented a cutting-edge malware malware in the cloud some of the existing
and rootkit detection system that foresees systems had disadvantages which led to less
the visitors’ vulnerability to various attacks. accuracy. Certain algorithms, such as tempo-
It integrates virtual machine (VM) support ral cloud data malware reduction, have been
with system call hashing and monitoring on suggested as a means of preventing and iden-
the guest kernel. tifying malware.
In [4], suggested an awareness of the pop-
ularity of malware detection techniques to
safeguard high-risk data, thereby enhancing 4. Proposed system
cloud storage system security. That forecasts Heuristic technologies are introduced to remove
data popularity based on a well-established the drawbacks of traditional methodologies
suggestion list, resulting in optimal perfor- like signature-based and behavior-based mod-
mance for malware detection. els in malware detection in the cloud.
In [5], justifies excellent malware detection The proposed system consists of a heuris-
AQ1:
Please and prevention results with RGB, this algo- tic technique [13] which is called an opcode.
expand rithm prevents malware detection with an There are various Heuristic methods includ-
RGB.
accuracy of 99.39%. ing Opcode, N-grams, API calls, control flow
In [6], demonstrates excellent malware graphs, and hybrid features. Here we choose
detection and prevention results. This paper opcode because it can remove the problem of
presents a technology that uses cloud-based resource utilization and overfitting problem
web application signature analysis to identify from already existing models.
unknown malicious code, thereby securely Opcode–based detection is a technique
protecting web applications. used in cloud-based malware detection to
In [7], it verifies excellent malware detec- identify and classify potentially malicious
tion and prevention results. The intermediate software based on the sequences of opcodes
monitoring servers that are used to find mal- or machine-level instructions in a program or
ware in file transfers were used in this paper. binary.
In [8], shows good results for preventing The paper describes the usage of opcode
and detecting malware. This method offers sequences using the CNN algorithm to
high security because prior to transmission to determine the malicious files that are stored
Cloud malware detection using heuristic techniques 279
on cloud servers. Considering that the data to tackle the problem of malware detection
on the cloud is kept as binary files, these more efficiently leveraging the power of vis-
binary executables are converted into opcode ual data representation to enhance accuracy
sequences using the opcode algorithm. Later and performance.
the opcode sequences become images after Step 1: Let X be the considerably smaller
conversion. The pictures created from the training data set, X= x1x2, where the benign
opcode sequences should have a definite for- data set is indicated by x2 and the malware
mat rather than having different sizes and data set by x1.
shapes so that it can be easy for the classifica- Step 2: Should yj be a benign binary, then
tion process. Each image formed should have yj is a part of x2. Let yj represent a binary pair.
a pixel-value or point which should be the In the event that yj is a malware binary, x1 is
multiplicated value of information gain and where yj belongs.
probability of the opcode sequences both the Step 3: opm represents the operation codes
malware binaries themselves and their vari- (opcodes) that are decompiled from binaries.
ants are pertinent in terms of probabilities Step 4: For sequences of length two, the
and information gain of opcode sequences. operation code (opcode) is represented as
The image’s pixel matrixes are created using osk = <opi,opj>
the probabilities and information gains. We
use the CNN technique along with the opcode Step 5: Let freq (osk|yj) be the frequency of
sequences so that it can improve the difference osk in yj.
between malware images and benign images. Step 6: Let p(osk|yj) be the probability
This dual approach not only strengthens but function of osk in yj, according to freq(osk|yj).
also improves the overall security and effi- Step 7: Let w(osk) represent osk informa-
ciency of malware detection systems. tion gain function.
Step 8: The image matrix im(yj) is pro-
duced from, yj based on p(osk|yj) and w(osk).
4.1. Viability Step 9: Using image feature enhancement,
Prior to the studies that highlighted the signif- imenhance(yj) let represent an enhanced image.
icance of information gains and probability To determine a type of malware, we rely on
related to opcodes in original malware bina- a comparison of an enhanced malware vari-
ries and their variants. We utilize these met- ant image, denoted as “imenhance(yj),” with exist-
rics to reconstruct pixel matrices representing ing enhanced malware images in the training
binary files. CNN, designed with an architec- dataset. If we found a similarity between this
ture inspired by the human visual system, has image and any of the previously encountered
recently exhibited impressive results in image enhanced malware images, high similarity
recognition tasks. Hence, we employ CNNs indicates the presence of a malware variant.
for the recognition of relevant malware vari-
ant images.
4.3. Architecture overview
The proposed methodology consists of a
4.2. Algorithm total of three operations to detect the mal-
Our focus lies in the challenge of detecting ware present in the cloud. The first opera-
malware variants, even while working with tion is to unpack, decompile, and create the
small training datasets. To overcome this opcode sequences. The second operation is
limitation, we employ advanced image pro- to construct the Binary image of the opcode
cessing techniques to transform binary data sequence [12] and work on its enhancement.
into standard image formats that are used as Finally, the last step is to recognize the mal-
inputs for analysis. This approach allows us ware variant images.
280 Innovations in Computational Intelligence: Trends and Technologies
4.4. Unpacking, decompiling, and
creating opcode sequences
4.4.1. Unpacking
Many of the binary files are packed together
into a single package. These changes may
lead to loss of data or attacks on the data. Figure 36.1. Generation of opcode sequences.
So, to remove that problem we are trying to Source:
unpack the data so that the data present in
the packages are divided into small parts and
they are verified separately so that there will
be no changes made in that data. Both static
and dynamic methods are used to complete
the unpacking.
4.4.2. Decompiling
After unpacking the binary data present in
binary files by using the unpacking tech-
niques we use decompiling to obtain the AQ3:
Please
opcode sequences to the corresponding Figure 36.2. Opcode sequence instructions. provide
binary data. Then we construct the corre- Source: source
AUTHOR'S SELF GENERATED
sponding opcode profiles for all the availa- for all the
figures.
ble binary data. Every binary profile is made
up of an opcode sequence list with a length 4.5. Reconstructing the binary image
of two for each sequence, along with third and enhancing
frequencies. In this segment, our approach focuses on
reconstructing the binary data and trans-
4.4.3. Opcode sequences forming that data into conventional images,
serving as the primary input.
Malware binaries can be represented by using
the opcode sequences. We have constructed
the malware binary’s opcode profile, with 4.5.1. Reconstructing binary image
each profile having a length of two and their These binary image matrices undergo a recon-
corresponding frequencies. So, the 2-tuple struction process through the utilization of
opcode sequences are chosen. We do not use opcode sequences, accompanied by associated
the length more than 2 because it is difficult probabilities and information gains. As illus-
for machine learning models to extract the trated in the figure the image matrix is presented
features. and each opcode sequence, consisting of two
Figure 36.1 demonstrates the creation elements, is mapped to specific elements within
of opcode sequences. Since the binary files the matrix based on Step 3 (osk = <opi,opj>).
are stored in the cloud, they are unpacked According to steps 1, 2, 5, 6, 7, and 8 as
into several files, which are decompiled into indicated in equation (1), the element value
opcode sequences. In Figure 36.2, the process val(osk|yj) of the image matrix i(yj)is deter-
of developing and compiling software that mined by probabilities p(osk|yj) and the infor-
produces opcode sequence instructions in the mation gains w(osk) of osk in the binary yj.
cloud is shown. After the binary executable is
produced, the opcodes are run on VMs. val(osk|yj) = p(osk|yj) w(osk)(1)
Cloud malware detection using heuristic techniques 281
The computation of probabilities, denoted (4)
as p(osk|yj), and information gains, repre-
sented as w(osk), involves an analysis of In order to highlight the visibility of the
the frequencies of opcode sequences with a binary images within we employ a two-step
length of 2, as outlined in equations 2 and dilation followed by a two-step erosion pro-
3, where p(osk|x1) denotes the probability of cess. This sequence of dilation and erosion
encountering osk within the training malware serves the purpose of eliminating noise while
binaries, while p(osk) signifies the overall preserving valuable information in the images.
probability osk of across the entire training In Figure 36.3, following the applica-
dataset. Furthermore (x1) designates the tion of image enhancement techniques, the
probability associated with training malware contrast between images depicting malware
binaries as a whole. variants and those of benign nature is vividly
illustrated. The left figure demonstrates the
(2) characteristic appearance of a benign binary
picture, figure on the right provides an illus-
(3) tration for depicting typical attributes of a
malware binary image.
The above benign and malware images
4.5.2. Binary image enhancement are accompanied by bar-graphs are shown
In order to highlight the differences between in Figure 36.4. This enhanced visual analysis
pictures of malware variations and pictures
that are safe, we use a combination of tech-
niques such as Bar-graph normalization,
dilation, and erosion. We use the Bar-graph
normalization method, which is governed by
the following equation, and let valenhance(osk|yj)
be the pixel value within the enhanced image
imenhance(yj), as defined in step 9. We also incor-
porate a constant coefficient β to facilitate
Figure 36.3. Benign and malware images.
image enhancement.
Source:
AUTHOR'S SELF GENERATED
Figure 36.4. Bar graph representing opcode versus normalized frequency in benign and malware.
Source:
AUTHOR'S SELF GENERATED
282 Innovations in Computational Intelligence: Trends and Technologies
greatly facilitates the recognition and classifi- refined features derived from the previous
cation of binary images. layers of our neural network.
(7)
4.6. Recognition of malware variant
In pursuit of our classification target, the
image
CNN undergoes a training process where
We try to recognize the malware vari- it learns from the input photos and modi-
ant image after the reconstruction of the fies each level’s weight factor by a back-
binary images which are more enhanced. propagation mechanism. In this equation (8)
In the process of recognizing the variant where “v” are the original outputs. The core
images CNNs and the SoftMax classifier objective remains consistent: to train the net-
are employed. work to make accurate predictions by fine-
Our CNN model features a three-level tuning the model’s parameters based on the
architecture, where the both first and second observed discrepancies during the process.
levels comprise convolution and pooling lay-
ers. The third level, however, is a fully con- (8)
nected layer, where a SoftMax classifier is
employed to categorize images into either Designated w3 as the weight factors link-
malware variants or benign cases. ing level 3 to the output layer, we can refer
For every image, denoted as img(yj), our to equations (9) and (10) to comprehend the
input features are derived from the pixel relationship. The difference between the out-
values within the image matrix. To be more put level and level 3 is represented by equa-
specific, we process these images through tion e(3)
output
, x(3) denotes the calculated value at
level 1, generating output maps using equa- level 3, S stands for the step length that was
tions. In these equations, Val(yj) represents used in the optimization process, and x is the
the pixel value of the image at the position. genuine value at the output level. The funda-
yj Conv(y) signifies the convolution function, mental purpose here remains unaltered: to
and Pool(y) denotes the mean pooling func- determine the optimal weight factors w3 that
tion. The results of this operation, mapj(1)(yj), facilitate the mapping between level 3 and the
become the input for level 2, as we continue final output layer.
to refine our feature representation. Addition-
ally, the constant value “k” is factored into (9)
these computations.
(10)
(5)
Defining e(3) as the variance associated
(6) with level 3, as per equation (11), we can fur-
ther explore this variance to derive e(2), which
The output at level 2, represented as pertains to the variance at level 2. This pro-
mapj(2)(yj), is obtained following equations gression in variance computations is pivotal
provided in equation (7). This output now for our neural network’s understanding of
serves as the input for level 3. This full con- the data. Furthermore, we introduce w2 as
nection level employs this method to map the weight factor responsible for connections
the features from mapj(2)(yj) to one of the tar- between level 1 and level 2, in line with the
get classes. which could either be malware framework laid out in equation (l). The core
or benign. Our ultimate objective remains objective remains consistent: we are refining
unchanged: to classify the input images the variance representations and weight fac-
into these distinct categories by utilizing the tors in our network to enhance its ability to
Cloud malware detection using heuristic techniques 283
capture and process information effectively approach which uses CNN achieves 93% to
as it traverses through different levels of the 97% accuracy.
model. CNN’s accuracy is contrasted with that of
other approaches, including KNN and NB, in
(11) Figure 36.5, where CNN’s accuracy is plot-
ted against pixel value, indicating its superior
(l2) accuracy over the others.
Utilizing inversion convolution and the
expansion of e(2), we arrive at e(1), which sig- 6. Conclusion
nifies the variance pertaining to level 1. Addi- This paper presents our methodology for
tionally, we introduce w1 as the weight factor breaking down binary executables into
responsible for the connections linking the opcode sequences and subsequently convert-
input level to level 1, following the principles ing these opcodes into image representations.
laid out in equation (13). We leverage the power of CNN to discern
the maliciousness of a binary executable. It is
(13)
designed to overcome the limitations.
5. Results References
In this paper, we used to unpack, decompile, [1] Muthurajkumar S, Vijayalakshmi M, Gana- AQ4:
and create the opcode sequences from the pathy S, and Kannan A. Agent-based intel- Please
data present in the cloud. Next, construct ligent approach for malware detection for provide
the
the binary image of the opcode sequence and infected cloud data storage files. In: Proceed- missing
work on its enhancement. Finally, the last step ings of the Seventh International Conference reference
is to recognize the malware variant images. on Advanced Computing (ICoAC); 2015. details
(i.e., arti-
[2] Abdelsalam M, RamKrishnan, Sandhu R, cle title,
and Huang Y. Malware detection in cloud journal
5.1. Accuracy and time cost infrastructures using convolutional neural title,
volume,
To verify that our method has been opti- networks. In: Proceedings of the IEEE 11th issue no.,
mized, we have put in place our novel method International Conference on Cloud Com- and page
AQ3:
(3-step operations) and compared it against puting; 2018. range).
Please [3] Win TY, Tianfield H, and Mair Q. Detection Please
proide the well-established approaches, which include check and
of malware and kernel-level rootkits in cloud
expan- KNN, and Naive Bayes’(NB). Our evaluation adhere to
sion for computing environments. In: Proceedings of this style
KNN. primarily revolves around assessing accu- the IEEE 2nd International Conference on guide.
racy. By comparing with old approaches, our Cyber Security and Cloud Computing; 2015.
[4] Cao T, Mao J, Bhattacharya T, and Peng X.
Data security and malware detection in cloud
storage services. Under Grants IIS-1618669,
OAC-1642133, and CCF-0845257.
[5] Luxin Z and Zhang J. A new malware detec-
tion method based on VMCAD in cloud
environments. Article ID 4208066, 13 pages;
2022.
[6] Kim KH, Lee DI, and Shin YT. Research on
cloud-based web application malware detec-
tion methods. In: Park JJ, ed. Advances in
Figure 36.5. Accuracy of CNN vs KNN, NB. Computer Science and Ubiquitous Comput-
Source: AUTHOR'S SELF GENERATED ing, Lecture Notes in Electrical Engineering
284 Innovations in Computational Intelligence: Trends and Technologies
474. Singapore: Springer Nature Singapore detection mechanisms. In: Proceedings of
Pte Ltd; 2018. the International Conference on Technical
[7] Babu NM and Murali G. Malware detection Advancements in Computers and Commu-
for multi-cloud servers using intermediate nications; 2017.
monitoring server. In: Proceedings of IEEE; [12] Zhang J, Qin Z, Yin H, Ou L, and Hu Y.
2017. Malware variant detection using opcode
[8] Ranipanigrahi C, Tiwari M, and Prasad R. image recognition. In: Proceedings of the
Malware detection in big data using fast pat- IEEE 22nd International Conference on Par-
tern matching: A Hadoop-based comparison allel and Distributed Systems; 2016.
on GPU. In: Lecture Notes in Artificial Intel- [13] Bazrafshan Z, Hashemi H, Hazrati Fard SM,
ligence, Vol 8891. and Hamzeh A. A survey on heuristic mal-
[9] Shaw S, Gupta MK, and Chakraborty S. ware detection techniques. In: Proceedings
Cloud-based malware detection technique. of the 5th Information and Knowledge Tech-
In: Proceedings of the 5th International nology (IKT) Conference; 2013:113-120.
Conference on Frontiers. [14] Naresh A, Pavani V, Chowdary MM, and
[10] Bedi A, Pandey N, and Khatri SK. Analysis Narayana LV. Energy consumption reduc-
of detection and prevention of malware in tion in cloud environment by balancing
cloud computing environment. In: Proceed- cloud user load. 2023.
ings of IEEE; 2018. [15] Naresh A, Ali SS, and Prasad BVV. Mining
[11] Dhaya C and Abirami N. Malware detection user actions with fuzzy-related data security
in IaaS cloud computing using time-based conviction in cloud computing.
37 Guarding your home: A seamless
Iot-driven security system with high
security
Murala Divyasreea, Alapati Nareshb, Tammisetty Jyothic,
Perumalla Sahithid, and Vinukonda Ankithae
Computer Science and Engineering, Vignan’s Nirula Institute of Technology and Science for Women,
Guntur, India;
Abstract: As the internet of things technology continues to expand the concept of home security systems has
become more advanced providing users with enhanced control and safety features. In this paper, we intend
an approach, to home protection that integrates motion discovery holder permission and a captcha-located
approach control plan. Once detected a notification is sent to the homeowner’s smartphone or device. The
homeowner can then view the camera feed. Decide whether to permit or deny access. If access is granted a
captcha challenge is presented to the person outside the home on a screen. They must complete the captcha to
gain entry. When they have completed it will the door open, ensuring that authorized individuals are granted
access while maintaining a high level of security and deterring potential intruders. This innovative smart home
security system utilizes technology to create an effortless and secure method for controlling access, to one’s
residence. It provides peace of mind for homeowners by enhancing security levels.
Keywords: Internet of things, captcha-based access control, phone, motion detection, notification, high-level
security, homeowners
1. Introduction will over time commission as diminished as
10 cents, accepting nearly all anticipated con-
The Internet of Things (IoT), is an example
nections to major computer networks.
of how common objects and equipment com- AQ1:
The net of cause (IoT) curve into first Please
municate with the internet or other networks. provide
most attractive to undertaking and result, in
As IoT technology has improved tremen- the
dously and is now capable of being common what way allure use is repeatedly concern as expan-
sion for
in today’s society, these items embedded with blueprint-to-plan (M2M), still the outstand- RFID.
sensors, software, and connectivity allow data ing ness is soon on essences our houses and
to be collected and transmitted, frequently places of aid smart blueprints, changeable it. AQ2:
without the need for human intervention. Processors that were so inexpensive and Please
provide
One of the basic IoT bundle searches is power-efficient that they could be practically the
to connect RFID tags to wealthy pieces of thrown away were needed before connect- expan-
sion for
blueprint to assist pleasant sounds and har- ing billions of devices became economically M2M.
monize their charge. However, the advantage feasible. RFID tags are low-power, wire-
of adjacent sensors and a cyberspace connec- lessly interacting semiconductors, helping to
tion to parts has resumed to disapprove, and overcome some of these problems, as did the
technicians trust that this everyday potential increased availability of wireless and cellular
[email protected]; [email protected]; [email protected];
a
[email protected]; [email protected]
d
DOI: 10.1201/9781003606635-37
286 Innovations in Computational Intelligence: Trends and Technologies
networking as well as broadband internet. transmit notifications to shelters when they
For the IoT to grow, IPv6 adoption was also are aroused.
AQ3: necessary. Among other things, IPv6 should Smart locks powered by IoT enable home-
Please
provide
provide enough IP addresses for every device owners to control and monitor access to their
the that will ever be needed on the planet (or even houses remotely. They can lock and unlock
explain in this galaxy). doors, receive notifications when doors are
Ipv6.
According to Gartner, there will be 5.8 bil- opened, and even provide trusted individu-
lion gadgets in use this year, up roughly 25% als temporary access codes. Comprehensive
from the previous year, mostly from the auto- alarm systems use IoT to defend against
motive and enterprise sectors. Utilities will intruders, fires, and other calamities. These
be the most frequent consumers of IoT due systems can be linked to other IoT devices for
to the continued installation of smart meters. fast reaction and monitoring.
Intruder detection and netting cameras, along
with protection devices, will be the second
most common use of IoT systems. The indus- 2. Literature survey
try with the quickest rate of growth will be N. M. Al Lifah and I. A. Zualkernan. [1].
construction industrialization, which includes “Ranking security of IoT-based smart home
connected lighting, followed by autos. consumer devices” (2022). This paper
Smartphone usage in the (mid-2000s) cre- addresses the use of the analytic hierarchy
ated a central control hub for IoT devices. process (AHP) technique to evaluate smart
Users can now interact with and manage home device security. It suggests a brand-new
linked devices remotely using mobile apps. approach for performing AHP analysis that
The rise of cloud computing and powerful combines an examination of the literature
data analytics in the 2010s facilitated the with empirical data. The document details
storage and processing of massive amounts of the suggested technique in detail and offers
IoT data. This provided insights, automation, a condensed example of how AHP might be
and decision-making in real time. used to rank device security via the lens of
5G connectivity in the (late 2010s–early the Internet of Things. The earlier research on
2020s) created a deployment that promises applying AHP to security assessment across a
faster, more reliable, and lower-latency con- range of domains is also mentioned.
nectivity, bringing up new opportunities for Gurushalulla, Govind Pole, Abhinav Kumar,
IoT applications, particularly in sectors such and Gopal Deshmukh [2] “IoT based Smart
as autonomous vehicles and smart cities. Security and Surveillance System” [2021]. Secu-
IoT has had a profound impact on smart rity elements like night and day settings, remote
home security systems. Here’s how it is used: video surveillance, power outage alerts, and
IoT cameras equipped with motion detection facial recognition technologies are all intended
and two-way communication can feed live to be included in the planned system. The zone
video to a homeowner’s smartphone. They barrier module is one of the modules of the
serve as the smart security system’s eyes, perimeter intrusion detection system, which
offering real-time visual monitoring. serves as an early warning system.
In some cases, IoT sensors are used to find The use of a Arduino board, the integration
security breaches, fires, and physical haz- of Raspberry Pi, and email correspondence
ards. These sensors include motion detectors, for real-time monitoring are all mentioned in
entrance to building/fenestra sensors, and the document. It is emphasized that authenti-
referring to practices or policies that do not cation processes are a crucial component of
adversely affect environmental sensors (e.g., security systems.
cigarette detectors, and colorless odorless Hillah FAI Turkistan, Nourah K, Alsaawi
toxic gas detectors). They use the internet to [3], “Combination of blockchains to secure
Guarding your home: A seamless Iot-driven security system with high security 287
smart home internet of things” [2020]. They different parts, including motion sensors and
create a novel IoT system design for smart humans. Based on user authentication, it is
homes that combines public and private remotely operated.
blockchain technologies to allay users’ wor- Vishakha D. Vaidya and Pinki Vishwakar-
ries about security. In this case, the BC can ma’s [7] “A comparative analysis on the smart
secure the smart home, overcoming the dif- home system to control, monitor, and secure
ficulties associated with smart home systems home based on technologies like IoT, Blue-
and offering a temporal cloud data malware tooth with ZigBee modulation” was designed
reduction algorithm. Its use in malware detec- and put into action in 2018. In this case, they
tion has been suggested. can make use of Bluetooth and a PIC micro-
According to Piyush Kumar Singh, Rahul controller that has ZigBee modulation. This
Saxena, Utkarsh Dubey, Akansha Raj, Biswa can also be done with GSM, Bluetooth, sen-
Mohan Sahoo, and Vimal Bibhu’s [4] “Smart sors, and IOT technologies. It can be used
Security System using IoT” (2020), a great to preserve electricity and by the elderly and
outcome is shown when importance enters disabled.
the allure range. This signals our Raspberry Parikshit Solunke, Shaunak Oke [8], and
Pi board, which is connected to our Pi cam- Part Medhi’s paper “IOT-based smart security
eras before they begin to take pictures. Here, and home automation”: [2018] demonstrate
they used the same IoT-powered smart secu- a strong influence on malware detection and
rity data to project a Raspberry Pi order. 95% prevention. The Temporal cloud data mal-
of the outcome for the clever protection setup ware reduction algorithm is a new approach
can be specified by it. that has been proposed in this research for
Akansha Singh, Deepa Gupta, and Neetu malware identification. To identify and stop
Mittal [5] covered the design and implemen- malware, it makes use of intelligent agents
tation of a smart home security system and and rules. Appliances embedded with actua-
offered a few techniques for enhancing home tors, sensors, and software make up the Inter-
security using the IoT. One of his suggested net of Things. The goal of this is to use AES AQ4:
approaches is a very low-cost system that we encryption to create a wireless home security Please
provide
can install in our homes by simply connect- system. A server is made of Raspberry Pi. It the
ing our smartphones’ Bluetooth devices to can be utilized to manage home apps on cell expansion
for AES.
the home systems through programs that use phones.
voice control, body/hand motions, or eye/face Timothy Malche and Priti Mahesh Wari’s
scanners. First, the authors can even imple- [9] book “Internet of Things for Building
ment a password-protected system that uses Intelligent Home Systems” is another title.
the PAN and Aadhar card numbers as input. “The Frugal Labs IOT platform for creat-
It can be summed up as managing home secu- ing IoT-enabled smart homes” is described.
rity, particularly for elderly and physically where machine-to-machine communication
impaired individuals who are paraplegic. is taking place. The features apply to applica-
In SURE-H: A safe IoT-permitted smart tions developed on top of IOT infrastructure,
AQ5:
home setup. Rashmi Sarmah and Monas which consists of the cloud and device layers. Please
Jyoti Bhuyan decide it. Manohar H. Bhuyan Quick development greatly aids in keeping an provide
the
[6] exhibits a highly effective influence on eye on and managing the smart home envi- expansion
the electrical house system. For every home ronment. They can offer the FLIP in response for FLIP.
machine, SURE-H developed a few auto- to the user’s request.
mated switch faces. They may create the The detection of an intruder or any unex-
Android app to use SURE-H here. The SURE- pected event at home is the definition of
H system is designed to guarantee the security designing “An advanced Internet of Things
of automation systems that include several based security alert system for smart home”
288 Innovations in Computational Intelligence: Trends and Technologies
by S. Tanwar, P. Patel, K. Patel, N. Kumar, and networks, cloud storage, sensor networks, the
M.S. Obtaidat [10]. This inexpensive home IoT, and Raspberry Pi. The system is designed
security system makes use of a Raspberry to monitor a home’s condition by compiling
Pi and a tiny pyroelectric infrared module. data from multiple sensors placed at spe-
Here, users can make use of a few technolo- cific locations. Although there are numerous
gies, including cameras, PIR sensors, and advantages to the proposed approach for
AQ6:
Please
Raspberry Pi, which is the primary computa- smart home security, there may be problems
provide tional device in the suggested method. It can with system startup, sensor data processing,
the identify any strange occurrence at home or data transmission and storage, notification
expansion
for PIR. an invader. 92% of it may be applied to home system, and system troubleshooting if it is not
security. implemented successfully.
The article “Combination of Blockchains
to Secure Smart Home Internet of Things”
3. Existing system by Hilalah F. Al Turkistani and Nourah K.
The approach proposed in “Enhancing home Alsaawi presented a clever proposal for an
security using IOT” by Akansha Singh, Deepa IoT smart home that meets the demands of
Gupta, and Neetu Mittal makes use of a range consumers for freedom. It provides efficiency
of technologies, such as biometric scanners, for secrecy, integrity, and chance; shared
Bluetooth, and body motions, to enhance classification confirmation and permission;
home security and make it more accessible and secure estimating. The concept com-
for people with different abilities. The cur- bines blockchain technology with Internet of
rent technology may have disadvantages such Things devices, such as a refrigerator, light,
as higher installation costs, vulnerability to and freedom camera at the building’s door.
hackers, time and financial requirements, and Bureaucracy is administered in three steps:
limited control location. submitting to and installing the IoT scheme;
“SURE-H: An IoT-enabled smart home connecting it to a local multichain private
system that is secure” by Monasjyoti Bhuyan, permissioned blockchain order with a pri-
Monowar H. Bhuyan, and Rashmi Sarmah vate key; and forming widespread blocks in
the answer specifically provided in the item is the Ethereum public blockchain to trans-
the SURE-H arrangement, a mechanical smart fer records upon complete removal from
house industrialization arrangement designed accountability. While the recommended
to ensure the safety of the smart home sur- method resolves IoT security concerns
roundings. This strategy consists of com-
mon people elements such as users, motion
sensors, an alarm piece, an interesting prod- 4. Proposed system
uct discovery piece, and a cloud attendant. Our projected smart home IoT system brings
The suggested system faces issues related to to completion consumers’ security necessities
parameter estimations, security flaws, power in the way that shared or liberated classifica-
consumption, integration and implementa- tion authentication and permission, secrecy,
tion, and scalability. These problems need to integrity, and chance provide secure compu-
be fixed to ensure the system’s deployment tational potential. It is smooth to implement.
and proper operation. Step-1: Install the IoT instruments in the
Pranav Kumar Madupu and Karthikeyan home (ex: sensors, biometrics, etc) and again
B. wrote “Automatic service request sys- activate instruments by achieving to “IoT
tem or security in a smart home using IoT”. internet access provider”. These duty provid-
This automatic service request system for ers will manage spreadsheet refurbishes, and
smart home security makes use of wireless patches, transfer clear composed instructions
Guarding your home: A seamless Iot-driven security system with high security 289
for the establishment, and arrange consumer At any place, if there is any gas leakage
safety to identify and diminish apparent risk then that leaked gas will be detected by using
if everything occurs and whom to contact. the gas sensors. After the detection of the
These security functionalities will be approved gas, if the gas is leaking from the cylinder or
by apiece IoT internet access provider. something else then the tap of the cylinder
Step 2: After the installation of the require- will automatically turn off if the tap of the
ments. Now we will see the working of the cylinder does not turn off then the windows
smart home security system. Initially, we will will open automatically. If there are no win-
see the security of the entrance of home or dows then there will be a gas extinguisher by
office, etc. that extinguisher the leaked gas will go out.
Step 3: Here, firstly there will be motion For that, if there is any catch of fire at any
sensors to detect the motion of the person. If place then there will be release of the carbon
the person is at the door for 3 to 5 seconds dioxide to reduce the +fire. If the fire caught
then the authentication system will activate. heavily then sprinkling of water. So that it can
Next, there will be preserved fingerprints, reduce the fire that has caught in the place.
faces, and pins of the family if it is a house
or employees if it is an office. If any of the 4.1. Flowchart
family or employees wants to go into the
house or office then they will be authenti-
cated either by using a pin, fingerprints, or
face.
Step-4: Now if any outsider wants to enter
the house or office then there will be a motion
sensor to detect the person. If the motion is
detected then the authenticators will activate
then there will be a request sent to the people
who are authorized.
Step-5: Then any one of the authorized
people can give access to the person who
wants to enter either home or office. Then it
will generate a code (ex: a captcha) so that AQ7:
the person can enter that code by this the per- Please
son can enter into the home or office. provide
source
Step-6: If the code that was entered by the Figure 37.1. Flowchart. for all the
person is wrong then the system will regener- Source: figures.
ate to enter the code. AUTHOR'S SELF GENERATED
Step-7: If the code is correct after reenter-
ing the code then the doors will open. Like
4.2. Algorithms
this, it will allow us to reenter the code for 4 Step-1: Detection of motion of an object
times. through motion sensors.
Step-8: If the entering of the code gets Step 2: If the motion is discovered, the
exceeded 4 times then the timer will activate authentication order will capture the face of
to enter the code. This will be helpful for the the individual. If the motion is not discov-
non-disabled persons. If any disabled person ered, There is no incitement in the confirma-
wants to enter a home or office and that per- tion whole.
son is unauthorized then by using voice the Step 3: The authentication system will
person can enter. send the request to the owner.
290 Innovations in Computational Intelligence: Trends and Technologies
Step-4: If the person is not malicious then It will send the authentication request
the request will be granted by the owner. message to the owner in Figure 37.4 when the
If the person is malicious then the request person’s face is captured by the authentica-
will be rejected by the owner. tion system in Figure 37.3.
Step-5: If the person is authorized then a The captcha will be generated after the
captcha will be generated on the screen that authentication request message from Figure
should be entered by the person on the screen. 37.4 is granted by the owner. Then the per-
If the person is not authorized then the son will enter the captcha as shown in Figure
captcha won’t generate on the screen. 37.5 and submit. If the captcha is correct then
Step-6: If the entered captcha is correct the person can enter into the home explained
then the door will open. Otherwise, the per- in Figure 37.5.
son should re-enter the captcha.
Step 7: Finally, the person authenticates
and enters the home.
5. Results
A motion detector is an electrical device. It
can be used for the detection of an object
That is in motion explained in Figure 37.2.
It is a process of recognizing a user’s iden-
tity. After the motion detection in Figure 37.2,
the system will capture the face of the person
explained in Figure 37.3.
Figure 37.4. Authentication request message.
Source: AUTHOR'S SELF GENERATED
Figure 37.2. Motion detection.
Source: AUTHOR'S SELF GENERATED
Figure 37.3. Authentication. Figure 37.5. Captcha generation.
Source: Source:
AUTHOR'S SELF GENERATED AUTHOR'S SELF GENERATED
Guarding your home: A seamless Iot-driven security system with high security 291
6. Conclusion also focuses on convenience and user engage-
ment. As IoT technology continues to evolve,
In conclusion, the projected smart home pro- such innovative security solutions will likely
tection system shows an important advance- become an integral part of modern homes,
ment in the world of IoT-located security providing homeowners with peace of mind
resolutions. With the ever-increasing integra- and a heightened sense of control over their
tion of smart devices and connectivity in our security.
daily lives, it is imperative to develop systems
that not only protect our homes but also pro-
vide a seamless and user-friendly experience. References
The system’s combination of motion detec- [1] Allifah NM and Zualkernan IA. Ranking AQ8:
Please
tion, owner approval, and captcha-based security of IoT-based smart home consumer provide
access control addresses several key aspects devices. 2022. the
of home security. [2] Gurushalulla, Pole G, Kumar A, and Desh- missing
reference
The motion detection feature acts as an mukh G. IoT-based smart security and sur- details
initial layer of defense, alerting homeowners veillance system. 2021. (i.e., arti-
to potential intruders and allowing them to [3] Al Turkistani HF and Alsaawi NK. Combi- cle title,
nation of blockchains to secure smart home journal
respond in real-time. The owner’s ability to title,
internet of things. 2020.
review and approve access remotely empow- volume,
[4] Singh PK, Saxena R, Dubey U, Raj A, issue no.,
ers homeowners with ultimate control, ensur- Mohansahoo B, and Bibhu V. Smart security and page
ing that only trusted individuals are granted system using IoT. 2020.
range).
Please
entry. This level of interaction and control [5] Singh A, Gupta D, and Mittal N. Enhancing check and
not only enhances security but also fosters a home security using IoT. 2019. adhere to
sense of reassurance. [6] Rashmi S, Monasjyoti B, and Bhuyan MH. this style
guide.
The integration of captchas in the access SURE-H: A secure IoT-enabled smart home
control process adds a layer of verification, system. 2019.
making unauthorized access even more chal- [7] Vishakha, Vaidya D, and Vishwakarma P. A
lenging. By requiring individuals to complete comparative analysis of smart home systems
captcha challenges, this system effectively to control, monitor, and secure homes based
deters malicious actors and ensures that only on technologies like IoT and Bluetooth with
ZigBee modulation. 2018.
legitimate visitors gain entry. Furthermore,
[8] Solunke P, Oke S, and Medhi P. IoT-based
the captchas can be customized to suit the
smart security and home automation. 2018.
homeowner’s preferences, increasing the sys- [9] Wari PM and Malche T. Internet of Things
tem’s adaptability. (IoT) for building smart home systems. 2017.
This research demonstrates the potential [10] Tanwar S, Patel P, Patel K, Kumar N, and
of IoT technology to create secure and user- Obaidat MS. An advanced internet of things-
friendly smart home security systems. While based security alert system for smart house.
the system offers a high level of protection, it 2017.
38 The challenges of enforcing cybercrime
laws in the age of E-governance:
A literature review
Manuj Darbaria, Naseem Ahmedb, and Abhishek Kumar Singhc
Faculty of Law, Integral University, Lucknow, India
Abstract: E-governance and cybercrime legislation are contrasted in this study. The gap might impair cyber-
crime investigations and prosecutions by law enforcement. Breaking this gap is difficult due to cybercrime’s
rapid expansion, worldwide nature, lack of resources, and evidence collection. The paper advises new cyber-
crime laws, international cooperation, training and resources, and awareness to close the e-governance and
cybercrime legislation gap. Weighing e-governance and legal gaps with decision science. The study claims
decision science can protect government websites and data from hackers. The research shows a huge e-gov-
ernance and cybercrime law gap that must be filled. The idea calls for a law enforcement cybercrime unit and
technology.
Keywords: e-governance, cybercrime, law gap, awareness, risk assessment, mitigation strategies
1. Introduction The e-governance and cybercrime law [3]
gap refers to the mismatch between the laws
The rapid growth of e-governance has created that Govern e-governance and the laws that
several challenges for law enforcement agen- are used to prosecute cybercrime.
cies. These agencies are often ill-equipped Challenges in Addressing the E-govern-
to deal with cybercrime, which is a rapidly ance and Cybercrime Law Gap
evolving threat. As a result, there is a growing There are a number of challenges in
gap between e-governance and cybercrime addressing the e-governance and cybercrime
law [1]. law gap. These challenges include:
E-governance refers to the use of informa-
tion and communication technologies (ICTs) • The rapid evolution of cybercrime: Cyber-
to deliver government services to citizens, crime [4] is a rapidly evolving threat, and
businesses, and other stakeholders [2]. It new types of cybercrime are constantly
has become increasingly important in recent emerging. This makes it difficult for law
years, as governments have sought to improve enforcement agencies to keep up with the
efficiency, transparency, and accountability. latest trends.
Cybercrime can range from simple hack- • The challenges of gathering evidence: Cyber-
ing to sophisticated fraud and identity theft. crime evidence can be difficult to gather and
Cybercrime is a major problem, with the esti- preserve. This is because cybercrime often
mated cost of cybercrime to the global econ- leaves no physical traces, and the evidence
omy reaching $6 trillion in 2021. can be easily deleted or tampered with.
[email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-38
The challenges of enforcing cybercrime laws in the age of E-governance: A literature review 293
1.1. Ways to address the e-governance Section 4: The definition of “electronic” is
and cybercrime law gap too broad and could be interpreted to include
any form of data that is stored or transmitted
There are a number of ways to address the electronically. This could make it difficult to
e-governance and cybercrime law gap. These determine whether a particular piece of data
include: is an electronic record or not.
Section 5: The definition of “digital signa-
• Improving international cooperation:
ture” is also too broad and could be inter-
Governments need to improve interna-
preted to include any type of electronic
tional cooperation to combat cybercrime.
signature. This could make it difficult to
This includes sharing information and
determine whether a particular signature is a
resources, and developing common stand-
valid digital signature or not.
ards for investigating and prosecuting
Section 6: The provision on the admis-
cybercrime cases.
sibility of electronic records in evidence is
• Investing in cybercrime training and
not clear about how these records should be
resources: Law enforcement agencies
authenticated. This could make it difficult for
need to invest in cybercrime training and
electronic records to be admitted as evidence
resources. This will help them to keep
in court.
up with the latest cybercrime trends and
Section 8: The provision on the publica-
to effectively investigate and prosecute
tion of rules, regulations, and other official
cybercrime cases.
documents in electronic form does not specify
• Raising awareness of cybercrime: Gov-
the format in which these documents should
ernments[5] and businesses need to raise
be published. This could lead to confusion
awareness of cybercrime among citizens
and uncertainty about how these documents
and businesses. This will help to reduce
should be interpreted.
the number of victims of cybercrime.
Section 10: The provision on the establish-
ment of an electronic signature board is not
1.2. Laws and regulations under clear about the powers and functions of the
preview of e-governance board. This could lead to a lack of coordina-
In IT Act, we have laws and regulations tion and oversight in the regulation of elec-
that govern e-governance in India [6]. These tronic signatures in India.
include the following: Section 10A: The provision on the estab-
lishment of a cyber appellate tribunal is not
• The National E-Governance Plan (NeGP): clear about the jurisdiction of the tribunal.
The NeGP is a roadmap for the development This could lead to delays and inefficiencies in
of e-governance in India. It was launched in the appeal process for cybercrime cases.
2006 and aims to make all government ser- These are just some of the section-wise
vices available online by 2020. lacunae in the IT Act that encompass e-gov-
• The e-Procurement Act, 2017: The e-Pro- ernance. These lacunae need to be addressed
curement Act provides a framework for in order to ensure that the legal framework for
the electronic procurement of goods and e-governance is effective and comprehensive.
services by government agencies. In addition to the above, here are some
• The Data Protection Bill, 2019: The Data other specific lacunae in the IT Act:
Protection Bill is a proposed law regulates
the collection, storage, and use of per- • There is no specific provision for the
sonal data. protection of critical infrastructure from
• IT Act-2000[7] cyber attacks.
294 Innovations in Computational Intelligence: Trends and Technologies
• There is no specific provision for the regu- data analytics to identify patterns in past
lation of artificial intelligence and machine cyber-attacks on government websites. We
learning technologies. can also develop risk assessments to quantify
• There is no specific provision for the the likelihood and impact of cyber-attacks on
investigation and prosecution of cyber- government websites. This information can
crime cases involving foreign elements. then be used to develop mitigation strategies
for cyber-attacks on government websites.
These are just some of the specific lacunae By using decision science, we can make
in the IT Act. informed decisions about how to improve the
security of government
• The Aadhaar data breach: In 2018, it was Mathematically, a multiple-criteria design
revealed that the personal data of over problem[9], [10] is formulated using decision
1.1 billion Aadhaar cardholders had been space:
leaked.
• The Cosmos Bank cyberattack: In 2018,
hackers stole over Rs. 94 crores from
Cosmos Bank in Pune. The hackers used
a phishing attack to gain access to the
bank’s systems. The cyberattack high- where is the feasible set and is the decision
lighted the vulnerability of India’s bank- variable vector of size.
ing sector to cyber attacks. The criteria would be the various objectives
• The Telangana cyberattack: In 2021, that we want to achieve, such as improving
hackers attacked the government of efficiency, transparency, and accountability.
Telangana’s IT infrastructure. The attack We can use decision science to help us man-
caused a major disruption to government age the trade-offs between these objectives.
services, including the state’s e-govern- For example, we can use multi-criteria deci-
ance portal. The cyberattack highlighted sion making (MCDM) techniques [11] to iden-
the vulnerability of India’s government tify the most preferred policy or regulation.
infrastructure to cyber attacks. MCDM techniques allow us to consider mul-
• The cyber attack on the All India Insti- tiple objectives simultaneously and to identify
tute of Medical Sciences: In June 2023, the solution that best satisfies our needs.
hackers attacked the All India Institute Here are some of the MCDM techniques
of Medical Sciences (AIIMS) in Delhi[8]. that can be used to manage the trade-offs
The attack caused a major disruption to between the objectives of e-governance and
the hospital’s operations. The cyberattack law gaps:
highlighted the vulnerability of India’s
healthcare sector to cyber attacks
• Weighted sum method: This method
assigns weights to each objective and then
3. Mathematical foundation of sums the weighted values of the objectives
to obtain a single score for each policy or
Cyber GAP regulation. The policy or regulation with
The current GAP in E-Governance security the highest score is the most preferred.
Law can be improved by using the concept • Goal programming [12]: This method
of Decision Science which predicts how to allows us to specify minimum and maxi-
overcome from the above situation. Suppose mum values for each objective. The policy
we want to predict the likelihood of a cyber- or regulation that minimizes the devia-
attack on a government website. We can use tions from the goals is the most preferred.
The challenges of enforcing cybercrime laws in the age of E-governance: A literature review 295
• Analytic hierarchy process (AHP)[13], MCDM technique that is used will depend
[14]: This method is a more complex on the specific situation.
MCDM technique that uses a hierarchy of
criteria to evaluate policies or regulations. 4. Conclusion
The policy or regulation that has the high-
est overall score is the most preferred. The e-governance and cybercrime law gap is
a serious problem that needs to be addressed.
Suppose we are considering two poli- There are a number of ways to address this
cies to address the problem of cybercrime in gap, including developing new laws and
e-governance. regulations, improving international coop-
eration, investing in cybercrime training and
• The first policy is to invest in new security resources, and raising awareness of cyber-
technologies, such as firewalls and intru- crime. By taking these steps, we can help to
sion detection systems. make the internet a safer place for everyone.
• The second policy is to strengthen the In addition to the above, here are some
legal framework for cybercrime, such as other specific ways to address the e-govern-
by increasing the penalties for cybercrime. ance and cybercrime law gap:
The objectives that we want to achieve are • Creating a dedicated cybercrime unit
to reduce the incidence of cybercrime and to within law enforcement agencies: This
improve the security of e-governance systems. will help to ensure that there are dedicated
We can assign weights to these objectives to resources and expertise available to inves-
reflect their relative importance to us. For tigate and prosecute cybercrime cases.
example, we may decide that reducing the • Using technology to combat cybercrime:
incidence of cybercrime is more important Law enforcement agencies can use tech-
than improving the security of e-governance nology, such as data analytics and artificial
systems [15]. intelligence, to help them to investigate
We can then use the weighted sum method and prosecute cybercrime cases.
to calculate a score for each policy [16]. The • Working with the private sector: Law
policy with the highest score is the most enforcement agencies can work with the
preferred. private sector to share information and
For example, suppose we assign the fol- resources to combat cybercrime.
lowing weights to the objectives: • Educating the public about cybercrime:
This will help to reduce the number of
• Reduce the incidence of cybercrime: 0.7 victims of cybercrime.
• Improve the security of e-governance sys-
tems: 0.3 References
• Then, we can calculate the score for each
[1] T. Ahmad, R. Aljafari, and V. Venkatesh, The
policy as follows:
Government of Jamaica’s electronic procure-
• Policy 1: (0.7)(0.8) + (0.3)(0.5) = 0.66 ment system: experiences and lessons learned,
• Policy 2: (0.7)(0.6) + (0.3)(0.9) = 0.72 INTR, vol. 29, no. 6, pp. 1571–1588, Dec.
2019, doi: 10.1108/INTR-02-2019-0044.
In this case, policy 2 is the most preferred [2] [M. Åkesson, P. Skålén, and B. Edvards-
policy, as it has a higher score than policy 1. son, E-government and service orientation:
This is just a simple example of how gaps between theory and practice, Interna-
MCDM techniques can be used to man- tional Journal of Public Sector Management,
age the trade-offs between the objectives vol. 21, no. 1, pp. 74–92, Jan. 2008, doi:
of e-governance and law gaps. The specific 10.1108/09513550810846122.
296 Innovations in Computational Intelligence: Trends and Technologies
[3] G. Edwards, Cybercrime Investigators Huynh, B. Le, K. Honda, M. Inuiguchi, and
Handbook, 1st ed. Wiley, 2019. doi: Y. Kohda, Eds., in Lecture Notes in Com-
10.1002/9781119596318. puter Science, vol. 14375. Cham: Springer
[4] S. Gordon and R. Ford, On the definition Nature Switzerland, 2023, pp. 85–96. doi:
and classification of cybercrime, J Comput 10.1007/978-3-031-46775-2_8.
Virol, vol. 2, no. 1, pp. 13–20, Aug. 2006, [10] A. Zagaria and A. Zennaro, A close look at
doi: 10.1007/s11416-006-0015-z. sociality in DSM criteria, Soc Psychiatry Psy-
[5] S. Y. Lee, J. M. Díaz-Puente, and S. Mar- chiatr Epidemiol, Nov. 2023, doi: 10.1007/
tin, The Contribution of Open Govern- s00127-023-02568-z.
ment to Prosperity of Society, International [11] H. Taherdoost and M. Madanchian, Multi-
Journal of Public Administration, vol. Criteria Decision Making (MCDM) Meth-
42, no. 2, pp. 144–157, Jan. 2019, doi: ods and Concepts, Encyclopedia, vol. 3,
10.1080/01900692.2017.1405446. no. 1, pp. 77–87, Jan. 2023, doi: 10.3390/
[6] S. Kethineni, Cybercrime in India: Laws, encyclopedia3010006.
Regulations, and Enforcement Mechanisms, [12] B. Liu and X. Chen, Uncertain Multiobjec-
in The Palgrave Handbook of International tive Programming and Uncertain Goal Pro-
Cybercrime and Cyberdeviance, T. J. Holt gramming, J. Uncertain. Anal. Appl., vol.
and A. M. Bossler, Eds., Cham: Springer 3, no. 1, p. 10, Dec. 2015, doi: 10.1186/
International Publishing, 2020, pp. 305– s40467-015-0036-6.
326. doi: 10.1007/978-3-319-78440-3_7. [13] O. Bozorg-Haddad, H. Loáiciga, and B. Zol-
[7] Government of India, Gazette of India: Indian ghadr-Asli, A Handbook on Multi-Attribute
IT Act 2000 (Registered NO. DL-33004/2000). Decision-Making Methods, 1st ed. Wiley,
Jun. 2000. Accessed: Nov. 10, 2023. [Online]. 2021. doi: 10.1002/9781119563501.
Available: https://eprocure.gov.in/cppp/ [14] O. Cooper, The Magic Of The Analytic Hier-
rulesandprocs/kbadqkdlcswfjdelrquehwux- archy Process (AHP), IJAHP, vol. 9, no. 3,
cfmijmuixngudufgbuubgubfugbububjxcgfvs- Dec. 2017, doi: 10.13033/ijahp.v9i3.519.
bdihbgfGhdfgFHytyhRtMjk4NzY= [15] M. A. K. Harahap, K. Kraugusteeliana, S. A.
[8] D. Kumar, AIIMS Delhi hit by fresh cyber- Pramono, O. Z. Jian, and A. M. Almaududi
attack for second time in a year, mint. Ausat, “The role of information technol-
Accessed: Nov. 10, 2023. [Online]. Avail- ogy in improving urban governance,” jmp,
able: https://www.livemint.com/news/ vol. 12, no. 1, pp. 371–379, May 2023, doi:
india/aiims-delhi-hit-by-fresh-cyberattacks- 10.33395/jmp.v12i1.12405.
details-here-11686061994629.html [16] H. Guo and N. J. Dorans, Using Weighted
[9] T. Entani, Different Types of Decision Cri- Sum Scores to Close the Gap Between DIF
teria in a Decision Problem, in Integrated Practice and Theory, J Educational Meas-
Uncertainty in Knowledge Modelling urement, vol. 57, no. 4, pp. 484–510, Dec.
and Decision Making, vol. 14375, V.-N. 2020, doi: 10.1111/jedm.12258.
39 A exhaustive review on advancements
and challenges in low power wireless
sensor networks
Manuj Darbari1,a, Naresh Chandrab2, Diwakar Yagyasenc1,b,
Anurag Tiwarid1,c, and Sandeep Kumar Mishrae1,d
1
CSE Department, BBDITM, Lucknow, India
2
Shambhunath Institute of Engineering and Technology, Prayagraj, India
Abstract: This paper presents a comprehensive examination of the challenges and corresponding solutions
associated with wireless sensor networks (WSNs), particularly focusing on energy constraints, communication
issues, network deployment, and security concerns. Through an in-depth analysis, it identifies the criticality of
energy efficiency due to the limited power resources of sensor nodes, especially in inaccessible environments.
Solutions such as energy harvesting from renewable sources and power-efficient communication protocols are
explored. The paper further delves into the intricacies of communication protocols, emphasizing the need for
adaptability and scalability to address the diverse requirements of WSN applications. It also discusses optimal
network deployment strategies that mitigate coverage issues and maximize energy utilization. Security is high-
lighted as a paramount issue, given the sensitive nature of the data handled by WSNs, and the paper outlines
various lightweight cryptographic measures and privacy-preserving techniques to safeguard against threats. A
case study on forest monitoring is presented to illustrate the real-world application of these solutions, showcas-
ing the effectiveness of energy-efficient protocols and adaptive strategies in extending the network’s lifespan
while ensuring robust performance. The paper concludes with insights into the future advancements expected in
WSN technologies, which promise to enhance their sustainability, autonomy, and security, further integrating
WSNs into the fabric of emerging Internet of Things.
Keywords: Energy waste, IoT, low power, WSN
1. Introduction communication protocols, node deployment,
and network security.
Wireless sensor networks (WSNs) have been
widely researched in the past decade, pri-
marily due to their potential applications 1.1. Energy efficiency
in fields such as environmental monitoring, Energy efficiency remains a crucial concern in
agriculture, health care, defense, transporta- LP-WSNs due to the typically limited battery
tion, and industrial automation. Low power life of sensor nodes and the often infeasibil-
WSNs (LP-WSNs) have garnered particular ity of battery replacement. Yang et al. (2020)
attention due to their promise of providing explore energy harvesting technologies,
highly efficient energy solutions. This lit- including solar, wind, and vibration, which
erature review outlines the advancements could effectively self-power WSNs. Mean-
and current challenges in LP-WSNs, explor- while, Harb et al. (2021) delve into energy-
ing key aspects such as power management, efficient design strategies for nodes, focusing
[email protected]; [email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-39
298 Innovations in Computational Intelligence: Trends and Technologies
on low-power hardware components and WSNs have rapidly evolved in recent years,
energy-efficient coding techniques. becoming a critical component of many tech-
nological fields, including industrial automa-
1.2. Communication protocols tion, environmental monitoring, health care,
and smart cities [1]. However, numerous
Communication protocols significantly influ- challenges continue to limit their widespread
ence the power consumption of LP-WSNs. In adoption and efficiency. This paper aims to
this vein, Chen et al. (2021) evaluate several identify key challenges and proposes poten-
routing protocols specifically designed for tial strategies for their resolution.
LP-WSNs, such as LEACH, PEGASIS, and
HEED. The authors underscore the trade-off
between energy efficiency and communica- 2. Challenges and Solutions in
tion reliability, highlighting the need for fur- WSNs
ther research in this area.
Akmehmet et al. (2022) survey various 2.1. Energy constraints
MAC protocols for LP-WSNs, including
S-MAC, T-MAC, and B-MAC. The authors a WSNs typically consist of a large number of
rgue that these protocols have shown prom- small, low-cost sensor nodes powered by bat-
ise in reducing idle listening, collision, and teries with limited energy resources [2]. In
overhearing - the primary sources of energy many scenarios, especially remote and hos-
waste in wireless communication. tile environments, it is infeasible to replace or
recharge these batteries, making energy effi-
ciency a crucial concern [3].
1.3. Node deployment Solution: Energy harvesting technologies,
Another research focus is on node deploy- such as solar, thermal, and vibration energy,
ment strategies to maximize network cover- can supplement battery power in certain envi-
age while minimizing power consumption. ronments [4]. Additionally, power-efficient
Liu et al. (2020) propose a novel algorithm communication protocols and smart sleeping
that utilizes a two-level hierarchical cluster- schedules can optimize energy use [5].
ing method for node deployment, signifi-
cantly improving the lifetime of LP-WSNs. 2.2. Communication issues
However, they point out that the proposed
model might not be practical in environments Communication protocols in WSNs must
with unpredictable changes. consider factors like energy efficiency, reli-
ability, scalability, and latency [6]. However,
designing a one-size-fits-all protocol is chal-
1.4. Network security lenging due to the varied nature of WSN
Despite advances in power management, com- applications.
munication protocols, and node deployment, Solution: A multitude of energy-efficient
network security continues to be a daunting routing and MAC protocols have been pro-
challenge for LP-WSNs. Zhang et al. (2022) posed for WSNs, such as LEACH, PEGASIS,
elucidates the unique security vulnerabilities S-MAC, and T-MAC [7]. Adaptive protocols,
associated with these networks, due to factors which can adjust parameters based on net-
like limited computational power and battery work conditions, also show promise [8]. The
life of the nodes. The authors propose using integration of energy-harvesting mechanisms
lightweight cryptographic solutions, but they such as solar, thermal, and vibration-based
also highlight the inherent trade-off between energy can bolster the battery life in certain
security and energy efficiency. contexts [4]. Moreover, the development and
A exhaustive review on advancements and challenges in low power wireless sensor networks 299
implementation of communication protocols amongst nodes. Additionally, data aggre-
that conserve power, alongside the utilization gation techniques that respect privacy can
of intelligent sleeping schedules, can further safeguard against the infringement of user
streamline energy consumption [5]. privacy.
2.3. Network deployment and 3. Case Study
coverage A WSN was deployed in a large forest area
Sensor node deployment impacts network to monitor environmental conditions, such
coverage, connectivity, and energy efficiency as temperature, humidity, and the presence of
[9]. A poorly planned deployment could lead harmful gases, to provide early warnings for
to coverage holes or energy wastage. forest fires.
Solution: Deployment strategies based on Challenge: The primary challenge was
optimization algorithms or swarm intelli- the network’s energy consumption, as sen-
gence can help achieve optimal sensor place- sor nodes were powered by batteries, and
ment [10]. Randomized deployment followed the forest’s remote location made battery
by a self-organization process could also be replacement difficult. Additionally, maintain-
effective in certain scenarios [11]. Solution: ing reliable communication for real-time data
Strategies that employ optimization algo- transfer without depleting the nodes’ energy
rithms or leverage the principles of swarm reserves was crucial.
intelligence can aid in securing an ideal place- An energy-efficient routing protocol,
ment of sensor nodes. Alternatively, a rand- LEACH (Low-Energy Adaptive Clustering
omized deployment approach followed by a Hierarchy), was implemented, allowing sen-
phase of self-organization might prove to be sor nodes to communicate their data to a
advantageous in certain circumstances. local cluster head that aggregated the infor-
mation before sending it to the base station.
This reduced the number of long-range trans-
2.4. Security and privacy missions, conserving energy.
WSNs often handle sensitive data and operate Furthermore, solar energy harvesting was
in unsecured environments, making them vul- integrated into the nodes to supplement the
nerable to various security threats, including battery power, ensuring a sustainable energy
eavesdropping, node tampering, and denial-
of-service attacks [12].
Solution: Lightweight cryptographic tech-
niques can protect the confidentiality and
integrity of data [13]. Intrusion detection sys-
tems can identify and mitigate threats, while
trust-based frameworks can promote coop- AQ7:
Please
erative behavior among nodes [14]. Privacy- provide
preserving data aggregation techniques can in-text
also ensure user privacy [15]. Solution: The citation
for Figure
application of lightweight encryption meth- 30.1
odologies can safeguard data confidentiality
and integrity. The employment of intrusion AQ2:
Please
detection systems can aid in identifying and provide
counteracting threats, while trust-centric Figure 39.1. WSN deployed in forest area. source for
frameworks can foster a spirit of cooperation Source: the figure.
AUTHOR'S SELF GENERATED
300 Innovations in Computational Intelligence: Trends and Technologies
source for the sensors. Adaptive protocols forests showcases the potential for broader
were also introduced, which adjusted the deployment in other critical areas such as
nodes’ activity levels based on environmental agriculture, healthcare, and urban planning.
data traffic, further conserving energy. The integration of lightweight cryptographic
The deployment of LEACH and the inte- techniques and privacy-preserving meth-
gration of energy harvesting significantly ods also ensures that data integrity and user
extended the WSN’s operational lifespan. privacy are maintained, even in unsecured
The network provided continuous moni- environments. Looking forward, the ongo-
toring with reduced energy consumption, ing evolution of WSN technologies will likely
resulting in fewer maintenance trips and focus on optimizing energy efficiency and net-
lower operational costs. Real-time data work self-sustainability while fortifying secu-
transmission allowed for timely responses rity protocols, ensuring that WSNs remain
to potential forest fires, demonstrating the indispensable in our increasingly intercon-
effectiveness of adaptive, energy-efficient nected world.
solutions in WSNs. In summary, this paper has identified and
addressed the key challenges facing WSNs,
focusing on energy constraints, commu-
4. Conclusion nication issues, network deployment, and
Despite the remarkable advancements in WSN security concerns. The implementation of
technology, numerous challenges persist. energy-efficient protocols like LEACH, cou-
However, the emergence of energy harvesting pled with energy harvesting techniques and
technologies, the design of energy-efficient adaptive communication strategies, has
communication protocols, the adoption of proven to be an effective solution to extend
smart node deployment strategies, and the the lifespan of WSNs while maintaining reli-
implementation of security measures provide able data transmission. The case study pre-
potential pathways to overcome these hur- sented on forest monitoring exemplifies the
dles. As WSNs continue to evolve, ongoing practical benefits of these solutions, reveal-
research and development are necessary to ing a substantial improvement in operational
address these challenges and unlock their full efficiency and a reduction in maintenance
potential. In conclusion, the exploration of costs.
challenges and solutions within WSNs (WSNs) As we advance, it becomes clear that
underscores the critical importance of energy the future of WSNs hinges on the continu-
efficiency, reliable communication, strategic ous innovation in energy harvesting, pro-
deployment, and robust security measures. tocol optimization, and security measures.
Through the implementation of adaptive The integration of advanced machine learn-
protocols, energy-harvesting technologies, ing algorithms for predictive analytics, the
and intelligent deployment strategies, WSNs development of more sophisticated intrusion
can overcome operational limitations. The detection systems, and the enhancement of
case study on forest monitoring demonstrates privacy-preserving data aggregation tech-
that with these solutions, WSNs can achieve niques are poised to elevate the capabilities of
sustainable, long-term monitoring with mini- WSNs. The potential for WSNs to contribute
mal environmental impact. As technology to smarter, safer, and more efficient ecosys-
advances, further refinement of these solu- tems is immense, especially as the Internet of
tions is expected, promising enhanced per- Things (IoT) continues to expand.
formance and broader application of WSNs The implications of such advancements
in various fields, the successful application extend far beyond the immediate improve-
of WSNs in monitoring environments like ments in network performance. They signal
A exhaustive review on advancements and challenges in low power wireless sensor networks 301
a shift towards more autonomous, resil- [7] W. B. Heinzelman, A. P. Chandrakasan, and
ient, and intelligent sensor networks capa- H. Balakrishnan, An application-specific
ble of supporting the data-driven demands protocol architecture for wireless microsen-
of modern society. By harnessing the power sor networks, IEEE Transactions on Wireless
of WSNs, we can expect to see significant Communications, vol. 1, no. 4, pp. 660–670,
2002.
impacts on environmental monitoring, dis-
[8] S. Singh and C. S. Raghavendra, PAMAS—
aster management, smart cities, and beyond.
Power aware multi-access protocol with
In essence, the evolution of WSNs not only signalling for ad hoc networks, ACM SIG-
reflects the progression of networking tech- COMM Computer Communication Review,
nologies but also the broader commitment vol. 28, no. 3, pp. 5–26, 1998.
to creating a more connected and responsive [9] G. Wang, G. Cao, and T. F. L. Porta, Move-
world. ment-assisted sensor deployment, in Proc.
INFOCOM, vol. 4, pp. 2469–2479, 2004.
[10] A. Howard, M. J. Mataric, and G. S.
References Sukhatme, Mobile sensor network deploy-
[1] K. Sohraby, D. Minoli, and T. Znati, Wireless ment using potential fields: A distributed,
sensor networks: technology, protocols, and scalable solution to the area coverage prob-
applications, John Wiley and Sons, 2007. lem, in Proc. DARS, vol. 6, pp. 299–308,
[2] I. F. Akyildiz, W. Su, Y. Sankarasubrama- 2002.
niam, and E. Cayirci, Wireless sensor net- [11] Y. Zou and K. Chakrabarty, Sensor deploy-
works: a survey, Computer Networks, vol. ment and target localization in distributed
38, no. 4, pp. 393–422, 2002. sensor networks, ACM Transactions on
[3] R. C. Shah and J. M. Rabaey, Energy aware Embedded Computing Systems (TECS), vol.
routing for low energy ad hoc sensor net- 3, no. 1, pp. 61–91, 2004.
works, in Proc. IEEE Wireless Communica- [12] A. D. Wood and J. A. Stankovic, Denial of
tions and Networking Conference (WCNC), service in sensor networks, Computer, vol.
vol. 1, pp. 350–355, 2002. 35, no. 10, pp. 54-62, 2002.
[4] S. Roundy, D. Steingart, L. Frechette, P. [13] C. Karlof and D. Wagner, Secure routing in
Wright, and J. Rabaey, Power sources for wireless sensor networks: Attacks and coun-
wireless sensor networks, in Proc. Euro- termeasures, Ad Hoc Networks, vol. 1, no.
pean Workshop Wireless Sensor Networks 2-3, pp. 293–315, 2003.
(EWSN), pp. 1–17, 2004. [14] R. Roman, J. Zhou, and J. Lopez, Applying
[5] W. Ye, J. Heidemann, and D. Estrin, An intrusion detection systems to wireless sen-
energy-efficient MAC protocol for wireless sor networks, in Proc. IEEE Consumer Com-
sensor networks, in Proc. INFOCOM, vol. munications and Networking Conference
3, pp. 1567–1576, 2002. (CCNC), pp. 640–644, 2006.
[6] K. Akkaya and M. Younis, A survey on rout- [15] P. Kamat, Y. Zhang, W. Trappe, and C.
ing protocols for wireless sensor networks, Ozturk, Enhancing source-location privacy
Ad Hoc Networks, vol. 3, no. 3, pp. 325– in sensor network routing, in Proc. ICDCS,
349, 2005. pp. 599–608, 2005.
40 Automated personality evaluation of
children based on emotional speech
Alapati Naresha, Basireddy Yasaswib, Golla Sri Lakshmic,
Konakandla Meghanad, and Allamudi Yaminie
Computer Science and Engineering, Vignan’s Nirula Institute of Technology and Science for Women,
Guntur
Abstract: Understanding children’s personalities is crucial for their emotional and social development. This
study proposes an innovative approach for automatic personality assessment in children based on their emo-
tional speech patterns. An essential component of human conduct, emotions are crucial to the manifestation of
one’s individuality. Our research leverages advanced machine learning techniques to extract valuable insights
from children’s vocal cues, aiming to provide a non-invasive and efficient means of personality assessment.
From the voice data, linguistic and acoustic characteristics are retrieved, including pitch, intonation, and speech
rate. CNN and MFCC algorithms are used in the creation of the developed model. The developed system is
capable of accurately predicting personality traits such as extroversion, agreeableness, conscientiousness, open-
ness, and neuroticism in children based on their emotional speech. Our results demonstrate promising accuracy
levels when compared to traditional personality assessment methods, showcasing the potential of this technol-
ogy in educational and clinical settings. This automated approach can aid educators, psychologists, and parents
in gaining deeper insights into children’s personalities, enabling more tailored support and interventions to
foster healthy development. The goal of automatic personality perception is to predict, using nonverbal cues,
the speaker’s behavior as viewed by the audience.
Keywords: MFCC, CNN, personality assessment, neuroticism, conscientiousness
1. Introduction psychologists. The characteristics of a child’s
personality play a crucial part in shaping
Understanding and assessing a child’s per-
their behavior, social relationships, and future
sonality is a difficult but crucial aspect of
outcomes. Traditional personality evalua-
their development and well-being. Emo-
tional speech analysis can help us understand tion methods can involve subjective observa-
a child’s emotional state, which is closely tions, questionnaires, and interviews that can
linked to their personality traits, emotional be laborious, biased, and human error. The
health, and psychological development. This proposed method offers a factual and data-
study suggests a novel approach for auto- driven solution to this issue. The extracted
matically evaluating children’s personalities MFCC characteristics are fed into a CNN-
by and analyzing emotional speech using based algorithm, a deep-learning architec-
convolutional neural networks (CNN) algo- ture that is well-known for its effectiveness
rithms and mel-frequency cepstral coefficients in image and pattern recognition. In this case,
(MFCC). Children’s personality evaluations the CNN is adjusted to process MFCC data,
are highly valued by parents, educators, and which allows it to learn and identify patterns
alapatinaresh13@ gmail.com; [email protected]; csrilakshmigolla2002@ gmail.com;
a
[email protected]; [email protected]
d
DOI: 10.1201/9781003606635-40
Automated personality evaluation of children based on emotional speech 303
linked to different personality traits and emo- filter for every data layer. One filter is used
tional states in children’s speech. to convolute a layer of data that is subjected
This study is significant because it has the to depth-wise separable convolution; each
potential to drastically change the way that output is then assigned a 191 filter. Lastly, it
children’s personalities are assessed. Through permits one more convolution operation to
the analysis of emotional speech, this process be performed on each filter and each layer of
is automated, allowing the system to produce output data. By doing this, the convolution
assessments that are more impartial, depend- operation’s calculation load may be lowered
able, and accurate. This work also contributes from 1/8 to 1/9 of the total CNN.
to the broader field of voice emotion analysis A CNN-based VGG16 architecture was AQ1:
and advances the use of deep learning algo- presented by Karen et al. [3] (Delplanque Please
rithms for speech-related tasks. The informa- 2017). The significance of depth in CNN net- provide
the
tion gathered from this study may contribute works is emphasized by VGG16. Each con- expan-
to the advancement of speech technology and volutional layer filter is 3 9 3 compressed as sion for
VGG16.
open the door to more accurate and useful opposed to CNN, which often uses a bigger
assessments of children’s personality traits filter. The last layer’s completely linked layer
and emotions, which would benefit their still contains 90% of the network’s param-
growth and well-being. eters, despite the convolution process being
more effective. As a result, The Inception V3
architecture was proposed by Christian et al.
2. Literature survey Strain and Joseph (2003) employing Global
Speech emotion recognition (SER), according Average Pooling technology. The layer’s prob-
to Teddy Surya Gunawan and Taiba Majid lem is capturing too many parameters and
Wani [1] has become a crucial part. Sophis- characteristics can be resolved by averaging
ticated voice processing systems and human- the pooling layer’s data. Limitations were
computer interaction. By recognizing and noted in the study, however they were not
categorizing the salient characteristics of a made clear. addressing constraints
speech signal that has previously undergone Personality is assessed using big five inven-
processing, an emotion recognition system tory questionnaires. The analysis is performed
(SER) can usually determine the speaker’s using the Berlin database of emotional speech
spectrum of emotions. The study provides (EMO-DB). An ANN is used for classification
readers with a cutting-edge grasp of the popu- after acoustic-prosodic information. Liu et al.
lar research issue by meticulously identifying (2013) [4] present an artificial neural network
and synthesizing recent pertinent literature (ANN)-based method for APP. extracted. The
linked to the numerous design components/ system’s forecast accuracy rate was 70%. The
methodologies of the SER system. The study evaluations of personality traits are founded
also points out that more reliable assessment on the opinions of human assessors. Subjec-
techniques are required to gauge the systems’ tive human assessments of personality may
effectiveness because SER systems’ precision not always coincide with objective personal-
is still not ideal. ity evaluations. Although 640 voice snippets
The Xception architecture was proposed from 322 speakers could be a suitable data-
by Francois et al. [2] (2010). Depth-wise set for study, it is somewhat small for train-
separable convolution technique reduces the ing sophisticated machine learning models,
quantity of data and significantly increases which may restrict the generalizability of the
the speed of convolution operations (Daunic findings.
2015). The standard CNN convolution oper- Zhang et al. (2018) [5] evaluated emotional
ation technique is used to convolve with every speech signals using a non-linear dynamic
304 Innovations in Computational Intelligence: Trends and Technologies
model based on chaotic components in the
discourse sound cycle. The chaotic aspects of
the emotional speech signals were subsequent
to extricating the non-straight components
of the close-to-home discourse signal and the
regularly utilized acoustic highlights (rhyth-
mic features and the Mayer inversion coef-
ficient (MFCC)).
Using the IITH-H and EMO-DB data-
bases, Gangamohan et al. (2017) discovered
ID paces of 76 and 69%, separately, by assess-
ing the Kullback-Leibler (KL) distance of the
excitation source signals [6].
3. Proposed methodology
A paradigm shift has occurred in the fields
of psychology and personality assessment in
an era characterized by rapid technological AQ2:
advancements. The fundamental idea of this Please
study is that emotional speech can provide provide
figure
deep insights into a person’s personality. Emo- citations
tions are markers of personality traits and in text.
dispositions in addition to being the outcome
of internal psychological states. A carefully
AQ3:
designed road map guides the model’s devel- Please
opment and construction, starting with data provide
source
collection and preprocessing to guarantee the for all the
accuracy and dependability of the input. The figures.
Figure 40.1. Workflow of the model.
next step involves using MFCC as the basis
for identifying key characteristics in the emo- Source:
AUTHOR'S SELF GENERATED
tionally charged speech samples. These char-
acteristics serve as the building blocks for the The CNN algorithm processes the features
CNNs and other key elements of our selected and produces the final result.
machine-learning architecture. We provide
an overview of the suggested model in this
summary, outlining its main phases from data 3.1. Data preprocessing
preprocessing to model training and testing. The first and most important task in this mod-
The ultimate objective is to clarify the pos- el’s initial phase is to gather a suitable data-
sible advantages and importance of auto- set, which is represented by the letter D. Pairs
mated personality assessment from emotional of audio waveforms (xi) and the labels that go
speech, thereby advancing our knowledge of with them (yi) are included in this dataset; xi
children’s growth and well-being in the con- represents the child’s speech, and yi represents
temporary era. the labels for personality traits. These labels
MFCC algorithm is used such that it takes might be categorical values that stand for
the speech as an input extracts the features various personality characteristics. The next
and gives the output to the CNN algorithm. crucial step is to divide the dataset into three
Automated personality evaluation of children based on emotional speech 305
subsets: a training set (Dtrain), a validation set Convolutional layers: These layers use
(Dval), and a test set (Dtest). This division facili- learnable weights (Wk) and biases (bk) to
tates the evaluation of the model’s generaliza- apply convolution operations; k denotes the
tion and performance. layer index. The MFCC data’s spatial patterns
can be captured with the aid of convolution.
3.2. MFCC feature extraction Activation Function: To provide nonlinearity
to the model and help it understand intricate
Our goal in this step is to identify pertinent correlations in the data, an activation function
features from the D audio samples. We utilize is usually utilized. rectified linear units are a
MFCC, a popular method for speech analysis, common example of activation functions.
to accomplish this. We use the MFCC extrac- Layers for pooling: These layers sample
tion procedure to produce a set of MFCC the feature maps while maintaining crucial
coefficients, referred to as MFCC(xi), for information and lowering the dimensionality
each audio sample xi in the dataset. Impor- of the data.
tant details about the child’s speech are cap- Completely linked Layers: These strata
tured by these coefficients, which are denoted link the convolutional layers’ output to a
as [c1, c2, ..., cP], where P is the number of single or multiple fully linked layers. They
MFCC coefficients. The feature vectors Xi = include more intricate feature interactions
[MFCC(xi)] that result from these computed and come with weights (wf) and biases (bf).
MFCC coefficients are then saved and will be Output Layer: The model can predict the
used as input data in the following stages. probability distribution of personality traits
for each input by using the SoftMax activa-
3.3. Architecture of CNN tion function in the final fully connected layer
for classification.
The structure of the CNN is the core of the
model. At this point, the CNN architecture
that will process the MFCC feature vectors is 3.4. Training
designed and specified. This network receives We define a critical learning component, dur-
Xi, a series of MFCC coefficients, as input. ing the training phase, the cross-entropy loss
The network is made up of various essential function is used. For all samples and personal-
parts: ity traits, this loss function computes the dif-
Figure 40.2 represents the convolution ference between the actual labels (yi,j) and the
and pooling process in CNN architecture. anticipated probabilities (pi,j). It is represented
by the notation L(wk, bk, wf, bf). Using an opti-
mizer, such as the well-known Adam optimizer,
the model’s parameters, including the convolu-
tional and fully connected weights and biases
(wk, bk, wf, bf), are iteratively changed. In order
to effectively train the model to make accurate
predictions and determine children’s personal-
ity traits from their emotional speech, the goal
is to minimize the loss of function.
3.5. Evaluation
Using the specially designed test dataset, Dtest,
Figure 40.2. CNN architecture. the trained model is put through a rigorous
Source: testing process to determine how well it can
AUTHOR'S SELF GENERATED
306 Innovations in Computational Intelligence: Trends and Technologies
automatically identify children’s personal- 15. Update the model’s wk, bk, wf, and bf
ity traits from emotional speech during the parameters to minimize the loss func-
evaluation phase. Computed performance tion. using an optimizer (like Adam, for
metrics include recall, accuracy, precision, example).
and F1-score, which give a thorough assess- 16. Examine the trained model’s performance
ment of how well the model classifies person- with the Dtest test dataset:
ality traits. This crucial phase guarantees the 17. Determine the F1-score, recall, accuracy,
validity of the model and its possible use in precision, and other classification metrics.
real-world scenarios involving the emotional
growth and development of children.
5. Result
The model gets the opportunity to advance
4. Algorithm possibly huge portrayals of character-related
discourse highlights by taking care of CNN
1. Gather the D = {(xi, yi)} dataset in which yi
with MFCCs as input characteristics: CNNs
is the label that corresponds to the audio
can automatically extract significant features
waveform xi.
from input data, including MFCCs, which
2. Divide the dataset into sets for test-
eliminates the need for the engineer to per-
ing (Dtest), validation (Dval), and training
form time-consuming feature engineering.
(D_train).
Thus our model predicts the emotions of
3. For every D audio sample xi:
an individual based on their speech data as
4. The formula for using MFCC to extract
shown in Figure 40.3. This model performs
MFCC coefficients is MFCC(xi) = [c1,
with high efficiency that helps in identifying
c2, ..., cP], where P is the total number of
the individual state of being.
MFCC coefficients.
This allows the model to gain discrimi-
5. Use Xi = [MFCC(xi)] to store the resulting
native elements straightforwardly from the
MFCC coefficients as features.
information, possibly distinguishing unob-
6. Describe the architecture of CNN:
trusive examples or nuances in the discourse
7. Input: an MFCC coefficient sequence
flags that may not be clear to human eyewit-
denoted by Xi.
nesses. It can also save time and effort when
8. Convolutional Layers: Apply convolu-
handcrafting features.
tional layers, where k is the layer index,
with weights wk and biases bk.
9. Activation Function: Make use of a ReLU-
style activation function.
10. Apply pooling layers to feature maps
before down sampling them.
11. Fully Connected Layers: Attach one or more
fully connected layers with weights wf and
biases bf to the convolutional layer’s result.
12. Output Layer: Softmax activation for
classification in the final fully connected
layer.
13. Describe the classification’s cross-entropy
loss function:
14. Where pi,j is the expected probability of
the j-th class for the i-th sample, L(wk, bk, Figure 40.3. Output.
wf, bf) = –Σi Σj yi,j * log(pi,j). Source:
AUTHOR'S SELF GENERATED
Automated personality evaluation of children based on emotional speech 307
6. Conclusion signals using the chaotic characteristics in
the speech sound process and extracted the
In conclusion, there is great potential for non-linear features of the emotional speech
understanding and promoting children’s emo- signal and the commonly used acoustic
tional well-being with the development of an features (rhythmic features and the Mayer
automated system for assessing children’s inversion coefficient (MFCC)). 2018.
personalities from emotional speech using [5] Smith FW and Rossit S. Identifying and
algorithms, MFCC, and CNN. This method detecting facial expressions of emotion in
reduces subjectivity in personality assess- peripheral vision. PLOS ONE. 2018;13(5).
ment and offers educators, psychologists, doi:10.1371/journal.pone.0197160.
and parents an objective and methodical way [6] Chen M, Zhou P, and Fortino G. Emotion
communication system. IEEE Access. 2017.
to assess children’s emotional states based
doi:10.1109/ACCESS.2016.2641480.
on speech. It also offers insightful informa- [7] Lee J and Tashev I. High-level feature repre-
tion. It may make it possible to identify emo- sentation using recurrent neural network for
tional problems early on, which may allow speech emotion recognition. In: Proceedings
for prompt support and intervention and, of the 16th Annual Conference International
eventually, better results for mental health. Speech Communication Association. 2015.
Additionally, by customizing interventions [8] Koolagudi SG and Rao KS. Emotion recog-
and instructional plans to match each child’s nition from speech: A review. Int J Speech
specific needs based on their personality traits Technol. 2012.
and emotional states, this technology can [9] Wakita H. Residual energy of linear predic-
support individualized support. tion to vowel and speaker recognition. IEEE
It can also be tested using various algo- Trans Audio Speech Signal Process. 1976.
rithms and video datasets. One important [10] Prasanna SRM, Gupta CS, and Yegnana-
rayana B. Extraction of speaker-specific
development for the system in the future
excitation information from linear predic-
would be its ability to support multiple lan- tion residual of speech. Speech Commun.
guages and cultures. 2006.
[11] Fastl H and Zwicker E. Psychoacoustics:
Facts and Models. 3rd ed. Springer; 2005.
References [12] Duda RO, Hart PE, and Stork DG. Pattern
[1] Wani TM and Gunawan TS. A comprehen- Classification. 2nd ed. John Wiley and Sons;
sive review of speech emotion recognition. 2000.
IEEE Access. 2021. [13] Scherer KR, and Scherer U. Speech behavior
[2] Francois, Karen et al. Development and and personality. Speech Evaluation in Psy-
evaluation of an emotional lexicon system chiatry. 1981.
for young children. Springer. June 22, 2019. [14] Kessous L, Castellano G, and Caridakis G.
[3] Liu et al. An artificial neural network (ANN) Multimodal emotion recognition in speech-
based approach is proposed for APP. Person- based interaction using facial expression,
ality is assessed using the Big Five Inventory body gesture, and acoustic analysis. 2010.
(BFI) questionnaires. 2013. [15] Pu X, Wu G, and Yuan C. Exploring overall
[4] Zhang et al. Applied a non-linear dynamic opinions for document-level sentiment clas-
model to analyze the emotional speech sification with structural SVM. 2019.
41 Redefining human activity recognition
with LSTM and GRU networks
Vasumathi Devi M.a, Sai Sreeja S.b, Anusha P.c, Mounika Bindu B.d,
and Sirisha M.e
Computer Science and Engineering, Vignan’s Nirula Institute of Technology and Science for Women,
Guntur, India
Abstract: Recognition of human activity is an important field in artificial intelligence, machine learning, and
deep learning (DL), applied in healthcare, sports analysis, security, and human-computer interaction. HAR uti-
AQ1:
lizes sensor data like accelerometers, gyroscopes, cameras, and audio sensors to identify human actions. Deep Please
learning, including convolutional neural networks and recurrent neural networks (RNNs), autonomously learns provide
and classifies activities. Data preprocessing and feature engineering are crucial. Validation ensures real-world the
accuracy. DL’s continuous evolution promises HAR advancements across domains. In our proposed model we compre-
hensive
used the LSTM and GRU Algorithm for Human Activity Recognition. The growing dataset volume necessitates
expansion
machine learning. Our LSTM and GRU, RNN-based model outperform existing methods on HAR datasets. for HAR.
This research paper presents a comprehensive investigation into HAR using Deep learning techniques. We
explore data collection methods, feature engineering approaches, and various DL models to accurately clas- AQ2:
sify human activities. Our study includes an in-depth analysis of performance metrics, model evaluation, and Please
provide
real-time applications. Through extensive experiments on diverse datasets, we demonstrate the effectiveness the
of DL-based HAR systems in achieving high accuracy and potential for real-world deployment. This research compre-
contributes to the ongoing efforts in developing robust and versatile HAR solutions. hensive
expansion
Keywords: RNN, LSTM, GRU, human activity recognition, neural network for LSTM
and GRU.
NOTE ADDED
1. Introduction at automatically learning intricate represen-
tations from raw data, making them suit- AQ3:
Deep learning-based HAR is a cutting-edge Please
able for HAR tasks. HAR, situated within provide
technology that gained significant attention
the broader field of artificial intelligence, the
in recent years [10]. It employs deep neural compre-
machine learning, and deep learning (DL),ADDED
NOTE hensive
networks to automatically detect, classify,
and comprehend human actions and behav- focuses on detecting and classifying human expansion
for HAR.
iors using sensor data or video inputs. HAR actions and movements using various sensors
finds applications in various domains, includ- like accelerometers, gyroscopes, and cameras AQ4:
ing healthcare, fitness tracking, security sur- [2]. DL techniques have revolutionized HAR, Please
veillance, and smart environments [1]. The offering exceptional accuracy and versatility provide
the
primary goal of HAR is to create systems in recognizing a wide range of human activi-
NOTE ADDED
expansion
capable of comprehending human move- ties. In this era of pervasive technology, HAR for LSTM
and GRU.
ments and actions. Deep learning techniques using DL has found applications in diverse
have proven highly effective in handling com- domains, from healthcare and fitness track-
plex, high-dimensional data for this purpose ing to security and augmented reality [3]. The
[4]. Deep neural networks, in particular, excel following introductory exploration delves
[email protected]; [email protected]; [email protected];
a
[email protected]; [email protected]
d
DOI: 10.1201/9781003606635-41
Redefining human activity recognition with LSTM and GRU networks 309
into the captivating domain of (HAR) using interpretability very little, lacks architec-
LSTM and GRU. We will explore the core tural details, and has a more extensive set of
principles of DL-based HAR, investigate data assessment metrics. [4] Inzamam Mashood
sources and preprocessing techniques, and Nasir et al. (IEEE 2021). Classifying human
discuss prevalent neural network architec- behaviors can benefit from the methods pre-
tures utilized for robust activity classification sented in this article, which include creating
[6]. Furthermore, we will examine real- 3D Cartesian-plane features, mining feature
world applications and prospects for future descriptors using geodesic distance, extract-
advancements, emphasizing the pivotal role ing interest-based segments, and more. Its
DL plays in comprehending and enhancing shortcomings, which include complexity,
human actions within our increasingly inter- dataset size, and lack of open-source code,
connected world [9]. may be resolved by future research. [5] Peng
To capture temporal correlations in time Zhihao et al. (IEEE 2020). The potential of
series problems, a memory cell structure neural networks with deep connections to
is integrated into both the GRU and the independently identify features in a smart
LSTM models of loop networks. A growing home is combined with the recently suc-
number of researchers are adopting these cessful recurrent and convolution neural
techniques for HAR due to DL’s outstand- networks in a novel multitask layer neural
ing capabilities [4]. Activity prediction has network (LSTM) model. Its drawback is its
applications in a wide range of domains, restricted dataset information. [6] Garima
including video surveillance, gait analysis, Aggarwal et al. (IEEE 2021. They have
behavior analysis, gesture identification, talked about two main models: 2-D CNN
health and exercise monitoring, and inter- and LSTM. According to how accurate they
active games [5]. are, they are compared. [7] Syed Hamza
Azeem et al. (IEEE 2021). Through ensemble
learning across various convolutional neural
2. Literature survey network (CNN) models, scientists devised a
[1] Marjan Gholamrezaii et al. (IEEE 2019). novel method for human behavior recogni-
this paper’s authors presented a novel frame- tion in this research. [8] Seref Recep Keskin
work that uses solely convolutional lay- et al. (IEEE 2020). This study examined pos-
ers in a 2D convolutional neural network. tural stress and spinal mobility using convo-
Its limitations should be considered while lutional and deep neural networks on two
analyzing heterogeneous datasets, process- sensors placed to the spine in addition to the
ing noisy input, and considering hardware pelvis of a healthy person. Potential over-
problems. [2] Md Maruf Hossain Shuvo et fitting, a constrained context, and a scant
al. (IEEE 2020). The research offers a via- dataset explanation are among its disadvan-
ble hybrid method for identifying human tages.[9] Magda B. Fayek et al. (Springer
activity that blends deep learning and con- 2022) Res Inc Conv LSTM, is a novel layer
ventional machine learning approaches. Its introduced in this work. Comparing their
drawbacks, however, include issues with suggested architecture to the Conv LSTM
privacy for practical implementation, and baseline architecture, accuracy increased by
noise handling. [3] Deepika Singh et al. 7%. Some of its disadvantages are hyperpa-
(Springer 2017). Three real-world data sets rameter tweaking, overfitting, interpretabil-
related to smart homes were used to train ity problems, and computational expense.
an LSTM recurrent neural network for this [10] Zhipeng Jin et al. (Springer 2021).
purpose. Some of its drawbacks are that The present study presents a unique frame-
it covers hyperparameter adjustment and work for video activity prediction: WLSTM
310 Innovations in Computational Intelligence: Trends and Technologies
combined with SME. Based on boundary 3.2. Algorithm
priors, we present an initial motion segmen-
Step 1: Take input values and declare hg(t–1)
tation method. Among its shortcomings are
and current input Xgt. Declare σ () = sigmoid
its interpretability, and data dependencies.
and tanh () =hyperbolic tangent are activa-
tion Functions.
3. Proposed model Step 2: Initialize the net parameters LSTM
net, calculate the forget gate as follows.
Our model leverages the advantages of LSTM
in combination with GRU algorithms to rec- (1)
ognize human behavior. The recurrent neu-
Now calculate the output of forget gate as
ral networks (RNN) family includes both of
follows:
these advanced NN topologies.
LSTM: By overcoming some of the draw- (2)
backs of conventional RNNs, LSTM aims
to capture the long-term dependencies in Step 3: calculate the input gate, it has two
sequential input. It does this by maintaining parts as shown below
a memory cell that is capable of storing and (3)
retrieving information over long sequences.
The widespread acceptance of LSTM is (4)
attributed in part to the mechanisms involved
Now calculate the output of the input gate
in each LSTM cell [6].
as follows:
GRU: GRU strikes a balance between
effectiveness and simplicity when compared (5)
to conventional RNNs and LSTMs, which
makes it an excellent choice for applications (6)
requiring sequence modeling [7]. The GRU, Step 4: Calculate the output gate, as below
operates with distinct gates and follows a dif-
ferent workflow than an RNN. To solve the (7)
problem with conventional RNN, GRU uti-
lizes the update gate and reset gate operation Step 5: now calculate:
approaches [9].
(8)
3.1. Dataset // GRU algorithm
Step 6: take the past information as input,
The dataset consists of 15 features of Human Calculate the updated As Follows:
Activities. The dataset contains about
12k+ labeled images including the valida- (9)
tion images. Test - contains 5400 images of
human activities. For these images, you are Step 7: Calculate the reset gate denoted by
required to make predictions as the respec- as follows:
tive class names – “calling,” “clapping,”
(10)
“cycling,” “dancing,” “drinking,” “eating,”
“fighting,” “hugging,” “laughing,” “listen- Step 8: Calculate the memory content
ing_to_music,” “running,” “sitting,” “sleep- which will use the reset gate to store the Rel-
ing,” “texting,” “using_laptop” and the train evant information from the past:
consists of 15 different folders each repre-
sents the activity [10]. (11)
Redefining human activity recognition with LSTM and GRU networks 311
Step 9: Now finally calculate htt-Vector splitting. On the training dataset, data pre-
which holds the information, Update-gate is processing operations like normalization,
Required. grey scaling, and label encoding are carried
out after segmentation.
(12)
Step 10: END 4.1. Normalization
To provide consistent input data, adjust each
4. Design image’s pixel values to a common range.
Designing HAR system using LSTM and
GRU networks with LSTM involves several 4.2. Grayscale conversion
key components and considerations [7]. Data Reducing the images to grayscale can help
preprocessing is the initial step, encompass- simplify the model and reduce the computa-
ing dataset collection, cleaning, and for- tional load.
matting. Data segmentation is crucial, as it
involves breaking the continuous data into 4.3. Label encoding
fixed-length segments, and creating sequences
that represent different activities. Addition- Converting categorical labels into numeri-
ally, feature extraction plays a vital role in cal representations for model training. There
extracting relevant features from these data have been three LSTM and GRU layers uti-
sequences, such as time-domain or frequency- lized, which include a dense layer then the
domain features. flattening layer.
It starts with the HAR dataset, which is
divided into training and testing datasets. 5. Implementation
Segmentation is carried out after dataset
The training dataset is sent into the LSTM
layer first, and subsequently the GRU layer.
In this way, three sets of LSTM and GRU lay-
ers are taken. The flattened layer receives the
output from these layers. A flattening layer is
used to restructure the output from the recur-
rent layers into a format that a dense layer
can handle. Furthermore, the flattened layer’s
output is fed into the dense layer as an input.
A dense layer is a completely linked layer that
make use of the output taken from flattened
AQ5:
Please layer to inform its final predictions [5]. The
provide outcome of the dense layer is the total result.
source
for all the
Design performance can be evaluated by sev-
figures. eral assessment metrics. Based on the supplied
image dataset, these metrics are used to assess
AQ6:
Please
how successfully the model detects human
provide actions. For classification tasks such as HAR,
citations common evaluation measures include accu-
for Fig-
ures and Figure 41.1. LSTM and GRU model racy, precision, and recall. These parameters
Tables in architecture. shed light on how well the algorithm can cat-
text.
Source: egorize different types of activity [8].
AUTHOR'S SELF GENERATED
312 Innovations in Computational Intelligence: Trends and Technologies
Hyperparameter tuning is another criti- Table 41.1. Comparison with other methods
cal aspect of the design. Adjustments must be Algorithms Accuracy (%)
made to regularization procedures like L2 or
dropout regularization, as well as batch size CNN 85%
and learning rate. The sequence length, repre- LSTM 90%
senting the length of input sequences, should ResNet 75%
be carefully chosen as it impacts the network’s LSTM and GRU (ours) 97%
ability to capture temporal dependencies Source:
effectively. Training is a fundamental phase,
with the choice of a suitable loss function and Figure 41.3 is about the accuracy for sev-
optimizer. Techniques like early stopping can eral models in which the proposed model
be implemented to prevent overfitting and achieved the highest among them which is
ensure better generalization. about 97%. With a precision score of 96%,
we were able to determine the percentage of
6. Results actual positive predictions among all positive
predictions generated by the model.
In our study on HAR using LSTM and GRU We were able to ascertain the proportion of
with the implementation of early stopping real accurate forecasts across all positive predic-
for reducing variance, we obtained highly tions produced by the model, achieving a preci-
promising results. With an astounding 97% sion score of 96%as shown in Figure 41.4.
accuracy rate, our algorithm demonstrated its
capacity to precisely categorize and identify
human activities. Real-world applications of
the proposed approach, such as healthcare,
sports analysis, or smart environments.
We evaluated the efficacy of our LSTM and
GRU models in our investigation of human
activity recognition against a number of earlier
models, including CNN, standalone LSTM,
and ResNet, based on existing literature and
benchmark results. While we did not directly
implement these alternative architectures in
our study, our model achieved a significantly Figure 41.3. Accuracy.
higher accuracy rate, by more than 10 to 20 Source: AUTHOR'S SELF GENERATED
percentage points in terms of accuracy.
Figure 41.2. Applications of 52. Figure 41.4. Precision.
Source: Source:
AUTHOR'S SELF GENERATED
AUTHOR'S SELF GENERATED
Redefining human activity recognition with LSTM and GRU networks 313
The recall score of 98 for our model indi- accuracy of the data used for training improv-
cated the percentage of factually positive ing as the total amount of epochs increases.
cases that were true positive prognostica-
tions. A high recall score indicates that the
model minimizes false negatives by success- 7. Conclusion
fully capturing the majority of positive cases. In summary, our research has explored
97% was the measured F1-score, which human activity recognition using two robust
balances recall and precision. This score takes structures for RNNs, LSTM and GRU, while
into account false negatives as well as false effectively reducing variability through the
positives to provide a fair evaluation of the strategic use of early stopping. Through
model’s overall performance. extensive experimentation and analysis,
These results collectively underscore the we’ve uncovered valuable insights and dem-
effectiveness of LSTM and GRU networks onstrated the potential of this approach to
with early stopping as a strategy for human enhance the accuracy and dependability of
activity recognition. The evaluation metrics activity recognition systems. By including
indicate that the approach exhibits excel- both LSTM and GRU networks in our study.
lent precision, recall, accuracy, and robust- Both architectures showed impressive perfor-
ness. The training accuracy of the epochs is mance, highlighting their ability to capture
displayed in the figure. Five epochs are used temporal dependencies and intricate patterns
to calculate the training accuracy, with the within activity data. This adaptability sug-
gests that LSTM and GRU networks can be
tailored to specific application domains and
data characteristics, providing flexibility to
researchers and practitioners.
References AQ7:
As per
[1] Gholamrezaii M and Almodarresi SMT. style the
Human activity recognition using 2D con- reference
volutional neural networks. In: 2019 27th citations
should be
Iranian Conference on Electrical Engineer- num-
Figure 41.5. Recall. ing (ICEE). IEEE; 2019:1682–1686. bered,
[2] Shuvo MMH, Ahmed N, Nouduri K, and please
Source: check and
AUTHOR'S SELF GENERATED Palaniappan K. A hybrid approach for provide
human activity recognition with support for
vector machine and 1D convolutional neural clarity.
network. In: 2020 IEEE Applied Imagery
Pattern Recognition Workshop (AIPR).
IEEE; 2020:1-5.Singh D, Merdivan E, Psy-
choula I, et al. Human activity recognition
using recurrent neural networks. In: Machine
Learning and Knowledge Extraction: First
IFIP TC 5, WG 8.4, 8.9, 12.9 International
Cross-Domain Conference, CD-MAKE
2017, Reggio, Italy, August 29–September 1,
2017, Proceedings 1. Springer International
Publishing; 2017:267–274.
Figure 41.6. F1-Score. [3] Nasir IM, Raza M, Shah JH, Khan MA, and
Source: Rehman A. Human action recognition using
AUTHOR'S SELF GENERATED
314 Innovations in Computational Intelligence: Trends and Technologies
machine learning in uncontrolled environ- networks. In: 2021 55th Annual Conference
ment. In: 2021 1st International Conference on Information Sciences and Systems (CISS).
on Artificial Intelligence and Data Analytics IEEE; 2021:1–5.
(CAIDA). IEEE; 2021:182–187. [7] Keskin ŞR, Gençdoğmuş A, Yıldırım B,
[4] Wang C and Peng Z. Deep learning model Doğan G, and Öztürk Y. DNN and CNN
for human activity recognition and predic- approach for human activity recognition. In:
tion in smart homes. In: 2020 International 2020 7th International Conference on Elec-
Conference on Intelligent Transportation, trical and Electronics Engineering (ICEEE).
Big Data and Smart City (ICITBS). IEEE; IEEE; 2020:254–258.
2020:741–744. [8] Khater S, Hadhoud M, and Fayek MB. A
[5] Khattar L, Kapoor C, and Aggarwal G. novel human activity recognition architec-
Analysis of human activity recognition using ture: using residual inception ConvLSTM
deep learning. In: 2021 11th International layer. Journal of Engineering and Applied
Conference on Cloud Computing, Data Sci- Science. 2022;69(1):45.
ence and Engineering (Confluence). IEEE; [9] Weng Z, Li W, and Jin Z. Human activity
2021:100–104. prediction using saliency-aware motion
[6] Zehra N, Azeem SH, and Farhan M. Human enhancement and weighted LSTM network.
activity recognition through ensemble EURASIP Journal on Image and Video Pro-
learning of multiple convolutional neural cessing. 2021;2021:1–23.
42 Cybersaferoutes: Protecting IoT
networks from hybrid attack with
security measures
Ganga Bhavani Pa, Lakshmi Tirapathamma Sb, Reshma SKc,
Swetha Bd, and Himaja Ne
Department of CSE, Vignan’s Nirula Institute of Technology and Science for Women Peda
Palakaluru, Guntur
Abstract: The world wide web has revolutionized interconnectivity, interweaving smart components into our
daily lives, homes, and critical infrastructure. However, this digital transformation has brought forth an emerg-
ing and complex threat landscape. This research explores the synergy of two potent cyber threats, namely the
Wormhole Attack and the Blackhole Attack, within the realm of IoT. Wormhole attacks involve the surrepti-
tious tunneling of data between distant points in an IoT network, potentially allowing malicious actors to
eavesdrop on, alter, or disrupt communications. On the other hand, Blackhole attacks involve the creation of
malicious nodes that attract data packets but intentionally drop or manipulate them. Both attacks exploit vul-
nerabilities in IoT communication protocols and can severely affect data integrity, privacy, and network perfor-
mance. IoT devices become increasingly integrated into our daily lives, and the security of these interconnected
systems takes on unprecedented importance. The convergence of Wormhole and Blackhole attacks serves as a
stark reminder that the IoT ecosystem is not immune to the tactics of sophisticated cyber adversaries.
Keywords: Wormhole Attack, Blackhole Attack, Cyber Threats, Digital Transformation, IoT Security, Data
Integrity, Privacy, Network Performance, Cyber Adversaries, Security Measures
1. Introduction Wormhole attacks are clandestine and
stealthy maneuvers within an IoT network that
The emergence of the World Wide Web surreptitiously tunnel data packets between
(WWW) of Things, or IoT, has brought about distant points, potentially enabling malicious
an unparalleled period of connectivity and actors to eavesdrop on sensitive communica-
innovation, revolutionizing how we commu- tions or disrupt data flows. Conversely, Black-
nicate with the digital and physical worlds. hole attacks involve the creation of rogue nodes
IoT has seamlessly integrated smart devices that attract data packets but subsequently dis-
into our homes, cities, industries, and critical card or manipulate them, leading to data loss
infrastructure [1], offering enhanced conveni- and network disruption. When these two for-
ence, efficiency, and automation. However, midable threats synergize, the consequences for
as the IoT ecosystem continues to expand, it IoT systems can be dire, encompassing compro-
has also exposed vulnerabilities that are ripe mised data integrity, disrupted network opera-
for exploitation, particularly through hybrid tions, and threats to user privacy.
attacks that ingeniously combine two potent The widespread use of IoT devices, such
threats: the Wormhole Attack and the Black- as industrial sensors, wearable fitness track-
hole Attack. ers, and smart thermostats, and autonomous
[email protected]; [email protected]; [email protected];
a
[email protected]; [email protected]
d
DOI: 10.1201/9781003606635-42
316 Innovations in Computational Intelligence: Trends and Technologies
vehicles, has heralded a new era of connec- Subash A; Arvin Danny CS; Vijayalakshmi
tivity and automation. Yet, with this unprec- M [5] In this paper, we discuss the increas-
edented growth comes a stark reality the ing threat of DDoS attacks in the internet
vulnerability of IoT ecosystems to sophisti- era. These attacks are on the rise, and attack-
cated cyber threats. When these two potent ers are well-equipped with resources. This
attack vectors are combined, the resulting scheme effectively addresses the packet frag-
hybrid attacks exploit the strengths of each, mentation issue while only slightly increasing
creating a formidable threat that can infil- storage needs.
trate IoT systems in a nuanced and multi- P. Sanju [6] In this paper, an innovative
pronged manner. In light of the evolving approach is presented that combines hybrid
threat landscape and the looming specter of metaheuristics with deep learning techniques
hybrid attacks, understanding these multi- to bolster intrusion detection capabilities in
faceted threats becomes imperative for all IoT systems.
IoT [2] stakeholders. By capitalizing on the Regonda Nagaraju, Jupeth Toriano Pen-
strengths and stealth of Wormhole and Black- tang, Shokhjakhon Abdufattokhov, Ricardo
hole attacks, these hybrid threats introduce Fernando CosioBorda, N. Mageswari, G.
a multifaceted and nuanced assault on IoT Uganya [7] this research paper combines the
systems power of hybridized strategies for optimiza-
As we stand on the brink of an era where tion with profound knowledge to establish
IoT devices become even more deeply inte- a robust attack prevention system for IoT
grated into our lives, safeguarding these inter- environments.
connected ecosystems is not merely an option. Amiya Kumar Sahu, Suraj Sharma, M.
Tanveer, Rohit Raja [8] This study introduces
an innovative security framework and an
2. Literature Survey attack detection mechanism, aiming to bridge
Nai The author Hossein [1] For this paper, existing gaps in the field.
researchers investigate a security problem Danish Javeed, Tianhan Gao, Muham-
related to attacker identification for a par- mad Taimoor Khan and Ijaz Ahmad [9] the
ticular kind a cyber-physical systems. These attained accuracy of our proposed scheme
networks are made from continual actors stands at an impressive 99.87%, coupled
that communicate with distinct computa- with an outstanding recall rate of 99.96%.
tional elements. Amit Sagu, Nasib Singh Gill, Preeti Gulia,
Yu-Cheng Chen, Santiago Grijalva, Dus- Jyotir Moy Chatterjee and Ishaani Priya-
tin Campbell, Vincent Mooney, and Tim darshini [10] In this current text, we introduce
Gieseking [2] In this study, we present a novel a mixed sensing model tailored for identifying
method for examining grid assaults: the use attacks within an IoT environment, employ-
of the Hybrid Assault Hypothesis (HAM). ing a comprehensive three-stage approach.
John Hale, Mauricio Papa, Michael Haney,
and Peter J. Hawrylak [3] In this paper, we
investigate a cyber-physical attack intended to 3. Proposed Model
get too hot an electrical appliance on an intel- As the Internet of Things (IoT) permeates
ligent power substation you are considering. every aspect of our lives, its promise of con-
Cheolhyeon Kwon; Inseok Hwang [4] for nectivity and convenience is accompanied by
the present instance, research on newspaper the growing specter of hybrid attacks, a men-
explore the design of strong commanders for acing combination of the Wormhole Attack
CPS that can with stand numerous kinds of and the Black Hole Attack. Our model begins
cyberattacks. with the initialization and setup of the IoT
Cybersaferoutes: Protecting IoT networks from hybrid attack with security measures 317
network, establishing various nodes, sen- Crucially, our model operates in real-time,
sors, and communication pathways. [3] Key continuously monitoring network traffic and
parameters such as distance (dn), the thresh- behavior. This proactive approach allows
old value (R), and the hop interval threshold for the early detection of hybrid attacks that
(Th) are configured to adapt to the unique blend the insidious Wormhole and Black
characteristics of the network. Hole tactics. By calculating the hop interval
Attack detection and identification are threshold (Th) based on parameters such as
achieved through the utilization of the Worm- distance (dn) and a threshold value (R), the
hole and Black Hole Attack Detection and model adapts to the specific context of the
Mitigation Algorithm. The presence of a Worm- IoT network, ensuring precise and dynamic
hole attack is inferred when the Hop Emission detection capabilities.
(Hem) equals 1, prompting a meticulous anal- The versatility of our model lies in its
ysis of the Hop Interval (Hi) in comparison adaptability. It continuously learns and
to the dynamically calculated threshold (Th). evolves through the incorporation of machine
In the event of a Wormhole attack, the model learning and artificial intelligence techniques.
responds decisively by marking it as “TRUE”
(W), thereby signifying the necessity to sever
the communication path between Ns and Nr. 3.1. Implementation of hybrid attack
Subsequently, secure communication is re- Step 0: Access information about the suspect
established via an alternate route, allowing Ns transmitter component Ns, along with its
and Nr to reconfigure their settings for a forti- opposite ending Nr..
fied communication channel. Step 1: Assign a cutoff for hops intervals (Th)
The model extends its vigilance to iden- based on parameters dn and R.
tify Black Hole attacks, promptly marking Step 2: Verify if Hem (Hop Emission) is equal
Black Hole (B) as “TRUE” when detected. to 1 (True).
In the presence of a Black Hole threat, simi- Step 3: If Hem equals 1, check if Hi (Hop
lar actions are taken, ensuring the network’s Interval) is greater than Th.
integrity is preserved. - If true, proceed to Step 4.
When neither Wormhole nor Black Hole - If false, state W as “YES.”
attacks are detected, the model deems the Step 4: State W as “NO.”
communication path safe, enabling uninter- Step 5: Verify if W is equal to “YES.”
rupted data exchange within the IoT network. Step 6: If W is “YES,” delink the Route con-
To ensure adaptability and continued necting Ns and Nr.
security, the model emphasizes continuous - Proceed to Step 7 to re-establish commu-
monitoring and learning. The integration of nication by a secure route.
machine learning and artificial intelligence Step 7: Allow Ns and Nr to reconfigure for
techniques allows the system to evolve, secure communication.
enhancing its ability to detect and mitigate Step 8: If W is not “YES,” consider the path
new hybrid attack patterns. safe and proceed to Step 9.
Our proposed model serves as a compre- Step 9: Verify if B (Black Hole) is equal to 1
hensive framework for fortifying IoT secu- (YES).
rity against the evolving threat of hybrid Step 10: If B equals 1, delink the route con-
attacks. By combining proactive monitoring, necting Ns and Nr.
[4] adaptive thresholds, and swift response - Proceed to Step 11 to rebuild contact
mechanisms, we aim to secure IoT networks through by a safe path.
in an era where these interconnected ecosys- Step 11: Permit Ns and Nr to reorganise to
tems are indispensable. for secure communication.
318 Innovations in Computational Intelligence: Trends and Technologies
Step 12: If B is not 1, consider the path safe In Hop Interval (Hi) vs. Threshold (Th) if
and allow communication to continue. Hem is 1, the algorithm checks if Hi (Hop
Interval) is greater than Th. Marking W as
Hop Interval Threshold (Th) here this algo- “TRUE” or “FALSE” here Depending on the
rithm sets a hop interval threshold (Th) based result of the previous comparison, the algo-
on parameters dn and R. This threshold likely rithm marks W as “TRUE” or “FALSE.” W
determines the maximum acceptable number could represent the path’s security status. If
of hops for a communication path to be con- W is “TRUE,” it may indicate that the path is
sidered secure. Adjusting Th appropriately is not secure due to excessive hops. In Delink-
crucial to balance security and performance. ing and Re-Establishing Communication
Hop Emission (Hem), means that the algo- If W is “TRUE,” the algorithm delinks the
rithm checks whether Hem is equal to 1. Hem path between Ns and Nr, essentially cutting
might be a flag or indicator of some kind, off communication. Then, it proceeds to re-
possibly indicating the initiation of commu- establish communication by finding a secure
nication between Ns and Nr. If Hem is 1, it route. This step ensures that communication
triggers further evaluation of the path. continues through a more secure path. The
algorithm allows Ns and Nr to reconfigure for
secure communication after delinking. This
likely involves finding a different, [5] more
secure path or making necessary adjustments
to communication parameters. Black Hole
(B) The algorithm checks if B (Black Hole) is
equal to 1. In Delinking and Re-Establishing
Communication (Again) if B is equal to 1,
similar to the case of W being “TRUE,” [6]
the algorithm delinks the path and re-estab-
lishes communication through a secure route.
In the event that W is marked as “TRUE,”
the algorithm proceeds to take essential
actions to safeguard the network. It initiates
the delinking of the communication path [7]
between the sender node Ns and its counter-
part Nr. This decisive step effectively severs
communication to prevent potential security
breaches. Subsequently, the algorithm endeav-
ors to re-establish secure communication by
AQ1:
Please
identifying an alternative, more secure route.
provide The model begins with the configuration
in-text
NOTE ADDED and initialization of the IoT network, encom-
citation
for figures passing a multitude of [8] interconnected
1-6. devices. Key parameters, such as distance
(dn), the threshold value (R), and the dynami-
AQ2: cally adaptable hop interval threshold (Th),
Please
provide are initialized to create a responsive security
source environment.
for all the Figure 42.1. Architecture diagram.
figures
Additionally, the algorithm incorporates a
Source: check for Black Hole attacks, examining the
AUTHOR'S SELF GENERATED
Cybersaferoutes: Protecting IoT networks from hybrid attack with security measures 319
variable B. A value of 1 for B indicates [9] the A hybrid attack in IoT that combines the
potential presence of a Black Hole attack, a black hole attack, where a malicious node
severe network issue, or a compromised net- drops all received packets, and the worm-
work node. hole attack, which creates a shortcut for data
transmission.
A hybrid attack in IoT, combining the
4. Results black hole and wormhole attacks, can lead to
In a hybrid attack involving the black hole data loss due to the malicious node dropping
and wormhole attacks in IoT, [10] the result- packets.
ing reduction in throughput can be quite
substantial.
The hybrid attack resulted in a reduced
packet delivery ratio (PDR), indicating a
higher rate of packet loss or failure in mes-
sage transmission.
The hybrid attack, which combines black
hole and wormhole attacks in IoT networks,
results in an elevated energy consumption.
Figure 42.4. Energy delay
Source: AUTHOR'S SELF GENERATED
Figure 42.2. Throughput.
Source: AUTHOR'S SELF GENERATED
Figure 42.5. Control overhead.
Source: AUTHOR'S SELF GENERATED
Figure 42.3. Packet delivery ratio. Figure 42.6. Loss.
Source: Source:
AUTHOR'S SELF GENERATED
AUTHOR'S SELF GENERATED
320 Innovations in Computational Intelligence: Trends and Technologies
5. Conclusion in electricity grid.” 2019 IEEE Texas Power
and Energy Conference (TPEC). IEEE, 2019.
This algorithm is a robust security mecha- [3] Hawrylak, Peter J., et al. “Using hybrid
nism that focuses on securing communication attack graphs to model cyber-physical
paths in a dynamic and adaptive manner. It attacks in the smart grid.” 2012 5th Inter-
addresses security threats, optimizes routing, national symposium on resilient control sys-
and maintains network resilience, ultimately tems. IEEE, 2012.
contributing to a more secure and reliable [4] Kwon, Cheolhyeon, and Inseok Hwang.
network environment. Path Optimization, “Hybrid robust controller design: Cyber
the algorithm emphasizes the importance of attack attenuation for cyber-physical sys-
optimizing the communication path. It prior- tems.” 52nd IEEE conference on decision
and control. IEEE, 2013.
itizes shorter, more secure routes over longer
[5] Subash, A., Arvin Danny, and M. Vijayalak-
and potentially less secure ones by delinking shmi. “An Enhanced Hybrid Scheme for IP
paths that exceed the hop threshold. The most Traceback.” 2023 4th International Confer-
important usage of this algorithm is Dynamic ence on Innovative Trends in Information
Security this algorithm provides a dynamic Technology (ICITIIT). IEEE, 2023.
approach to security. It doesn’t rely solely [6] Sanju, P. “Enhancing Intrusion Detection in
on fixed configurations but adapts to chang- IoT Systems: A Hybrid Metaheuristics-Deep
ing conditions. When issues like excessive Learning Approach with Ensemble of Recur-
hops or Black Hole conditions are detected, rent Neural Networks.” Journal of Engineer-
it takes immediate action to address them. ing Research (2023): 100122.
By actively assessing the hop interval, hop [7] Nagaraju, Regonda, et al. “Attack preven-
emission, and the presence of black holes, it tion in IoT through hybrid optimization
acts as a shield against hybrid wormhole and mechanism and deep learning framework.”
Measurement: Sensors 24 (2022): 100431.
black hole attacks.
[8] Tanveer, M. “Internet of Things attack detection
using hybrid Deep Learning Model.” (2021).
[9] Javeed, Danish, et al. “A hybrid deep learn-
References ing-driven SDN enabled mechanism for
[1] Mosenia, Arsalan, and Niraj K. Jha. “A com- secure communication in Internet of Things
prehensive study of the security of internet- (IoT).” Sensors 21.14 (2021): 4884.
of-things.” IEEE Transactions on emerging [10] Sagu, Amit, et al. “A hybrid deep learning model
topics in computing 5.4 (2016): 586–602. with self-improved optimization algorithm for
[2] Chen, Yu-Cheng, et al. “A hybrid attack detection of security attacks in the IoT environ-
model for cyber-physical security assessment ment.” Future Internet 14.10 (2022).
43 Harmful content classification in social
media using gated recurrent units and
bidirectional encoder representations
from transformer
Sujatha V.a, Tejaswi Y.b, Pravalika V.c, Pavani P.d, and Ch Sravanie
Vignan’s Nirula Institute of Technology and Science for Women, Guntur, India
Abstract: Harmful content on the internet poses a significant challenge to maintaining a safe and inclusive
online environment. To remove this issue, we propose a novel approach for harmful content classification,
combining the power of recurrent neural networks and pre-trained transformers. Specifically, we utilize gated
recurrent units (GRUs) for text preprocessing and bidirectional encoder representations from transformers
(BERT) for the final classification task. In the preprocessing stage, GRUs are employed to capture sequential
dependencies in textual data, allowing for the effective extraction of context-based features from the input text.
This preprocessing step is crucial for understanding the nuanced structure and context of potentially harmful
content, such as hate speech, cyberbullying, or offensive language. Subsequently, the pre-processed data is fed
into a fine-tuned BERT model for classification. BERT is a state-of-the-art transformer model that excels in
understanding the semantics of text, making it particularly well-suited for the nuanced and context-dependent
nature of harmful content detection. By leveraging BERT’s pre-trained contextual embeddings, we can effi-
ciently classify text into different categories of harm, including hate speech, harassment, or misinformation.
Our proposed approach is capable of handling multi-class classification, making it versatile for a huge range of
harmful content identification tasks. Furthermore, the combination of GRU-based preprocessing and BERT-
based classification offers a powerful and adaptable solution for detecting harmful content, contributing to the
ongoing efforts to create safer online spaces, and promoting responsible content moderation.
Keywords: Natural language processing, toxicity detection, gated recurrent unit, bidirectional encoder repre-
sentations from transformers, convolutional neural network
1. Introduction threatens online review services’ long-term
survival. According to studies, there are many
In the modern world, the Internet serves as bogus reviews on websites. Consequently, it’s
both a platform for opinion expression and a critical to automatically detect false reviews
tool for knowledge acquisition. Fraud involv- on online forums and give users more accu-
ing customer reviews of online products or rate information. Consequently, it’s critical to
services that are essential to a company or automatically detect false reviews on online
brand represents one of the most significant forums and give users more accurate infor-
problems that exists currently. The prolifera- mation The goal of producing a fake review is
tion of fake ratings is turning into a major to influence consumer’s willingness to shop in
problem. since it deceives customers into mak- addition to raising the product’s star rating.
ing poor purchasing decisions and critically Reviews that evoke favorable feelings will
a
[email protected]; [email protected]; [email protected]; dgppavani58@gmail.
com; [email protected]
DOI: 10.1201/9781003606635-43
322 Innovations in Computational Intelligence: Trends and Technologies
increase readers’ preference for the products, 2. Literature survey
negative reviews will decrease favorability.
The majority of the time, businesses engage Kanwal Yousaf et all [5] It refers to bidirec- AQ3:
specialists to write false remarks. tional LSTM, convolutional neural network Please
provide
These experts are compensated for posting (CNN) IEEE [2022], the usage of pre-trained the
both positive and negative reviews of goods transformer language models and word NOTEexpan-
ADDED
or brands, which greatly aids in promoting embeddings do not directly encode offensive sion for
LSTM.
a certain company. Traditional methods for remarks due to their construction from clean
fake review detection [4] have relied on rule- corpora.
based heuristics, lexical analysis, and statisti- Wonhel Lee1 et all [6] In order to choose
NOTE
AQ1: ADDED
cal approaches. While these techniques have the best extraction and normalization method
Reference
offered some degree of success, they often and create a safeguarding system that shields
citations
are not in struggle to capture the subtle nuances and young people using the internet from haz-
sequence.
context-dependent cues that distinguish gen- ardous contents, several different algorithms
Please
check and uine from fake reviews. This paper focuses were tried, including TF-IDF, TF-ICF, IG,
review for
on the utilization of two state-of-the-art NLP CHI, log TF, and SVM.
clarity.
architectures [1]: Gated recurrent unit (GRU) Hina Tufail [7] et al: SKL-Based Fake
and bidirectional encoder representations Review Detection, SVM, KNN, and logistic
AQ2:
Please from the transformer (BERT), in the context regression to thoroughly examine Positive-
provide
NOTE ADDED of fake review recognition These architec- unlabeled learning strategies, a semi-supervised
the
expansion tures have demonstrated remarkable capabil- learning methodology.
for NLP.
ities in capturing complex linguistic patterns Ying Qian et al [8] The performance of
and contextual information, enabling them emotion classification models is enhanced by
to discern even the most sophisticated decep- the combination of local and global semantics
tive reviews. The use of GRU in preprocess- in CNN, self-aware neural network mecha-
ing aims to capture sequential dependencies nism, and neural network PCC.
within textual data, allowing our model to Rohit Kumar Kaliyar et al [9] Conventional
better understand the context and nuances of Neural Network (CNN), LSTM examines
potentially harmful content.[3] GRU’s abil- the issue of fake news from the perspective of
ity to retain long-term dependencies, coupled various echo chambers found in social media
with its computational efficiency, positions it data, which are regarded as a collection of
as a potent tool for text data preparation. individuals with similar viewpoints on any
Our goal in this research is to empiri- given social issue.
cally demonstrate the superior performance
of GRU and BERT classifiers [2] compared 3. Related work
to traditional methods. By leveraging their
advanced capabilities in sequential data pro- 3.1. Proposed approaches for toxic
cessing and contextual understanding.
Through this research, we hope to improve
content detection
the effectiveness of hostile review detection The deliberate writing of fake content with
while simultaneously making a larger dis- the goal of distorting information makes it
course on leveraging advanced NLP tech- difficult to detect. Typically, regular, and obvi-
niques for addressing challenges in online ous word forms found in the corpus are used
commerce and consumer trust. The results to create pre-trained embeddings. Informa-
of this analysis have important consequences tion is lost because these representations do
for e-commerce platforms, consumers, and not include all the words used in the hurtful
the broader online retail ecosystem. comments.
Harmful content classification in social media using gated recurrent units and bidirectional encoder 323
3.1.1. Data preparation The data have been split into three subsets:
a training set for model training, a validation
The current method primarily concentrates
set for hyperparameter tuning and model
on different types of supervised machine-
selection, and a test set for evaluating the
learning classification algorithms. Techniques
model’s performance.
such as bidirectional LSTM, CNN, SVM,
KNN, and logistic regression have been used
to achieve accuracy above 90 Unfortunately, 3.3. Algorithms
these methods weren’t enough to train large This research work uses deep learning for pre-
datasets quickly. So, in our study, we used processing the GRU algorithm and machine
GRU and BERT algorithm combined model learning for classification which uses BERT
building an easy-to-use site for fake comment algorithms. Many natural processing tasks
identification and training on huge datasets can be applied to these models, and they per-
form well on them. This research technique
3.1.2. Data sources makes use of the models indicated below.
Social Media Platforms: The information
may have been gathered from well-known 3.4. GRU
social media sites with a lot of harmful con- The reset gate () determines how much of the
tent, such as Facebook, Twitter, and others. past information should be discarded, while the
Web scraping: relevant textual data from update gate () controls the balance between the
public postings and comments on social previous hidden state and the Seeker hidden
media may have been gathered using auto- state (). The Seeker hidden state is a combina-
mated online scraping techniques. tion of the previous hidden state and the input,
and the final hidden state is determined by a lin-
3.2. Data preprocessing ear interpolation between the previous hidden
state and the Seeker hidden state. These equa-
Text Normalization: To standardize the text tions allow the GRU algorithm to capture and
data, common text normalization procedures update information over sequential input, mak-
like lowercasing, deleting special characters, ing it suitable for various tasks such as natural
and handling slang or informal language language processing and time-series analysis.
may have been used. To prepare the textual
material for future processing, it can undergo
tokenization into words or sub-word units.
3.5. GRU formula
N-gram extraction is the process of taking 1. Reset gate:
n-grams, or consecutive sequences of n ele- rst = s igmoid(Wtirsxt + Wthrsht–1 + bshrs +
ments, out of the text data in order to extract bsirs)
contextual information for the purpose of 2. Update gate:
identifying potentially dangerous content. udt = s igmoid(Wtiudxt + Wthudht–1 + bshud +
Stopword Elimination: In order to enhance bsiud)
the classification model’s performance, com- 3. Seeker hidden state:
mon words with little meaning, including “and,” hst = t anh(Wtihsxt + rt(Wthhs * ht–1 + bshsn) +
“the,” and “is,” may have been eliminated. bsihs
Lemmatization or stemming: Words may 4. Hidden state:
have been reduced to their base or root form ht = (1 – udt) * hst + udtht–1
using morphological analysis approaches, • rst is the input at time t.
which can aid in lowering the dimensionality • ht–1 is the hidden state from the previ-
of the feature space. ous time step.
324 Innovations in Computational Intelligence: Trends and Technologies
• rst is the reset gate, which determines
how much of the past information to
forget.
• udt is the update gate, which deter-
mines how much of the new informa-
tion to keep.
• hst is the new memory content.
• sigmoid represents the sigmoid activa- AQ5: All
tion function. figure
citation
• tanh represents the hyperbolic tangent are miss-
activation function. ing in
• Wt and bs are weight matrices and the text,
please
bias vectors, respectively, which are provide
learned during training the
appro-
priate
3.6. BERT citations.
In this paper, BERT serves as a powerful nat- AQ6:
Please
ural language processing model that excels at provide
understanding the contextual meaning and Figure 43.1. Hybrid GRU-BERT sentinel. source
for all the
nuances of text. Its primary role as a classifier Source:
figures.
AUTHOR'S SELF GENERATED
lies in extracting meaningful features from the
preprocessed review text and utilizing them
to make accurate predictions about whether schedules, gradient clipping, and warmup
a review is genuine or fake. BERT employs steps are often used for BERT models to fine-
a deep transformer architecture that enables tune performance.
it to capture intricate relationships between
words, phrases, and sentences. 3.7. Algorithm
Furthermore, BERT’s proficiency in han-
1. Input Xtrain, Ytrain, Hyper Parameters.
dling long-range dependencies and its ability
Xtrain: Training input data
to comprehend the overall content of a piece
Ytrain: Training labels for the input data
of text is particularly advantageous in fake
2. Hyper Parameters for GRU: Number of
review classification. Reviews often consist of
GRU units, learning rate, etc.
multiple sentences or even paragraphs, and
Hyper Parameters for BERT: BERT model
BERT’s architecture enables it to consider the
configuration, Learning rate, etc.,
entire context, rather than just isolated words
3. Initializing Parameters:
or phrases.
GRU Parameters:
Overall, BERT’s role as a classifier in fake
Weights and biases for the GRU gates
review detection is pivotal, as it leverages its
and hidden state transformation.
deep contextual understanding to make highly
BERT Parameters:
accurate and nuanced predictions regarding
Load pre-trained BERT model.
the authenticity of reviews. Hyperparameters
4. Initial step of GRU Preprocessing, set ini-
AQ4: for GRU and BERT models typically include
Please tial hidden state: ht–1 = 0
learning rate, batch size, number of epochs,
provideADDED
NOTE 5. Process the input Sequence through GRU
the dropout rate, optimizer type, and weight
expan- for each time step t 1 to 4
decay. For optimization strategies, common
sion for rst ← { Wtirsxt, Wthrsht–1, bshrs, bsirs}
RMS and choices include Adam, RMSprop, and SGD
SGD. with momentum. Additionally, learning rate udt ← {Wtiudxt, Wthudht–1, bshud, bsiud}
Harmful content classification in social media using gated recurrent units and bidirectional encoder 325
hst ← {Wtihsxt, rt * (Wthhsht–1, bshsn), bsihs} Existing paper 1 achieved an accuracy of
85.28% while existing paper 2 demonstrated
ht ← {(1 – udt) * hst, udtht–1}
95.14%. In contrast, the proposed paper
6. Obtain the Final Hidden state outperformed both with an impressive 98%
Fgru ← {ht} accuracy.
7. Tokenize and encode the reviews using a
BERT tokenizer pass the tokenized input
through the BERT and obtain the classifi-
cation output.
8. Compute the loss function and do the
backpropagation and Optimization
9. Repeat the 5 to 8 steps for a certain num-
ber of epochs or until convergence.
10. Preprocess input speech sequences using
GRU (steps 3 to 7), then input pre-
processed data into BERT classifier for
class prediction, optionally choose a class Figure 43.2. Accuracy.
with higher probability. Source: AUTHOR'S SELF GENERATED
4. Experimental analysis
4.1. Dataset
The Toxic comment classification challenge
dataset is used and it is available on the
Kaggle website. this dataset has 7 columns
toxic, threat, severe toxic, obscene, insult,
dentity_hate [13]. The advantage of using
this GRU and BERT is that it gives better
output.
Figure 43.3. Precision.
5. Result Source:
AUTHOR'S SELF GENERATED
By considering our results, we got efficient
accuracy, precision, recall, F1 score, and low
loss compared with the Existing model. These
algorithms are thoroughly evaluated for their
effectiveness in identifying and classifying
harmful content on social media platforms,
taking parameters such as accuracy, preci-
sion, recall, and F1 score. In our experiment,
we have considered 4 epochs in the range of
25,50,75 and 100. This research not only
provides practical implications for improv-
ing social media but also lays the foundation
for future advancements in ongoing efforts to Figure 43.4 Recall
create safer online spaces. Source:
AUTHOR'S SELF GENERATED
326 Innovations in Computational Intelligence: Trends and Technologies
Existing Paper 1 achieved a precision of Acknowledgment
90.28% while Existing Paper 2 demonstrated
We, the authors gratefully acknowledge the
94.96%. In contrast, the Proposed Paper out-
staff and authority of the computer science
performed both with an impressive 98.29%
and engineering department for their coop-
precision.
eration in the research
Proposed Paper achieved a remarkable
98.69% recall, surpassing Existing Paper 1
(81.49%) and Existing Paper 2 (95.28%). 6. Conclusion
Existing Paper 1 achieved a F1-Score of
To detect hateful and abusive comments on
81.39%, while Existing Paper 2 demonstrated
social media, this study complements exist-
95.96%. In contrast, the Proposed Paper out-
ing data sources with a rich dataset. The
performed both with an impressive 98.49%
study uses multiple deep learning-based
F1-Score. `
classification techniques, based on the harm-
Existing Paper 1 achieved a loss of 0.45%,
ful comment classification challenge data-
while Existing Paper 2 demonstrated 0.48%.
set. The effectiveness of these algorithms in
In contrast, the Proposed Paper outperformed
locating and classifying dangerous content
both with less loss of 0.19% loss.
on social media platforms is carefully evalu-
ated. Using BERT for classification and GRU
for preprocessing together is a powerful way
to improve the identification of hazard-
ous information on social media [10]. This
method not only increases the precision with
which hazardous content is categorized, but
it also lays the groundwork for the creation
of more robust content moderation systems
in the ever-changing social media environ-
ment. The suggested hybrid approach out-
performs current techniques for classifying
dangerous information, with a 98 percent
accuracy rate. we have trained our model on
one particular language and we are looking
Figure 43.5. F1-Score. forward to training our model on different
Source: languages in the dataset if we have similar
AUTHOR'S SELF GENERATED
data with positive and negative reviews then
the model trained itself by backpropagation
for better output.
References
[1] Bangyal WH, Qasim R, Rehman NU,
Ahmad Z, Dar H, Rukhsar L, Aman Z, and
Ahmad J. Detection of fake news text clas-
sification on COVID-19 using deep learning
approaches. 2021.
[2] Eke CI, Norman AA, and Shuib L. Context-
based feature technique for sarcasm identi-
Figure 43.6. Loss. fication in benchmark datasets using deep
Source: learning and BERT model. IEEE. 2021.
AUTHOR'S SELF GENERATED
Harmful content classification in social media using gated recurrent units and bidirectional encoder 327
[3] Zhang Z, Robinson D, and Tepper J. [7] Qian Y and Huang J. A self-attentive convo-
Detecting hate speech on Twitter using a lutional neural network for emotion classifi-
convolution-GRU based deep neural net-
cation on user-generated content. 2021.
work. 2018; 03 June 2018. [8] Tufail H, Ashraf MU, Alsubhi K, and Aljah-
[4] Fan H, Du W, Dahou A, et al. Social media dali HM. The effect of fake reviews on
toxicity classification using deep learning. e-commerce during and after the COVID-19
Electronics. 2021. pandemic. 2021.
[5] Yousaf K. A deep learning-based approach [9] Kaliyar RK, Goswami A, and Narang P. Fake
for inappropriate content detection and clas- Bert: Fake news detection in social media with
sification of YouTube videos. IEEE. 2022. a BERT-based deep learning approach. 2021.
[6] Lee W, Lee SS, Chung S, and Dongun. Harm- [10] Mazari AC, Boudoukhani N, and Djeffal A.
ful contents classification using the harmful BERT-based ensemble learning for multi-
word. 2021. aspect hate speech detection. Springer. 2023.
44 Detection of missprounciation using
deep learning
Naga Vardhani Ba, Naga Nandini Kb, Nandhakeerthi Mc, and
Rupanjali Md
Department of CSE, Vignan’s Nirula Institute of Technology and Science for Women, Peda
Palakaluru, Guntur
Abstract: The voice recognition community is currently paying more attention to “mispronunciation detection”.
This study interest and the focus of this work are primarily driven by two areas of application: speech recogni-
tion and language learning adaption. There are many Systems for CALL that use CAPT methods. This thesis
introduces a new text-dependent mispronunciation method based on AFCC for text. This approach is demon-
strated to perform better than the traditional HMM method based on MFCCs. To assist language learners in
recognizing and correcting pronunciation errors, a PCA-based system for mispronunciation detection and clas-
sification is also developed. Two projects have been investigated in order to improve voice recognition through
adaptation. As one method of making grammar-based name recognition adaptive, the initial one enhances name
awareness by teaching permissible variances when pronouncing names. The second project involves detecting
accents by looking at how fundamental vowels vary in speech with accents. It has been demonstrated that this
method, which detects accents using both acoustic and phonetic information, is effective with accented English.
To enhance name and speech recognition, these apps able to incorporated into a foreign phone system that is
automated. Based on a short speech sample, it estimates the accent of the caller. For better detection outcomes,
it moves away from the default switching from an accent-adaptive speech recognition engine once the type of
accents is identified.
Keywords: Convulution Neural Network, BiLSTM, Detection of misprounciation, Deep Learning, CALL,
MFCC
1. Introduction that can effectively detect mispronunciations.
The CNN extracts relevant features from the
Mispronunciation detection is a task in speech
audio, and the BiLSTM captures the tempo-
processing where we aim to identify errors
ral dependencies and context. This combined
or deviations in the pronunciation of words
or phrases. CNNs, or Convolutional Neural approach allows the system to learn and iden-
Nets, are commonly used verbally processing tify patterns of mispronunciations.
jobs to extract relevant features from audio In mispronunciation detection, CNNs are
data. They can capture patterns and varia- used to extract relevant features from audio
tions in the speech signal that are indicative data. These features capture important pat-
of mispronunciations. On the other hand terns and variations in the speech signal. The
BiLSTM models are recurrent neural net- CNN architecture consists of completely
works that can capture temporal dependen- linked layers, pooling layers, and convolu-
cies in the speech signal. By combining CNN tional layers. The filters that convolutional
and BiLSTM models, we can create a system layers apply to the contributions audio,
[email protected]; [email protected]; [email protected];
a
DOI: 10.1201/9781003606635-44
Detection of missprounciation using deep learning 329
extracting local qualities. The layer pooling synthesis Daniel Korzekwa. The three
downsample the output, lowering the num- cutting-edge methods to produce appro-
ber of dimensions. Finally, the completely priately pronounced and mispronounced
linked levels combine The characteristics and synthetic speech are founded on T2S, S2S,
make predictions. and P2P conversion. We demonstrate that
BiLSTMs, on the other hand, are a kind these methods help set a new standard for
of RNN, or recurrent neural network, that the area and enhance three machine learn-
is able to recognize temporal connections ing algorithms’ accuracy model faults. We
in the speech signal. They process the input offer a comprehensive analysis of three dis-
sequence both backwards and forwards. To tinct speech-generating methods depend-
detect mispronunciations, a dataset of cor- ing on the conversion of P2P, T2S, and S2S
rectly pronounced and mispronounced words for identifying pronunciation problems.
or phrases is used for training the model. The 3. Identification of Mispronounced Words
CNN extracts features from the audio, and the in Stockle, Perdigo. Find candidate word
BiLSTM captures the temporal dependencies. pronunciations in the first step by per-
Most people try to learn as many words mitting syllable repeats and erroneous
as they can when they first begin to learn a beginnings. In In a subsequent stage, the
language. As a result, they can first learn a pronouncability of word candidates is
new word, then put it together into a phrase determined. For a free phone loop and a
and, hopefully, a sentence. Finally, they try to word-spotting model in the vicinity of the
communicate with people whose native lan- candidate segmentation, the log-likelihood
guage is not English but who share an interest ratiois determined to be the most effective
in speaking it due to circumstances at work, feature for classifying mispronunciations.
in the schools where they work, or even with It can help anticipate how well a kid will
people for whom English is their mother read, and it is necessary to look into the
tongue or who use it as a lingua franca, in effects of the improvements provided
other words, they are native English speakers. by the approach suggested in this paper.
First, there is a chance of inaccuracies in
the phonetic recognizer’s output. The sub-
2. Literature Survey sequent issue this is the arbitrary manu-
1. Enhanced mispronunciation detec- ally noting appropriate terms additionaly
tion using logistic regression classifiers pronunciation errors.
founded on transferrable skills and DNN/ 4. Automatic Estimation of Young Read-
Yong Wang. 2015 Using LR classifiers ers’ Proficiency for Advanced Literacy
based on transfer learning and an auditory Evaluation by Matthew P. Black, Eleven
model acquired on a DNN, we can get Human Assessors assessed the children’s
better mispronunciation detection perfor- overall reading abilities in the first stage
mance. The DNN training with improved of the research using the audio recordings.
discrimination refines the model of sound We examined how the evaluator’s experi-
developed via the traditional HMM-based ence affected the agreement amongst the
GMM method. To give language learners assessors. It enables evaluators to auto-
more meaningful diagnosis feedback, we matically model and forecast assessors’
will look into the detection of prosodic, high-level judgments for a specific read-
insertional, and deletional mispronuncia- ing assignment as well as adapt to the
tions faults in the future. subjects’ speaking styles. When it comes
2. Training in computer-assisted pronun- to supporting material that is both audio-
ciation: Nearly all you require is speech and video-based.
330 Innovations in Computational Intelligence: Trends and Technologies
5. Pronunciation Scoring Using Dynamic 3. Proposed Model
Time Warping and Pronunciation Quality
Geetika Dand, Charu Gupta, Pankaj Dad- Convolutional Neural Nets have demon-
heech. This work suggests an innovative strated their highly efficient in a range of
approach to getting beyond these restric- computer vision jobs, and their application
tions by combining the prosodic, fluent, extends to the realm of audio and speech
accurate, and completeness ratings. The processing. In the context of mispronuncia-
recommended approach yields results that tion detection, CNNs play a pivotal role in
may be customized for both novice and capturing spatial patterns within audio spec-
seasoned learners, and it closely reflects. trograms. Spectrograms provide a visual rep-
The mean score of experts in the dataset. resentation of the frequency content of audio
6. CNN Properties of the Transfer Learn- signals over time. This allows the model to
ing-Based Arabic Phoneme Model: Mis- discern subtle variations in how phonemes
pronunciation Detection Mustanser are articulated, contributing significantly
Ali Ghazanfar. This work explores the to the overall accuracy of pronunciation
application of deep convolutional neu- assessment.
ral networks for Arabic phoneme mis- BiLSTMs, by processing input sequences in
pronunciation identification. In order both forward and backward directions, pro-
to determine which CNN layer’s fea- vide a holistic understanding of the temporal
tures are the best, we extracted features dynamics of speech. They are well-suited to
and applied the classification algorithms grasp long-term dependencies and contex-
KNN, SVM, and NN to the charac- tual information, which is crucial in discern-
teristics of every layer. Within order to ing variations in pronunciation. The synergy
determine which layer features are the between CNNs and BiLSTMs enables the
best. We took characteristics out of sev- model to leverage both spatial and tempo-
eral CNN strata and applied categoriz- ral insights, resulting in a more nuanced and
ing algorithms. Each layer’s features are accurate analysis of speech patterns.
assigned KNN, SVM, and NN. Mispronunciation detection using a hybrid
7. Fellow, a Dual-Pass Structure for Error of BiLSTM and CNN is a powerful approach
Identification and Correction in Com- that leverages the strengths of both architec-
puter-Aided Speech Recognition Training tures. This hybrid model excels in capturing
Its goal is to get over the problems with both spatial and temporal features, making it
the current Extended Recognition Net- particularly effective in discerning subtle var-
works (ERNs)-based technique. iations in pronunciation. The combination of
8. With deep neural networks with many CNN and BiLSTM addresses the limitations
distributions, Kun Li, Xiaojun Qian pro- of previous models. Oversimplified models
vide an diagnosis and recognition system might struggle with complex pronunciation
for mispronunciation in conversation in patterns, lacking the capacity to capture both
English (L2). Its purpose is to get over the spatial and temporal aspects effectively. The
problems with the current Extended Rec- CNN-BiLSTM hybrid overcomes this limita-
ognition Networks (ERNs)-based tech- tion, providing a more comprehensive analy-
nique. By adding the expected Together sis of speech data. During training, the CNN
with the official transcriptions and the learns to recognize spatial patterns indicative
target words’ phonetic mistake patterns, of different pronunciations.
the ERNs limit the search space and make For the training of this model we have
use of the current state of automated voice used the dataset named Common voice.
recognition technology. The Common Voice dataset is a large-scale,
Detection of missprounciation using deep learning 331
open-source initiative developed by Mozilla The training process involves defining
to collect and share multilingual and diverse appropriate loss functions and optimizers,
speech data for training automatic speech with a keen focus on monitoring validation
recognition (ASR) systems. loss to prevent overfitting. Evaluation metrics
Mispronunciation detection using CNN that shed light on the model’s performance
and BiLSTM with the Common Voice dataset include accuracy, precision, recall, and F1
is a complex and multi-faceted process that score. Testing on a separate dataset allows
involves several critical steps. Starting with for an unbiased assessment, and misclassifi-
the preparation of the Common Voice dataset, cations are analyzed to identify patterns and
including downloading, preprocessing, and guide further refinements.
splitting dividing test, validation, and training The mispronunciation detection model,
sets, the project aims to create a robust model employing CNN, BiLSTM with the Common
capable of accurately detecting mispronun- Voice dataset, exhibits a myriad of advan-
ciations across diverse languages and accents. tages that collectively contribute to its efficacy
Moving to the model architecture (Fig. 44.1), a in tackling the challenges inherent in diverse
hybrid of BiLSTM, CNN is employed. CNN languages, accents, and audio patterns. The
layers are utilized for spatial feature extracti- Common Voice dataset encompasses a rich
on, capturing patterns in spectrograms, while tapestry of languages, and the model excels at
BiLSTM layers handle temporal analysis, extracting spatial features from audio spec-
capturing dependencies in the audio data. trograms, making it adept at recognizing pro-
nunciation nuances across diverse linguistic
contexts. This adaptability is crucial in real-
world scenarios where language variations
are abundant. The incorporation of Bidirec-
tional LSTMs is a key factor in the model’s
success.
Accent robustness is a notable advantage
of this model. Accents pose a significant chal-
lenge in mispronunciation detection due to
their variability. However, the adaptability of
the CNN-BiLSTM model to both spatial and
temporal features makes it resilient to accent
variations. The effective handling of tempo-
ral dependencies is a hallmark of the BiLSTM
layers. Accent robustness is a notable advan-
tage of this model. Accents pose a significant
AQ1: challenge in mispronunciation detection due
Please
provide to their variability. However, the adaptability
TATION ADDED
in-text of the CNN-BiLSTM model to both spatial
citation
for figure
and temporal features makes it resilient to
1. accent variations. The effective handling of
temporal dependencies is a hallmark of the
AQ2: BiLSTM layers.
Please
provide While the mispronunciation detection
source Figure 44.1. CNN and BiLSTM architecture model employing CNN and BiLSTM with
for all the model.
figures
the Common Voice dataset offers notable
Source: advantages, it is crucial to acknowledge and
AUTHOR'S SELF GENERATED
332 Innovations in Computational Intelligence: Trends and Technologies
understand its limitations. These limitations
span aspects of data variability, generaliza-
tion challenges, potential biases, and the com-
putational complexity inherent in the model.
One significant limitation stems from the var-
iability inherent in speech data. Handling of
noise and variations in acoustic conditions is
a challenging aspect.
Computational complexity is a practi-
cal limitation, particularly in resource-con-
strained environments. The CNN-BiLSTM
model, being a deep learning architecture, Figure 44.2. Accuracy.
requires significant computational power Source: AUTHOR'S SELF GENERATED
for training and inference. Strategies such
as model quantization or using lightweight
architectures may be necessary to address this
limitation. In conclusion, while the mispro-
nunciation detection model with CNN and
BiLSTM leveraging the Common Voice data-
set offers notable advantages, it is imperative
to recognize its limitations. These limitations
encompass challenges related to data vari-
ability, generalization, biases, interpretability,
noise handling, computational complexity,
and the need for ongoing dataset enrichment.
Acknowledging these limitations is essential Figure 44.3. Loss.
for the responsible deployment and continu- Source:
ous improvement of mispronunciation detec- AUTHOR'S SELF GENERATED
tion models in real-world applications. case, it’s 96%, of the examples were correctly
identified by the model.
Training loss in deep learning measures
4. Results how well a neural network fits training data.
Detecting mispronunciations using a hybrid Minimizing this loss shows the network’s
BiLSTM and CNN achieved an accuracy of learning and improved pattern recognition.
96%. In addition to accuracy, you can evalu- Lower training loss indicates a well-trained
ate the model’s performance using param- model.
eters including F1-score, recall, and precision. Precision gauges a model’s accuracy in
These metrics offer a more thorough evalua- predicting positive instances among all its
tion of the model’s ability to correctly classify positive predictions. RNN and BiLSTM mod-
mispronunciations. els typically achieve good precision in many
When compared with existing Bi-LSTM tasks but may face challenges capturing long-
and RNN model our proposed model gave range dependencies in data.
the best results. The evaluation metrics com- Recall quantifies how well a model can
paring the three models are given in the identify all relevant instances among all
Figures 44.2–44.4. actual relevant instances in a dataset. The
Accuracy is the most straightforward met- proposed Attention-based model outper-
ric, symbolizing the percentage of accurately forms CNN and BiLSTM models in terms of
classified samples of the total dataset. In your recall, as illustrated in Figure 44.5, indicating
Detection of missprounciation using deep learning 333
its superior ability to capture and identify rel- 5. Conclusion
evant data points.
The proposed model outperforms exist- In this work, we suggested several approaches
ing models with a higher F1-Score in Figure to detection of mispronunciation improve-
44.6. Series1 corresponds to BiLSTM, Series2 ment. The CNN-BILSTM, initially we
to CNN, highlighting the superior balance of extracted complex features from the Spectro-
precision and recall achieved by the proposed grams that are produced from the datasets
model. using CNN. And this output from the CNN
is given as an input for the process of BIL-
STM Which analysed the process in both for-
ward and backward approach and produced
an acccurate results. The process of overfit-
ting is been resolved through this model using
Batch Normalization which each and every
hidden layer is mean normalized converting
the datasets into batches which overcome the
problem of overfitting.
The future improvement of this model is
that not only detecting the mispronunced
Figure 44.4. Precision. words but also can guide the person with the
corrected words by using the help of AI. AI-
Source: AUTHOR'S SELF GENERATED
driven correction systems can analyze a user’s
speech in real-time, highlighting mispronunci-
ations and suggesting correct pronunciations.
They offer visual and auditory feedback,
guided practice exercises, and track a user’s
progress over time. These systems provide
customized recommendations for improving
pronunciation, considering individual chal-
lenges and speech factors.
References AQ3:
Please
[1] Cohen, Herman, and R. T. Craig. “The his- provide
Figure 44.5. Recall. in-text
tory of speech communication: The emer- citation
Source: AUTHOR'S SELF GENERATED gence of a discipline.” Annandale, VA: for all
Speech Communication Association. Plural- references
ism. Disagreement, and the Status of Argu-
ment in the Public Sphere 137 (1995).
[2] Hu, Wenping, et al. “Improved mispronun-
ciation detection with deep neural network
trained acoustic models and transfer learning
based logistic regression classifiers.” Speech
Communication 67 (2015): 154–166.
[3] Çalik, Şükrü Selim, Ayhan Küçükmanisa,
and Zeynep Hilal Kilimci. «Deep Learning-
Based Pronunciation Detection of Arabic
Phonemes.» 2022 International Conference
Figure 44.6. F1-Score. on INnovations in Intelligent SysTems and
Source: Applications (INISTA). IEEE, 2022.
AUTHOR'S SELF GENERATED
334 Innovations in Computational Intelligence: Trends and Technologies
[4] Proença, Jorge, et al. “Mispronunciation [6] Li, Kun, Xiaojun Qian, and Helen Meng.
Detection in Children’s Reading of Sen- “Mispronunciation detection and diagnosis
tences.” IEEE/ACM Transactions on Audio, in l2 english speech using multidistribution
Speech, and Language Processing 26.7 deep neural networks.” IEEE/ACM Transac-
(2018): 1207–1219. tions on Audio, Speech, and Language Pro-
[5] Black, Matthew P., Joseph Tepperman, and cessing 25.1 (2016): 193–207.
Shrikanth S. Narayanan. “Automatic predic- [7] Qian, Xiaojun, Helen Meng, and Frank Soong.
tion of children’s reading ability for high- “A two-pass framework of mispronunciation
level literacy assessment.” IEEE Transactions detection and diagnosis for computer-aided
on Audio, Speech, and Language Processing pronunciation training.” IEEE/ACM Transac-
19.4 (2010): 1015–1028. tions on Audio, Speech, and Language Pro-
cessing 24.6 (2016): 1020–1028.
45 IOT-enabled cloud solutions for reliable
health monitoring
Lakshmi Sravani G.a, Lalitha Chandrika N.b, Lavanya S.c,
Vijaya Lakshmi S.d, and Sri Lakshmi Prasanna S.e
Computer Scienceand Engineering, VNITSW, Guntur, India
Abstract: Contemporary hospitals are meeting Client needs with new technology like smartwatches or the web
of items. Wearable devices are nothing but smart watches, wrist bands etc., because it provides more freedom
for far removed patient’s connections via the web of technologies and the tracking of medical data. The new
technology offers more features and improvements to the current healthcare services. When introducing wear-
able technology to the healthcare industry, there are a number of security concerns that must be taken into
account, including Medical information is safe and private. We present a comprehensive approach to building a
secure health monitoring system that leverages cloud computing and IOT technologies.
Keywords: A system of multivital signals, electronic wearables, Internet of things, monitoring system for health,
preprocessing of data, protection of privacy, security issues, the XLNET deep learning algorithm, wearables
cloud computing
1. Introduction the security algorithms. And here, I used
an algorithm called Advanced Encryption
There is a sharp rise in the number of patients Standard (AES). The medical sector runs on
in many nations, and it is getting harder for safe evaluating environments in the cloud.
people to see primary care physicians or The things that are on the Internet system
other providers. The past few decades, the encrypts patient data and stores it in a cloud
emergence of wearable technology and Inter- database using a 128-bit AES approach. With
net of things (IOT) has enhanced the remote a login passkey that may be used to decrypt
patient monitoring to improve patient quality cloud data back into its original format, only
of treatment. Body sensor networks make it authorized users are permitted access to the
very convenient to identify a patient’s physi- information maintained in the cloud.
cal abnormalities and administer the appro-
priate care when needed. Sensors pick up
physiological information Such as heart rate, 2. Literature Survey
blood pressure, temperature, and electrocar- [1] Remote Case Pall determining in con-
diogram). By employing a permutation-based junction with the World of impact system
training strategy, it enhances the autoregres- for tracking and classification. The paper dis-
sive language modeling of models such as cusses the security challenges of using IOT in
generative pretrained transformer (GPT), remote case monitoring. It offers perceptiv-
message, etc. To provide security for the data ity into encryption, access control, and secure
we get from the IOT devices we should use data transmission.
[email protected]; [email protected]; [email protected];
a
[email protected]; [email protected]
d
DOI: 10.1201/9781003606635-45
336 Innovations in Computational Intelligence: Trends and Technologies
[2] Author Sabzevar Branch published this so, and The World Wide Web of The events
paper on 8 September Springer (2021) An IoT for naturally options in an isolated location.
-based wellness tracking system that takes [8] Author Mohammed Almaiah pub-
into account the connected home’s clever data lished this on IEE Article (2017). Advanced
for case-centered health surveillance. Pall treatment operation IOT bias help track
computing in conjunction with an Online of the administration of medicines and the
Impacts (IoT) structure for faraway situation response to the treatment and reduce medi-
watching (RPM) is a novel approach that has cal crimes. Smart Health Solution Integrat-
the potential to transform medicine. ing IOT and pall. Medical data availability
[3] Author Ashish Singh published this of electronic medical records allow cases to
paper on 2 September Springer (2022) The admit quality care and help healthcare pro-
primary disadvantages of this approach is viders make the right medical opinions and
that maintaining the shop’s premises and help complications
ensuring the security of the massive volume of [9] Author Ghulam Muhammad published
continuously generated information present an article in IEEE in January 2017. In this
tough obstacles for the traditional archive work, the viability of tracking speech disease
form. The main objective of this endeavor is with the IOT and pall integrated. We suggest
to address the socially harmful interest prob- an architecture in which conversations are
lem outlined by the European Communities. obtained via the Internet of Things and sent
[4] Zulfiqar Ali Ghulam published this to a smartphone-like serving gadget.
paper on 09 March IEEE (2017) in nations [10] Author M. Shamim Hossain pub-
that are developing and developed alike, such lished a paper in Elsevier (2017) to this end.
as the United States and India, health concerns This paper describes an IOT-grounded health
among the elderly are a major business. Due monitoring frame, where health monitoring
to various health issues, an important num- signals are authenticated. While IOT pledges
ber of medical setups attract elderly citizens. to reduce the cost of healthcare in the long
[5] A protect patiently health tracking sys- term, the cost of its perpetration in hospitals
tem based on IOT A case tracking solution and staff training is relatively high.
based on IOT enables patients to benefit from
healthcare-related programs. Settling in their
houses in a distant spot 3. Proposed Model
[6] Authors Hui Shao et al. published an In our model, I have employed a deep learn-
article in the Magazine of Transformed Tech- ing method to extract and classify data from
nology and Intelligent Environments in 2018 the physical activity dataset. This algorithm
.This comprehensive review covers color- collects necessary information and data,
ful aspects of IOT in healthcare, including which is then encrypted using the AES tech-
patient monitoring, security, and sequestra- nique for security. Let’s see the work flow of
tion. Case’s sequestration, in this instance, XLNet algorithm
safety and protection are absolutely crucial. Algo:
Techniques for Encrypting Healthcare Step 1: Starts
Journalism based on Internet of Things and Initialize Transform layers and Attention
Cloud Computing for Software in Healthcare heads as L and H
Emergencies Transform layers=L
[7]. The disease tracker that is discussed Attention head=H
in the above article is virus-free, affordable, Step 2:
actual time, and accountable. It offers imme- Declare Query, key and Value matrices as Q,
diate fashion reveal surveillance, pall doing K, and V
IOT-enabled cloud solutions for reliable health monitoring 337
Step 3: Step 8:
Calculate dot product matrices in step 2 with Compute Gradient during fine-tuning by the
Input vector below formula
Step 4: Step 9:
Compute attention scores by applying Soft- Evaluate θ by descent update rule as below
Max function and multiply with Value vector
Step 10: End
Tasks involving natural language process-
ing in data preprocessing can be handled by
Step 5:
XLNet. As part of the data preprocessing
Calculate feed-forward transformation by
pipeline, XLNet can be used for text classi-
using ReLU activation function
fication, sentiment analysis, or named entity
recognition if you need to analyze or classify
textual data. XLNet is useful for deriving
insights from textual patient data, including
Step 6: electronic health records (EHRs). Based on
Apply normalization to Xp called Output past patient records, this can help with fore-
casting, spotting trends, and offering indi-
vidualized care suggestions. For chatbots,
voice assistants, or text-based interactions,
XLNet can be used to effectively compre-
Step 7: hend and respond to user requests if the
Compute gradient loss to update the model health monitoring system needs to process
parameters. Gradient Calculation Parameter and understand natural language queries or
(θ): interactions with users, abnormalities or pos-
sible security breaches can be found. While
XLNet can be an asset for all kinds of NLP-
related tasks within your project, it’s vital to
keep in mind that NLP is not the only aspect
of secure health monitoring. In addition, real-
time data processing, data transfer, storage,
AQ1:
Please
provide
figure
citations
in text.
AQ2:
Please
provide
source
for all the
figures.
Figure 21.1. Architecture of reliable health monitoring.
Source: AUTHOR'S SELF GENERATED
338 Innovations in Computational Intelligence: Trends and Technologies
security, and interaction with IOT devices are A famous symmetrical encryption method
all involved. security of data, compliance, and is called AES, or Enhanced Cryptography
the encryption of communications between System.
cloud servers, Internet of Things devices, and This indicates that it generates a total of
other system components. 128 /192/256bits of encrypted cipher text as
output after receiving 128/192/256 bits as
1. Wearable technology is used by the user input.
(Smart watch, wrist bands, fitness, tracker,
etc.).
2. Health data is gathered and sent to 4. Experimental Results
The cloud via wearable technology. The graph represents the models of three dif-
a. Real-time or periodic data collection ferent those are:
is done. CNN, DLMNN and XLNET
b. Information is sent to the cloud safely.
3. Storage on the Cloud
a. Cloud storage platform called Google
Cloud securely store data.
b. To protect data when it’s not in use,
use AES encryption.
4. Data preprocessing a. Data is cleaned and
formatted during the preprocessing stage.
a. Missing values are addressed and the
data is ready for examination.
5. Extraction and Classification Using Deep
Learning
a. The XLNet algorithm is employed for
feature extraction and categorization Figure 21.3. Accuracy data.
of health data. Source:
AUTHOR'S SELF GENERATED
b. The features and classifications that
were extracted are saved for later study. Accuracy data comparison analysis is
based on a plot with the y-axis representing
the accuracy metric and the x-axis represent-
ing the epochs.
Figure 21.2. AES structure. Figure 21.4. Precision data.
Source: Source:
AUTHOR'S SELF GENERATED AUTHOR'S SELF GENERATED
IOT-enabled cloud solutions for reliable health monitoring 339
Based on the precision data comparison
graph over precision and epochs, it is evi-
dent that the XLNET consistently outper-
forms as stays at a higher value compared to
the DLMNN which show a relatively stable
precision.
Based on the recall data comparison graph,
with the y-axis representing the recall score
and the x-axis representing epoch values, it
appears that the XLNET model consistently
has a higher value over epochs indicating a
higher recall score and outperforms both the Figure 21.7. F-score data.
DLMNN model, which shows intermediate Source:
AUTHOR'S SELF GENERATED
performance and then followed by Random
CNN in terms of recall.
The loss data comparison graph above epoch value whereas CNN model keeps on
shows that the DLMNN is constant over the increasing and the XLNET model have few
fluctuations with some dips and spikes over
the epoch value
Here in the F-score Data Comparison
graph, the three models have minimal fluc-
tuations over the epoch value in which only
the XLNET model increase with the value
whereas the other two CNN and DLMNN
model have dips and spikes and there by
decrease in the score can be seen for these
two models.
Figure 21.5. Recall data.
5. Conclusion
Source: Finally, using services on the cloud provided
AUTHOR'S SELF GENERATED by IP in the area of healthcare monitoring
may significantly improve the dependability
and precision of healthcare institutions. Doc-
tors can offer individuals individualized care.
References
[1] Somayeh Iranpak, Asadollah Shahbahrami
and Hassan Shakeri. (2021). Remote patient
monitoring and classifying using the inter-
net of things platform combined with cloud
computing. Journal of Big Data.8, 90821.
[2] Ashish Singh and Kakali Chatterjee. (2022).
Edge computing based secure health moni-
toring framework for electronic health-
Figure 21.6. Loss data. care system. Complex Intelligent Systems.
Source: 26, 90222, 1205–1220.
AUTHOR'S SELF GENERATED
340 Innovations in Computational Intelligence: Trends and Technologies
[3] Zulfiqar Ali, Ghulam Muhammad, Moham- IoT and Cloud: A Case Study of Voice
med F. Alhamid. (2017). An Automatic Pathology Monitoring.55 (1), 11917, 69–73,
Health Monitoring System for Patients Suf- DOI: 10.1109/MCOM.2017.1600425CM.
fering From Voice Complications in Smart [7] Naif Al Mudawi. (2022). Integration
Cities.5, 30917, 3900–3908. DOI:10.1109/ of IoT and Fog Computing in Health-
ACCESS.2017.2680467. care Based the Smart Intensive Units. 10,
[4] Naveen, R. K. Sharma, Anil Ramachandran 60222, 59906–59918, DOI:10.1109/
Nair. (2019). IoT-based Secure Healthcare ACCESS.2022.3179704.
Monitoring System.101719. DOI:10.1109/ [8] V Tamilselvi, S Sribalaji, P Vigneshwaran, P
ICECCT.2019.8868984. Vinu, J. GeethaRamani. (2020). IoT Based
[5] Samira Akhbarifar, Hamid Haj Seyyed Health Monitoring System.42320, DOI:
Javadi, Amir Masoud Rahmani and Mehdi 10.1109/ICACCS48705.2020.9074192.
Hosseinzadeh. (2020). Secure Health Mon- [9] Lorin Anton, Theodor Borangiu, Silviu Raile-
itoring Communication Systems Based anu, Iulia Iacob and Silvia Anton. (2018).
on IoT and Cloud Computing for Medi- Managing Patient Observation Sheets in
cal Emergency Applications.27, 697–713, Hospitals Using Cloud Services.331, 91318.
111620. [10] Ebrahim Al Alkeem, Dina Shehada, Chan
[6] [Ghulam Muhammad, SK Md Mizanur Rah- Yeob Yeun, M. Jamal Zemerly and Jiankun
man, Abdulhameed Alelaiwi, Atif Alamri. Hu. (2017). New secure healthcare system
(2017).Smart Health Solution Integrating using cloud of things. 20, 50517, 2211–2229.
46 Detection of melanoma using
DenseNet-based adaptive weighted loss
function
Rishi Agrawal
Department of CEA, GLA University, Mathura, India
Abstract: Skin cancer is currently the most prevalent and terrible disease on a global scale. Malignant mela-
noma (MM) is less common than other forms of skin cancer in terms of incidence, but it has the highest mor-
tality rate. Lesion removal can be used to treat MM when it is diagnosed at an earlier stage. However, the risk
of death increases if the diagnosis has been made later. So an early melanoma detection method is proposed
using the DenseNet-based Adaptive Weighted Loss function. The experimental work was performed on the
ISIC-2019 Dataset. The proposed model scored a balanced accuracy of 68.4% which is better than traditional
deep learning methods.
Keywords: Adaptive weighted, DenseNet, ISIC-2019, dataset loss function, malignant melanoma
1. Introduction lesion kinds and the noise in histopathology
images, creating an accurate automated skin
Malignant Melanoma (MM) is a massive lesion classification strategy is one of the
entity of cancer type, but it particularly affects promising challenges. Samples and statistical
light-colored skin people who really are sen- distribution of ISIC 2019 dataset is given in
sitive to UV rays [1]. Malignant cancer or Figure 46.1.
melanoma can be categorized into two parts To classify skin lesions, CAD systems
melanocytic and non-melanocytic. Basal cell must first perform image pre-processing,
carcinoma (BCC) and squamous cell carci- noise removal, lesion delineation, feature
noma (SCC) are the two types of non-mel- extraction from the lesion region, and lesion
anocytic cancers, whereas melanoma is a type classification using conventional IP or ML
of melanocytic skin cancer [2]. BCC and SCC algorithms [4]. For picture normalization
are typically combined and indicated as usual and classification, a variety of image pre-
skin cancer. processing techniques have been used, includ-
In recent decades, computerized dermos- ing contrast enhancement (CE), chromaticity
copy image classification using CAD methods modification, and image filtering. For image
has been introduced [3]. These modules are scaling and noise reduction, image boundary
designed to assist doctors in making deci- forecasting and segmentation are essential
sions, particularly when doing so quickly processes. For computing skin lesion segmen-
and with little prior knowledge. Numerous tation, conventional IP techniques such as
semi- or fully automated models were antici- statistical thresholding, clustering, or active
pated for use in diagnosis. Due to the simi- contours have been widely used by [5]. The
lar morphological appearance of several skin most recent segmentation methods have been
[email protected]
DOI: 10.1201/9781003606635-46
342 Innovations in Computational Intelligence: Trends and Technologies
introduced in recent years with the use of objects of various sizes in images. It featured
supervised ML techniques. the innovative Inception module, which used
Numerous clinical classifications can be multiple filter sizes within a single layer.
AQ1: carried out using CNNs and their impres-
Please
define
sive standard [6]. Additionally, CNN-related 2.2. ResNet[14]
CNN methods were mostly used for the classifica-
One of the most famous variants of ResNet
in full tion of skin lesions and performed very well
at first is ResNet-50, which has 50 layers. There
in comparison to other methods. Contrary
mention are also other variants like ResNet-18,
to conventional methods, large CNN-related
ResNet-101, and so on, which vary in terms
methods to classify skin lesions have been
of the number of layers and model complex-
applied to real or previously processed skin
ity. These deeper architectures have been
lesion pictures without image segmentation.
shown to perform exceptionally well on a
The influence of skin cancer detection on
variety of applications. ResNet and its vari-
CNN-based classifiers has not yet been fully
ants have significantly contributed to the suc-
revealed, despite the best classification results
cess of deep learning in various domains by
from Convolutional neural approaches to
enabling the training of very deep networks
skin infection classification without lesion
without facing severe optimization chal-
segmentation methodologies [7]. Only a small
lenges. ResNet introduces the “bottleneck”
number of studies have used lesion segmenta-
architecture within each residual block. This
tion features to improve model performance
helps in reducing computational complexity
in CNN-related classification workflow.
while retaining representational.
CNN employed for melanoma or non-mel-
anoma skin cancer scanning is not validated
in medical trials, which is still regarded as a 2.3. ResNeXt[15]
severe limitation [8–11]. The classification ResNeXt is a deep learning architecture that
of benign and malign lesions is carried out enhances the capabilities of CNNs by intro-
by ML in numerous diagnosing procedures, ducing grouped convolutions and the con-
predominantly, teledermatology, portable cept of cardinality. Researchers continue to
sectors, and custom vaccinations, in addition explore and adapt its principles for various
to the straightforward procedure. Knowing applications in image analysis and beyond.
what kind of DL methodology one to spec- Similar to ResNet, ResNeXt uses bottleneck
ify the usage and application to control and architectures in its residual blocks. ResNeXt
improve the methods used by individuals. architectures can vary in depth and cardinal-
ity. A typical ResNeXt model might consist
of multiple residual blocks stacked together,
2. Related Work with each block containing grouped convolu-
Recent research explores many Deep CNN tional layers.
models that can be efficiently applied for the
detection of melanoma.
2.1. Inception [13]
Here are some of the notable versions of the
Inception architecture: The Inception models AQ2:
Please
are computationally efficient due to their use provide
of 1 × 1 convolutions for dimensionality reduc- Figure 46.1. Samples and statistical distribution source
tion and the parallel processing of features. of ISIC 2019 Dataset [12]. for all the
figures.
The use of multiple filters makes it robust to Source:
AUTHOR'S SELF GENERATED
Detection of melanoma using DenseNet-based adaptive weighted loss function 343
3. Proposed Method 3.4. Data Imbalance Handling
For detection of melanoma different phases Skin image datasets may suffer from class
are applied: imbalance, where certain classes have fewer
samples than others. Address this issue by
3.1. Data Preparation using techniques like class weighting, over-
sampling, or undersampling to ensure that the
In this step, collect, and preprocess a data- model doesn’t bias toward the majority class.
set of skin images and dataset is labeled Data imbalance is managed by adaptive
with appropriate class labels, such as benign, weighting loss function [16]:
malignant, or specific skin diseases. Data aug-
mentation like random rotations and flips are • A weighting coefficient Ci is used to make
used to increase the diversity of dataset. training effective.
• The weighting coefficient is Cy.
3.2. Model Selection • Variable p is used to detect outliers.
DenseNet architecture is used on the basis of • Hyperparameter α is used for the fine-
dataset size and complexity. Transfer learn- tuning of weights.
ing and Fine-tuning is applied to make model • The adaptive weighting loss function can
faster and efficient. be defined as shown in Eq. (4).
3.3. Lesion Segmentation (4)
Segmentation is a crucial step in skin lesion
categorization. In addition to addressing
issues like color changes, the presence of hair,
and pathological inconsistency, an analysis 3.5. Classification and Detection
tools method can detect diseased areas.
• The Sobel–Feldman operator is used to By following the above steps and customizing
compute gradients initially, with either a them to the skin image classification task, the
fixed Gaussian kernel of (3 × 3) proposed model effectively used DenseNet to
• Each channel should be divided into equal- develop an accurate and reliable model for
sized units (4, 8, 12, etc.), which should skin image analysis and diagnosis as shown
then be arranged in decreasing size. Each in Figure 46.2. DenseNet-201 is explored for
block is now given a weight based on the the proposed model [17]. The following steps
gradient’s strength. as shown in Eq. (1), are used to apply DenseNet201 architecture
Eq. (2) and Eq. (3). as shown in Figure 46.3.
• A convolutional Neural network is
(1) applied to Image m0.
• The Network contains N layers and non-
(2) linear transformation is implemented.
• Skip-connection is used to overcome the
problem of vanishing gradient. The iden-
(3) tity function used in architecture can be
defined as [18] Eq. (5):
(5)
where the weight coefficient w(i=1, ...,
4) reflects threshold values in relation to • The features maps of all preceding layers
computed gradient. as shown in Fig. 3. Eq. (6):
344 Innovations in Computational Intelligence: Trends and Technologies
m0, m1, …, mp–1 as input: A comparison analysis is demonstrated in
Table 46.1. 30 epochs are experimented to
(6) check the accuracy of proposed method as
given in Figure 46.4.
Table 46.1. Comparison with other state-of-the-
art methods
Model BACC Model BACC
VGG-11 61.7 ResNet-34 66.3
Figure 46.2. Melanoma detecting Uging
VGG-13 62.2 ResNet-50 66.7
DenseNet-201.
VGG-16 62.5 ResNet-101 66.9 AQ3:
Source: AUTHOR'S SELF GENERATED Please
VGG-19 63.1 SENet-50 66.6 provide
ResNet-18 65.8 Proposed 68.4 source for
the Table.
Source: AUTHOR'S SELF GENERATED
Figure 46.3. Feature concatenation process of
DenseNet [17].
Source: AUTHOR'S SELF GENERATED
Figure 46.4. Proposed model performance graph
4. Result Analysis for training and validation.
In medical diagnosis, recall is used to assess Source: AUTHOR'S SELF GENERATED
how effectively a diagnostic test or model can
identify individuals with a particular medical
condition. High recall means that the model is
5. Conclusion
good at catching cases with the disease, mini- In this paper, detection of Melanoma Using
mizing the risk of false negatives. The pro- DenseNet-based adaptive weighted loss func-
posed method is implemented on ISIC2019 tion is performed. The proposed method used
Dataset. In cancer screening, a high recall rate regularization techniques like dropout and
is crucial because missing a cancer diagnosis weight decay to prevent overfitting. The model
can have life-threatening consequences. For used techniques like gradient-based visuali-
example, in mammography for breast cancer zation or heatmaps to interpret the model’s
detection, high recall or BACC ensures that predictions and understand which regions or
the majority of actual cancer cases are identi- features in the skin images are contributing to
fied. The balanced accuracy (BACC) can be the classification decisions. The experimental
calculated as Eq. (7): work was performed on the ISIC-2019 Data-
set. The proposed model scored a balanced
accuracy of 68.4% which is better than tradi-
(7)
tional Deep Learning methods.
Detection of melanoma using DenseNet-based adaptive weighted loss function 345
References detection in crowded scenes. Multimedia
Tools and Applications, 1–24
[1] Soyer, H. P., Smolle, J., Kerl, H., and Stettner, [10] Singh, J. P.and Kumar, M. (2022). Chaotic
H. (1987). Early diagnosis of malignant mel- whale-atom search optimization-based deep
anoma by surface microscopy. The Lancet, stacked auto encoder for crowd behaviour
330(8562), 803. recognition. Journal of Experimental and
[2] Argenziano, G., Soyer, H. P., Chimenti, S., Theoretical Artificial Intelligence, 1–25
Talamini, R., Corona, R., Sera, F., and Kopf, [11] Agarwal, A., Gupta, B. K., Kumar, K., and
A. W. (2003). Dermoscopy of pigmented skin Agrawal, R. (2023, March). A Neural Net-
lesions: results of a consensus meeting via work based Concept to Improve Downscal-
the Internet. Journal of the American Acad- ing Accuracy of Coarse Resolution Satellite
emy of Dermatology, 48(5), 679–693.Han, Imagery for Parameter Extraction. In 2023
J., Colditz, G. A., and Hunter, D. J. (2006). 6th International Conference on Informa-
Risk factors for skin cancers: a nested case– tion Systems and Computer Networks
control study within the Nurses’ Health (ISCON) (pp. 1–5). IEEE.
Study. International journal of epidemiol- [12] Li, Y., and Ercisli, S. (2023). Explain-
ogy, 35(6), 1514–1521. able human-in-the-loop healthcare image
[3] P. Schmid-Saugeona, J. Guillodb, and J.-P. information quality assessment and selec-
Thirana, Towards a computer-aided diagno- tion. CAAI Transactions on Intelligence
sis system for pigmented skin lesions, Com- Technology.
puterized Med. Imag. Graph., vol. 27, no. 1, [13] Szegedy, C., Vanhoucke, V., Ioffe, S., Shlens,
pp. 6578, Jan. 2003. J., and Wojna, Z. (2016). Rethinking the
[4] A. I. of Health and Welfare, Cancer in Aus- inception architecture for computer vision.
tralia 2017, vol. Cancer series no. 101, Can- In Proceedings of the IEEE conference on
berra, ACT, Australia: Australian Institute of computer vision and pattern recognition
Welfare, 2017. (pp. 2818–2826).
[5] E. Silverberg, C. C. Boring, and T. S. Squires, [14] Wu, Z., Shen, C., and Van Den Hengel, A.
Cancer statistics, 1990, CA, Cancer J. Clini- (2019). Wider or deeper: Revisiting the
cians, vol. 40, pp. 9–26, 1990. resnet model for visual recognition. Pattern
[6] J. Han, G. A. Colditz, and D. J. Hunter, Risk Recognition, 90, 119–133
factors for skin cancers: A nested casecontrol [15] Zhou, T., Zhao, Y., and Wu, J. (2021, Janu-
study within the Nurses’ health study, Int. ary). Resnext and res2net structures for
J. Epidemiol, vol. 35, no. 6, pp. 15141521, speaker verification. In 2021 IEEE Spoken
Dec. 2006. Language Technology Workshop (SLT) (pp.
[7] G. Argenziano, G. Fabbrocini, P. Carli, V. De 301–307). IEEE.
Giorgi, E. Sammarco, and M. Delno, Epilu- [16] Yao, P., Shen, S., Xu, M., Liu, P., Zhang, F.,
minescence microscopy for the diagnosis of Xing, J.,and Xu, R. X. (2021). Single model
doubtful melanocytic skin lesions. Compari- deep learning on imbalanced small datasets
son of the ABCD rule of dermoscopy and for skin lesion classification. IEEE transac-
a new 7-point checklist based on pattern tions on medical imaging, 41(5), 1242–1254.
analysis, Arch Dermatol, vol. 134, no. 12, [17] Huang, G., Liu, Z., Van Der Maaten, L., and
pp. 701563, 1998. Weinberger, K. Q. (2017). Densely connected
[8] Agarwal, A., Kumar, S., and Singh, D. convolutional networks. In Proceedings of
(2021). An adaptive technique to detect and the IEEE conference on computer vision and
remove shadow from drone data. Journal of pattern recognition (pp. 4700–4708).
the Indian Society of Remote Sensing, 49(3), [18] Wu, Z., Shen, C., and Van Den Hengel, A.
491–498. (2019). Wider or deeper: Revisiting the
[9] Singh, J. P., and Kumar, M. (2023). Chrono- resnet model for visual recognition. Pattern
logical ant lion optimizer-based deep convo- Recognition, 90, 119–133.
lutional neural network for panic behavior
47 Advanced reconfigured solar arrays to
attenuate shading loss under partial
shading conditions: experimental
validation
Vijay Laxmi Mishraa, Yogesh K. Chauhanb, and Kripa S. Vermac
Electrical Engineering, Kamla Nehru Institute of Technology, Sultanpur, Uttar Pradesh, India
Abstract: The traditional total cross-tied (TCT) model and an advanced permutation combination reconfigura-
tion (PCR) are employed to analyze the performance of the solar arrays under realistic partial shading condi-
tions (PSCs). The model uses a single-diode solar cell using the MATLAB/SIMULINK tool. The performance
indicators are evaluated to show the high potency of advanced PCR over TCT under various PSCs. Results
show that the advanced PCR has minimized the shading loss (SL) by 9.28% and 76.78% under lamp post-shape
shading and corner shading (CS) against TCT. Also, advanced PCR has maximized the global power (GP) by
14.55% which has been verified by a hardware prototype under CS. These outstanding results of the advanced
PCR support the practicality of the suggested technique (PCR).
Keywords: Efficiency, global power, partial shading conditions, permutation combination reconfiguration, total
cross-tied
1. Introduction under dwarf-broad PSCs [3]. Modified cou-
ple matching circuitry reduces the shading
As compared to non-renewable sources
loss by 57.9% under varying irradiations as
of energy, the renewable source of energy
compared to TCT [4]. Prime-number-linked
plays a vital role in keeping the environ-
reconfiguration lowers the shading loss
ment clean and pollution-free. Solar energy
excels because of its wide availability, clean between 4% and 5% under the considered
nature, cost-effectiveness, and noiseless PSCs [5]. Permutation-combination-based
behavior [1]. Solar energy can be effectively reconfiguration (PCR) improves efficiency (η)
harnessed by installing solar panels. How- up to 12.57% under border shading as com-
ever, due to partial shading conditions (PSCs) pared to TCT [6]. The intelligent reconfigu-
like nearby buildings, passing clouds, air- ration model smoothens the output power
planes, bird droppings, etc., the solar arrays plot and maximizes the production around
fail to produce maximum power and lead 25.15% during inclined shading [7]. Under
to shading loss (SL) [2]. To overcome these inverted-U shading, the calcudoku reconfig-
problems, the reconfiguration of solar mod- ured solar model reduces the shading loss by
ules is performed. The conventional total 5.18% over TCT [6]. Compared to TCT, the
cross-tied (TCT) model is compared with the Kenken model improves the η by 13.16%
advanced odd-even (OE) model. OE causes under zig-zag shading [6]. The range-kutta-
maximum power extraction up to 31.01W inspired solar model upgrades the production
[email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-47
Advanced reconfigured solar arrays to attenuate shading loss under partial shading conditions 347
by 41.32% under shading cases over TCT
(Nassef et al., 2022). A distance-based recon-
figuration enhances the η between 12.36%
to 42.13% under all four shading cases
(Mishra et al., 2023). The cross-kit reconfigu-
ration (CKR) improved the performance by
10.12% to 26.12% over TCT under all con-
sidered shading cases (Faisal et al. 2023). The Figure 47.1. Electrical circuit of a solar cell.
reconfigured solar model increases the wire Source: AUTHOR'S SELF GENERATED
length. To study its effect on the performance
of solar arrays, authors introduced enhanced
skyscraper reconfiguration. It reduced 50%
wire length as compared to another consid- AQ1:
Please
ered reconfiguration under various consid- provide
ered PSCs (Mishra et al., 2023). As compared in-text
to TCT, the novel Ramanujan’s reconfigura- citation
for Figure
tion (NRR) improved η up to 3.16% under 23.1.
the third PSCs (Mishra et al., 2023). The nov-
AQ2:
elty of the proposed work as compared to the Figure 47.2. Solar module connections. Please
existing works is illustrated below: Source:
AUTHOR'S SELF GENERATED provide
source
for all the
• In this paper, various real-time shadings Figure 47.2. It is connected in TCT orien- figures.
are investigated to discuss the perfor- tation i.e., in series-parallel form (Figure
mance of solar arrays. 47.2 (a)). Since this structure involves many
• The conventional TCT is experimentally cross-connections, hence TCT is complex in
validated with an advanced reconfigured design [5]. The obtained TCT is swapped
PCR model. across each column to obtain an advanced
• Various performance parameters are stud- PCR model (Figure 47.2 (b)) [6]. In the pro- AQ3: As
ied to assess the behavior of TCT and posed work, experimental validation is done per style,
the refer-
advanced PCR model carefully. to show the authenticity of the TCT and ence list
advanced PCR. It comprises a 4 × 4 solar and its
citation
2. Solar Photovoltaic System and array arranged in TCT form. The ratings of should be
a single solar module are: open-circuit volt- num-
Considered Interconnection age (VOC) = 21.59 V, short-circuit current bered,
please
Schemes (ISC) = 0.63 A, peak voltage (VPE) = 17.30 V, amend
peak current (IPE) = 0.59 A, peak power (PPE) through-
out the
2.1. Modeling of Solar Cell = 10 W. Later the modules are swapped to file.
Eq. (1) is employed to design the solar cell test for PCR connection as shown in Figure
(Mishra et al., 2023). 47.3. Further, a voltmeter is used to read the
voltage data, an ammeter records the corre-
(1) sponding current values, a temperature sen-
sor detects the surrounding temperature, a
solar meter reads the irradiation level, and
2.2. Solar Array and Experimental
a rheostat is employed that acts as a load.
Setup A PSC is introduced by using thin sheets of
A symmetrical 4 × 4 solar array has various colors as per the amount of irradi-
been employed in this study as shown in ance on each shaded solar module.
348 Innovations in Computational Intelligence: Trends and Technologies
(3)
4. Results and Discussions
This study investigates traditional TCT and
Figure 47.3. Experimental setup. an advanced PCR under two shading cases.
Their effects are comprehensively studied as
Source:
AUTHOR'S SELF GENERATED discussed below:
3. Shading Patterns and Row 4.1. Output Curves under Shading
Current Formation Conditions
Figure 47.5 shows output curves under the
The proposed work incorporates two real-
considered shading patterns. Under uniform
istic shading conditions (a) lamp post-shape
shading (LPS) and (b) corner shading (CS) as
shown in Figure 47.4.
Figure 47.4. Shading conditions.
Source:
AUTHOR'S SELF GENERATED
The corresponding row-current generation
under CS for TCT and PCR are discussed as
shown by Eqs. (2) and (3) respectively. Simi-
lar row-current generation can be calculated
under LPS.
Figure 47.5. Output curves (a) TCT (b) PCR.
(2)
Source:
AUTHOR'S SELF GENERATED
Advanced reconfigured solar arrays to attenuate shading loss under partial shading conditions 349
conditions, both TCT and PCR generate 4.4 Execution Ratio (ER)
160 W. Under various shadings (LPS and CS),
It is calculated by Eq. (5) [5].
multiple power peaks are generated. This
leads to the reduction in global power gen- (5)
eration as depicted in Figure 47.6.
For TCT under LPS and CS, ER is 58.46%
4.2. Quantitative Results and 71.82% respectively. The advanced PCR
improves it to 61.99% and 84.05% under the
The quantitative results obtained for the pro- LPS and CS respectively. Figure 47.8 shows
posed work under various shading cases are the execution ratio plot for TCT and PCR
depicted in Table 47.1. under shading conditions.
4.3. Shading Loss (SL)
It is calculated by Eq. (4) [6].
(4)
The advanced PCR excels in reducing the
SL by 9.28% and 76.78% under both the
shading conditions respectively. Figure 47.7
shows a shading loss plot for TCT and PCR
under shading conditions.
Figure 47.7. Shading loss plot.
Source: AUTHOR'S SELF GENERATED
Figure 47.6. Global power plot. Figure 47.8. Execution ratio plot.
Source: AUTHOR'S SELF GENERATED Source: AUTHOR'S SELF GENERATED
Table 47.1. Quantitative Simulation Results
Shadings Topology GP (W) SL (%) ER (%) FF (%)
Uniform TCT, PCR 160 0 0 77.76
LPS TCT 94.26 41.53 58.46 65.21
AQ4: CS 115.8 28.18 71.82 55.87
Please LPS PCR 99.96 38 61.99 68.9
provide
source for CS 135.53 15.94 84.05 68.84
the Table.
Source:
AUTHOR'S SELF GENERATED
350 Innovations in Computational Intelligence: Trends and Technologies
4.5. Fill Factor (FF)
It is calculated by Eq. (6) (Yadav et al., 2020).
(6)
The advanced PCR improves the FF by
5.35% and 18.84% under both the shading
conditions respectively. Figure 47.9 shows
the FF plot for TCT and PCR under different
shadings.
Figure 47.9. Fill factor plot.
Source: AUTHOR'S SELF GENERATED
Figure 47.10. Experimental curves.
Source:
4.6. Proposed Study under Corner AUTHOR'S SELF GENERATED
Shading: Experimental
• The advanced PCR attenuates the SL by
Validation
9.28% and 76.78% under both the shad-
The proposed work is verified experimentally ing conditions respectively.
under CS. The advanced PCR attains maximum • The advanced PCR improves ER to
power at 135.53 W whereas the conventional 61.99% and 84.05% under the LPS and
TCT reduces it to 115.80 W. The experimental CS respectively.
output curve is depicted in Figure 47.10. • Reduced cross-connections and more shift
of the maximum power peak (GP) on the
output curves are the key reasons for the
5. Conclusion and Future Scope
excellent performance of the advanced
An extensive study has been performed to vali- PCR over TCT under the shading cases.
date the outstanding performance of advanced In the future, this work can be extended
PCR over conventional TCT. It is done under for large-size solar arrays by adding small 4
two realistic shading conditions. Four perfor- × 4 solar arrays. This will help in extracting
mance indices have been calculated. Following more power at the commercial level.
are the few key results of this study:
• As compared to conventional TCT, the
advanced PCR maximizes the GP by References
5.70% and 14.55% under considered [1] Nihanth, M. S. S., Ram, J. P., Pillai, D. S.,
shadings respectively. Ghias, A. M. Y. M., Garg, A., and Rajasekar,
Advanced reconfigured solar arrays to attenuate shading loss under partial shading conditions 351
N. (2019). Enhanced power production in Model Under Realistic Partial Shading Con-
PV arrays using a new skyscraper puzzle ditions. International Conference on Green
based one-time reconfiguration procedure Energy, Computing and Sustainable Tech-
under partial shade conditions (PSCs). Sol. nology (GECOST) 258–263.
Energy 194:209–224. [9] Mishra, V. L., Chauhan, Y. K., and Verma,
[2] Nasiruddin, I., Khatoon, S., Jalil, M. F., and K. S. (2022). A critical review on advanced
Bansal, R. C. (2019). Shade diffusion of reconfigured models and metaheuristics-
partial shaded PV array by using odd-even based MPPT to address complex shadings
structure. Sol. Energy 181:519–529. of the solar array. Energy Convers. Manag.
[3] Yadav, K., Kumar, B., and Swaroop, D. 269:116099–116135.
(2020). Mitigation of Mismatch Power [10] Nassef, A. M., Houssein, E. H., Helmy, B.
Losses of PV Array under Partial Shading E., Fathy, A., Alghaythi, M. L., and Rezk
Condition using novel Odd Even Configura- H. (2022). Optimal reconfiguration strat-
tion. Energy Reports 6:427–437. egy based on modified Runge Kutta opti-
[4] Sugumar, S., Winston, D. P., and Pravin, mizer to mitigate partial shading condition
M. (2021). A novel on-time partial shad- in photovoltaic systems. Energy Reports
ing detection technique for electrical recon- 8:7242–7262.
figuration in solar PV system. Sol. Energy [11] Mishra, V. L., Chauhan, Y. K., and Verma, K.
225:1009–1025. S. (2023). A new physics-based solar recon-
[5] Rezazadeh, S., Moradzadeh, A., Hashemza- figured model to enhance efficiency under
deh, S. M., Pourhossein, K., Ivatloo, B. M., partial shading conditions: Experimental fea-
and Hosseini, S. H. (2021). A novel prime sibility. Int. J. Green Energy 21(3):589–624.
numbers-based PV array reconfiguration [12] Faisal, M., Sharma, D., and Bansal, R. C.
solution to produce maximum energy under (2023). Cross kit reconfiguration algorithm
partial shade conditions. Sustain. Energy for enhanced output power of PV array dur-
Technol. Assessments 47:101498–101522. ing shading mismatch conditions. Optik
[6] Mishra, V. L., Chauhan, Y. K., and Verma, (Stuttg) 288:171218–171244.
K. S. (2022). A novel PV array reconfigu- [13] Mishra, V. L., Chauhan, Y. K., and Verma, K.
ration approach to mitigate non-uniform S. (2023). A comprehensive investigation of a
irradiation effect. Energy Convers. Manag. solar array with wire length under partial shad-
265:115728, 2022. ing conditions. Energy Sources, Part A Recover.
[7] Kannan, E., Avudaiappan, M., Kaliyaperu- Util. Environ. Eff. 45(4):10217–10241.
mal, S., Muthusamy, S., Panchal, H., Pan- [14] Mishra, V. L., Chauhan, Y. K., and Verma,
diyan, S., and Azhaganantham A. (2022). A K. S. (2023). A novel reconfiguration of
novel strategy for implementation of intelli- the solar array to enhance peak power and
gent techniques in solar photovoltaic arrays efficiency under partial shading conditions:
to improve the performance and various experimental validation. Clean Energy
comparison of partial shading mitigating 7(4):824-843.
techniques. Energy Sources, Part A Recover. [15] Mishra, V. L., Chauhan, Y. K., and Verma, K.
Util. Environ. Eff. 44(2):3079–3099. S. (2023). Various Modeling Approaches of
[8] Mishra, V. L., Chauhan, Y. K., and Verma, Photovoltaic Module: A Comparative Anal-
K. S. (2022). Peak Power Enhancement by ysis. Majlesi Journal of Electrical Engineer-
Novel Reconfiguration Scheme of a Solar ing 17(2):117–131.
48 Understanding the relationship of
quantum mechanics and machine
learning
Sushamana Sharma
Department of Physics, Jodhpur Institute of Engineering and Technology, Jodhpur, India
Abstract: The field of quantum mechanics is yet to be explored fully but still has great applications particularly
in quantum information processing. The need of alliance of quantum mechanics and machine learning (ML)
is discussed. The speed up achieved in quantum computation is the main reason of looking forward to use
quantum mechanics to reduce time complexity in learning process. Quantum superposition seems to be able
to reduce sample complexity of ML algorithm by reducing number of examples required to learn a task. The
hurdles in this path and the possible techniques to conquer these are examined at the understanding level. The
main aspects of quantum ML are discussed.
Keywords: Machine learning, quantum mechanics, quantum machine learning, supervised machine learning
1. Introduction that can be overcome by using quantum
computers.
The domain of computer science has
In 1900, Max Planck introduced the concept
increased extremely from simple simulation
of quantization to the world and successfully
of physical state to the analysis of complex
explained the problem of 1900 i.e. the black
data, i.e., from household use to astronomical
body radiation. Great physicists contributed
use, from entertainment to medical use. As
to the development of quantum theory nota-
dependence of human on computer increases,
more research is carried out to improve the bly are Albert Einstein, Heisenberg, deBroglie,
existing computational capacity as well as to E. Schrödinger, Dirac and many more. Once
provide more facilities for the betterment of Einstein said “if quantum mechanics were
human life. Internet of Things, big data, cyber correct then the world would be crazy”, it
security, machine learning (ML), artificial is very true as the world is going crazy over
intelligence, quantum information processing the incredible power the quantum mechanics
are the recent technological development in holds. The famous EPR paradox (Einstein et AQ1: As
per style,
this direction. The ML domain is a subset of al., 1935) questions the completeness of the the refer-
artificial intelligence whose literal meaning is quantum theory with the argument that the ence list
and its
“learning of a machine”. There are number of description of the physical reality provided citation
examples in our daily life where ML is used by quantum mechanics is not complete. Ein- should be
to make our life easier like current condition stein et al. imagined a pair of particles whose num-
bered,
of traffic by Google map, Google translation, quantum state is correlated with each other please
chat bots. In all these examples the data size irrespective of the distance between them. If amend
through-
is very large which cannot be processed effi- an observer making a measurement on one out the
ciently and speedily by classical computers particle to get a result, the quantum state of file.
[email protected]
DOI: 10.1201/9781003606635-48
Understanding the relationship of quantum mechanics and machine learning 353
the other particle collapses instantaneously. is given by |c0|2(|c1|2). The simplest qubit has
It could be explained in two ways: (1) the the form |ψ⟩ = 1 (|0⟩ + |1⟩) with probabil-
interaction between particles is faster than the √2
speed of light or (2) the correlation between ity of having outcome 0 or 1 is equal i.e. the
particles decides the final quantum state in probability of finding an electron in up spin
advance which implies the incompleteness of is 50% of the time likewise in down spin,
quantum mechanics. This correlation in quan- its chances are 50%. The density matrix is
tum states is known as quantum entanglement defined as ρ = |ψ⟩⟨ψ| having the value of trace
(Horodecki et al., 2009). Despite of the com- one and is semi-definite positive, helping in
plexities, quantum theory is undoubtedly the defining probability distribution which reveal
most successful quantitative theory of nature the nature of state whether classical or non-
that provides description of physical proper- classical. The negative values of probability
ties. Its applicability ranges from elementary distribution indicate the nonclassical behav-
particles like quarks to gigantic objects like ior of the state. There are three distribution
galaxies, from lasers and transistors to nuclear defined namely, Glauber–Sudarshan distribu-
magnetic resonators and quantum computers. tion, Wigner distribution and Husimi func-
The field of computation is also changing to tion. The Glauber–Sudarshan distribution
provide faster and secure processing and it is used to describe light in optical phase space
achievable with the help of quantum states like and has many applications in laser theory
squeezed states (Teich and Saleh, 1989), entan- and coherence theory. For nonclassical states
gled states. The use of quantum laws ensures like entangled state, the Wigner probability
security of the information in the channel distribution is non-positive and using Husimi
and very fast processing. Quantum informa- function, maximally nonclassical state can
tion processing (Nielsen and Chuang 2010) is be generated. Recent advances in the field
well-established concept nowadays. Less runt- of information processing and technology
ime and applicability of quantum laws hints use quantum laws to have faster and secure
the use of quantum computation in ML. An processing because of its capability to man-
attempt is made to explain the emerging field age big data size efficiently. The nonclassi-
of quantum ML to a general reader to encour- cal states such as squeezed states, entangled
age them. states are the gateway of not-so-far future
technology. It is possible to perform a large
number of operations in parallel in quantum
2. Quantum Mechanics and computer which makes them different from
Computation classical computer and this process of per-
In quantum mechanics, a physical state is rep- forming many tasks in parallel is known as
resented by a mathematical function which quantum parallelism. Two main constraints
by the superposition principle can be writ- on quantum information processing are: no
states, mathematically |ψ⟩ = ∑ncn|φn⟩ where
ten as the linear combination of all possible cloning theorem and not faster than light
communication. Since quantum computing
cn is the probability amplitude of the result obeys laws of quantum mechanics making it
be in state |φn⟩. The classical information is
of the measurement made on the system to impossible to know the current state of the
system in contrast with the case in classical
expressed in terms of 0 and 1 known as bit computers where the current state is known
while the quantum information is the super- to the observer. In quantum information pro-
position of the two bits 0 and 1 and known cessing, the qubit is considered as a single
as Qubit (Quantum Bit) given by |ψ⟩ = c0|0⟩ bit and the computation is nothing but the
+ c1|1⟩ with |c0|2 + |c1|2 = 1. The probability application of sequence of unitary matrices
of outcome of the measurement to be 0 (1) on the initialized state vector of the system
354 Innovations in Computational Intelligence: Trends and Technologies
whose result is obtained by the probability to increase the accuracy and to make predic-
distribution of the final density matrix. The tions on the basis of their experience. In ML,
logic gates can be considered as the applica- the data is fed to the system to train the algo-
tion of unitary matrices on small number of rithm to analyze the data, prepare appropri-
qubits and to record multiple qubits, regis- ate model to perform certain task and make
ters are arranged using tensor product which predictions. Based on the process of learning,
induces interference effect responsible for ML is categorized in four types mentioned
evaluation of a function on every point in the below.
domain simultaneously. This property makes
the quantum computers to perform some • Supervised ML in which the results of
tasks faster than classical system, sometimes past data analysis is used to study the
exponentially faster. For example, the Shor’s new data i.e. the algorithm trained the
integer factorization algorithm for quantum machine to generate a function based on
computers takes time 𝒪((log N)3) while the the conclusions derived for known input-
1 2 output pairs (labelled data) e.g. prediction
classical devices take time 𝒪((log N)3)(log N)3) of price of land or house (see Figure 48.1).
to factorize an integer N which outperforms For precise predictions, the ML models
the RSA encryption protocol (Montanaro used labelled data which is a data set hav-
2016). Here, the notation 𝒪(f(m)) is taken ing information of similar type data. For
from computer science to compare the per- example, a labelled data named cat has all
formance of algorithms. To characterize a the feature of a cat like its shape, height,
problem, “m” parameters are required then color, eyes, ears and others. Now someone
the upper bound on the asymptotic scaling of provides information of height, weight,
the algorithm will be the function f(m) i.e. it color of an animal to the system then ML
depends on the size of the input. will predict outcome using the labelled
The efficiency of quantum computer relies data and if the input maps with the label
on how efficiently a quantum algorithm can
run which requires large scale interference.
Interference is very fragile in nature making
the large scale computation almost impossi-
ble with precision therefore some techniques
are needed to convert impossible to possible.
There is a threshold in the physical error i.e.
the disturbance in interference below which
there is a possibility to correct the quantum
system and to make it work efficiently with-
out disturbing the interference using some
codes known as quantum error correction
(Ciliberto et al., 2018).
3. Machine Learning
The term ML was first used by Arthur Samuel
in 1959 during his research in the game of
checkers where the computer was not pro- AQ2:
Please
grammed explicitly to perform certain task. Figure 48.1. Showing labeled data for provide
In other words, it is the learning of a machine supervised ML. source for
just like the human learns by improving skills Source: AUTHOR'S SELF GENERATED
the figure.
Understanding the relationship of quantum mechanics and machine learning 355
cat, the best prediction will be a cat. similar data and then will separate the
Another example can be the estimation of images of dogs from cats. The examples
a price of a land in a given locality. The of association rule are stock market pre-
labelled data will have all the information dictions, suggestions after surfing particu-
like the price in the same locality in past lar website like “people who visit this site
years, the price of land in its neighbor- also visit”.
hood, size of land, location of land like a Flowchart of clustering method:
corner, future projects nearby and others.
When one feeds information for rate esti- Random raw data
mation, the ML will access that labelled ↓
data and make accurate prediction. formation of clustering of similar kind data
The flowchart of supervised ML is given ↓
above. ML processing using cluster
↓
Random raw data Accurate prediction
↓
Data labelling of similar kind data Flowchart of association rule:
↓
ML processing using input – output data set Random raw data
↓ ↓
Accurate prediction Data collection based on relationship
↓
• As we humans learn something from oth- ML processing
ers experience while something is learnt ↓
through our own experience. Just like Accurate prediction
human learning process, machine also
learnt from labelled data set as well as • In semi-supervised learning, algorithms
something from its own. How machine are fed with labelled and unlabelled data
learns from its own experience is the sec- i.e. it is a mixture of supervised and unsu-
ond type of ML i.e. Unsupervised learn- pervised ML. The machine learnt from
ing where the machine is trained without labelled as well as from unlabelled data
labelled data to make conclusions from to predict accurately just like supervised
the new data set. The machine itself cat- ML. it is preferable when very large data
egorizes data from the given data set is unlabelled and it is not desirable to label
based on the data type and data size. it. In simple words, the raw data is pro-
The grouping of data based on similari- cessed by unsupervised learning and then
ties is a method of unsupervised learning the intermediate prediction is processed
known as clustering while classification through labelled data set for most accu-
of data on the basis of some relationship rate prediction. It is mainly use to solve
is known as association rule. Third tech- regression and classification problems like
nique is the dimensionality reduction used text classification.
to reduce the data set size without com- • The reinforcement ML requires no cor-
promising with the integrity of the data rectness of input–output data set and it
set as much as possible. How can ML be is learning through surrounding, experi-
used to differentiate between the pictures ence and continuous improvement after
of dogs and cats? Here the machine will every observation e.g. driverless car. It is
collect the data and then make cluster of just like we humans learn i.e. collecting
356 Innovations in Computational Intelligence: Trends and Technologies
data by observation then making a deci- solution of a learning problem 𝒪(N3) is where
sion and then the feedback whether it was N is the number of examples in the training
a good decision. Here, the observation data. Using different types of regularization
means the available data, the prediction or optimization methods it can be improved
or decision based on it and then the feed- to 𝒪(N2). Which type of regularization strat-
back to know whether the action taken egy will be used, is selected by the sampling
was beneficial. to be performed. The training time reduces
significantly only for the dataset of moderate
Out of these four, the well-established and size but the modern dataset is quite big and
mature theory is the supervised ML which still increasing, the time is too demanding.
learns by putting the data set into mathe-
matical foundation where from a learner can
approximate an unknown function. There 4. Quantum Machine Learning
are mainly two different models for learn- The large data size increases computational
ing, probably approximately correct model complexity and computational time. These
proposed (Valiant 1984) and the statistical are the two main hurdles in the wide use
learning theory (Vapnik 1998). Valiant pro- of ML using classical computation. Paral-
posed a precise computational model of the lel processing due to quantum superposition
learning which consists of a mechanism for is the key feature for the speed up in quan-
the collection of information, learning pro- tum computation and if the training set can
cedure and assurance whether the learning be converted into superposed form, quan-
can be done in polynomial number of steps in tum computation may help in reducing the
simple words it gives correct complexity-the- runtime of ML algorithm. Now the question
oretic meaning of a concept class to be learn- arises, can we do this, if yes, how? For quan-
able. In this protocol, the learner must have tum computers, the dataset must be in quan-
the relevant data to exemplify the concept to tum form in other words dataset describing
be learned by the machine and the examples the state of quantum system and its evolution.
are distributed in a probabilistic way (Valiant To collect information, measurement is done
1984). The significance of the examples with on the system. In classical system, there is
the concept is tested by a routine Oracle. In no loss of information as independent meas-
the statistical learning model, the selection urements are done on the sub-system but in
of function is based on the actual data avail- quantum system due to non-local behavior
able. In any model, efficiency of the learner of particle, it is impractical to make a joint
represented by two parameters one is sample measurement on the quantum system having
complexity and the other is time complexity. N subsystems as well as it is not preferable to
To learn a function, the number of examples measure individual subsystem as it costs loss
required is a measure of sample complex- of data. There are two possible ways to get
ity that means the minimum the number of quantum data: one is to use the data in quan-
examples, the lower is the complexity and tum form i.e. data obtained by the process in
the minimum runtime of the algorithm gives which input argument and output of a quan-
measure of time complexity. The aim of ML tum function describe the quantum state of
is to create a statistical estimator which fits a system just like in quantum chemistry and
well with the new examples and assure good the other is to convert data stored in classi-
prediction performance. To improve the time cal form. The former option is preferable for
complexity, optimization techniques are quantum ML but it is not always possible to
used. In context of classical computation, the have information in quantum form. With the
time complexity of the process of finding the second choice, there must be some procedure
Understanding the relationship of quantum mechanics and machine learning 357
to get quantum states from classical informa- computation quadratically (Grover and
tion which cost the uncertainty in getting the Rudolph 2002). This is how time complexity
desired runtime in ML for classical data and can be dealt in ML using the laws of quantum
the computational expenses. The quantum mechanics. For sample complexity, there are
random access memory (QRAM) may trans- number of research papers on quantum prob-
form the classical information into quantum ably approximately correct model (Servedio
superposition. Just like RAM structure con- and Gortler 2004; Atici and Servedio 2005;
sisting of input register, memory array and Zhang 2010; Arunachalam and Wolf 2017).
output register, QRAM has the same archi- For instance, the quantum example for a con-
tecture for qubits that means it accesses mem- cept to be learned c:{0,1}n→{0,1} returns a
ory in quantum superposition. It is possible if (n+1) qubit state as per some uniform distri-
the input register i has addresses in superpo- bution D(x) provided (Bshouty and Jackson
sition state mathematically the input register 2008) i.e.
reads as ∑jψj|j⟩i and when processed through
QRAM, it gets superposed with the data in
data register d and returns what is known as
quantum data. There is a quantum algorithm for the
formation of clusters of data which is being
implemented in few-qubit Noisy Interme-
diate-Scale Quantum devices (Bermejo and
Here, Dj is the content stored at location Orus 2023). The idea is to use maximally
j in memory array. It is shown in reference orthogonal states in Hilbert space to address
(Giovannetti et al., 2008) that QRAM can large clusters with fewer qubits.
encode N d-dimensional vectors in ℝd into
log (Nd) qubits in time 𝒪(log (Nd)) requiring
the physical resources of the scale 𝒪(Nd). This 5. Conclusion
exponential scaling makes this idea uncertain It is difficult to conclude here as it is still grow-
in its physical realization as well as its compu- ing field in research by the quantum physics
tational benefits which are addressed by (Aar- and computer science community. It is a try
onson 2015, 292). Besides scaling issue, this to understand these two fields and their col-
concept also faces a problem with the active, laboration in a better way. There is no experi-
non-active and error-free number of compo- mental evidence of the exponential speed up
nents required (Arunachalam et al., 2015). in quantum computation and the encoding
Grover and Rudolph also proposed a model of classical information in quantum super-
based on efficiently integrable probability position efficiently. Having efficient quan-
distribution. According to this, a quantum tum hardware to support polynomial speed
superposition state can be generated having up will help in large scale ML but there is
vaguely all the information of classical prob- no single method to realize physically quan-
ability distribution. If there were an efficient tum computer in labs even though different
classical algorithm, the given probability dis- methods like ion trap, photon qubits, nuclear
tribution can be transformed into quantum magnetic resonance, optical lattice may be
superposed distribution or atleast in mixed used. The clubbing of quantum mechan-
state ρ = ∑ipi|ψ⟩⟨ψ| where pi is the probabil- ics and ML will definitely improve classical
ity distribution in classical form. Grover and ML in terms of sample complexity and time
Rudolph showed that the application of this complexity. These play significant role when
technique for the estimation of amplitude of the system is dealing with large size data. By
Fourier component is faster than in classical improving efficiency of QRAM large dataset
358 Innovations in Computational Intelligence: Trends and Technologies
can be handled and utilized for quantum [10] Aaronson, Scott. 2015. Read the fine print
ML. But it is evident that the quantum effects Nature Physics 11, (April):291–293. https://
make learning easy as compared to classical doi.org/10.1038/nphys3272.
case depending on the type of learning model [11] Arunachalam, S., Gheorghiu, V., Jochym-
we used. O’Connor, T., Mosca, M., and Srinivasan,
P. V. (2015). On the robustness of bucket
brigade quantum RAM. New journal of
Physics 17:123010(1–17).
References [12] Grover, Lov, and Rudolph, Terry 2002. Cre-
[1] Einstein, A., Podolsky, B., and Rosen, N. ating superpositions that correspond to effi-
(1935). Can quantum mechanical descrip- ciently integrable probability distributions
tion of physical reality be considered com- http://arxiv.org/abs/quant-ph/0208112.
plete?. Physical Review 47(10):777–780. [13] Servedio, Rocco. A., and Gortler, Steven
[2] Horodecki, R., Horodecki, P., Horodecki, J. (2004). Equivalences and separations
M., and Horodecki, K. (2009). Quantum between quantum and classical learnability
entanglement. Reviews of Modern Physics SIAM J. Comput. 33, no. 5:1067–1092.
81(2):865–942. [14] Atici, Alp, and Servedio, Rocco. A. 2005.
[3] Teich, M. C., and Saleh, B. E. 1989. Squeezed Improved Bounds on Quantum Learning
state of light Quantum Opt. 1, no. 2:153– Algorithms Quantum Inf. Process 4, no. 5
191. DOI 10.1088/0954-8998/1/2/006. (October):355–386.
[4] Nielsen, M. A., and I.L. Chuang. 2010. [15] Zhang, Chi 2010. An improved lower bound
Quantum Computation and Quantum Infor- on query complexity for quantum PAC
mation. 10th ed. Cambridge University Press. learning Inf. Process. Lett. 111, no. 1(Decem-
[5] Montanaro, Ashley. 2016. Quantum algo- ber):40–45. https://doi.org/10.1016/j.ipl.
rithms: an overview NPJ Quantum Infor- 2010.10.007.
mation 2, (January):15023. https://doi. [16] Arunachalam, Srinivasan, and de Wolf, R.
org/10.1038/npjqi.2015.23. 2017. Optimal quantum sample complexity
[6] Ciliberto, C., Herbster, M., Ialongo, A. D., of learning algorithms In Proc. 32nd Compu-
Pontil, M., Rocchetto, A., Severini, S., and tational Complexity Conference, edited by
Wossnig, L. (2018). Quantum machine Ryan O’Donnell, LIPIcs. Germany:Dagstuhl
learning: a classical perspective. Proc. R. Publishing.
Soc. A 474(2209): 20170551(1–26). http:// [17] Bshouty, Nader H., and Jackson, Jaf-
doi.org/10.1098/rspa.2017.0551. frey C. 2008. Learning DNF over the
[7] Valiant, L. G. 1984. A theory of the learn- Uniform Distribution using a Quantum
able Commun. ACM. 27, no. 11 (Novem- Example Oracle SIAM J. Comput. 28, no.
ber):1134–1142. https://doi.org/10.1145/ 3(January):1136–1153. doi.org/10.1137/
1968.1972. S0097539795293123.
[8] Vapnik, Vladimir N. 1998. Statistical learn- [18] Bermejo, Pablo, and Orus, Roman 2023.
ing theory. Vol. 1. New York NY:Wiley. Variational quantum and quantum-
[9] Giovannetti, V., Lloyd, S., and Maccone, L. inspired clustering Scientific Reports 13,
(2008). Quantum Random Access Memory. (August):13284. https://doi.org/10.1038/
Phys. Rev. Lett. 100(16): 160501. s41598-023-39771-6
49 Employability prediction: A machine
learning model for computer science
students
Abhishek Bhattacherjeea and Vinay Kukrejab
Department of Computer Science and Engineering, Chitkara University Institute of Engineering and
Technology, Punjab, India
Abstract: The current study targets to find a suitable machine learning (ML) model that helps in predicting
computer science students’ employability. Three different ML models have been implemented on the collected
data after consideration of six features. The data is collected from 210 students and the algorithms that have
been used are a Support vector machine, decision tree, and artificial neural network. Different performance
metrics have been used while implementing ML algorithms. The metrics that have been used are accuracy,
F1-score, precision, recall, and Matthew’s correlation coefficient (MCC). SVM has outperformed the other two
algorithms in the case of Accuracy, Recall, F1-score, and MCC. These values for SVM are 88.10%, 96.97%,
92.75% and 0.6159 respectively whereas for DT is 83.33%, 89.19%, 90.41% and 0.2701 and for ANN is
85.71%, 96.88%, 91.18% and 0.5705. It can be seen that SVM has outperformed the other two algorithms.
This study can be very beneficial to colleges and universities and will help them to put stress on only the features
that students are lacking for employability.
Keywords: Artificial neural network, classification, decision tree, employability, support vector machine
1. Introduction with appropriate training that increases their
competence level of skills that are required
The nation’s development, economy, and by employers. Now, institutes and universi-
growth primarily depend on education. ties have the responsibility to inculcate the
Education helps every country strengthen technical, soft, and personal skills, etc. in
its national competitive status by providing the students that make them employable.
suitable and qualitative manpower to the Before students can complete their gradua-
society. To impart quality education, univer- tion, institutes and universities must see the AQ1: As
per style,
sities, schools, and affiliated colleges play a skills requirements that students must have to the refer-
vital role. The higher education universi- work extensively before sitting for the inter- ence list
ties and institutes especially play a vital role views. Nowadays, education targets getting and its
citations
as they educate students with edge-based the students employment. should be
technologies that help the students in their In the past, many researchers have worked num-
bered,
employment. Nowadays, most of the work on student employability prediction mod- please
is automated which helps computer science els with the help of data mining techniques, check and
confirm
graduates to get hired quickly due to a vast machine learning (ML) models, and statisti- for
number of computer-related jobs (Casuat and cal models (Fok et al. 2018), (Bharambe et clarity.
Festijo 2019). Even employers spend a lot of al. 2017). These papers have tried to match
money on the candidates by providing them up the students’ skills with the companies’
[email protected]; [email protected]
a
DOI: 10.1201/9781003606635-49
360 Innovations in Computational Intelligence: Trends and Technologies
required skills. Researchers in the past have communication skills, and core employabil-
used KNN, random forest, SVM, decision tree ity skills. The researchers (Finch et al. 2013)
(DT), Naïve Bayes, etc. (Casuat and Festijo have tried to explore the factors that impact
2019), (Bharambe et al. 2017), (Nagaria and undergraduate employability. They have
Senthil Velan 2020) whereas some have tried identified 17 factors and further that have
to build their instrument scale for student categorized these 17 factors into five and
employability (Rothwell, Herbert, and Roth- they have also calculated the relative impor-
well 2008), (Behle 2020). The current study tance of these identified categories. The
has implemented different ML algorithms study aims to inculcate soft skills and prob-
after figuring out the skillset that is needed lem-solving skills in undergraduate students
by the students for a job opportunity. The that help in their employability. The authors
current study has also compared the results (Fok et al. 2018) have used the deep learning
of different ML algorithms with a detailed model to know the students’ development
discussion. performance. They have evaluated both aca-
demic as well as non-academic performance.
1.1. The layout of this paper They have used 25% of the collected data
for training and 75% for testing and have
This paper has described the related work in achieved an accuracy between 80 to 91 per-
Section 2. Section 3 has discussed the pro- cent. Moreover, they have tried to figure out
posed method that is followed by Section 4 the strengths and weaknesses of the students.
which focuses on materials and methods. The They have also identified the eligible com-
last section 5 mentioned the results and con- panies in which students can participate by
clusion of the study. matching their skill set. The authors (Nesrine
Mezhoudi, Rawan Alghamdi, Rim Aljunaid
2021) have conducted a detailed survey and
2. Related Work worked on the four objectives. Firstly, they
The literature shows that researchers have have identified the different data mining
worked on many features for the students’ applications that have been used in the past
employability prediction and many differ- for employability prediction. Secondly, the
ent algorithms have been implemented. The survey was conducted to identify the skills
authors (Casuat and Festijo 2019) have that students lack to be employable, thirdly
worked on features like soft skills and com- they have stressed the adjusting the curricu-
munication skills. They have used algo- lum according to their future employment,
rithms like DTs, support vector machines and lastly, they have tried to see the long
(SVMs), and random forests. The research- term demands of the market that helps in the
ers (Kenayathulla and Idris 2019) have con- students’ employment. An ensemble model
ducted a study in Malaysia and have tried (R. R. Maaliw, K. A. C. Quing, A. C. Lag-
to figure out the skills that are required for man, B. H. Ugalde 2022) was created by the
employability. They have tried to differen- researchers to predict the students’ employ-
tiate between the competence of skills and ability. They have used different data mining
the importance of skills that are required techniques for this. The model that has used
for employment. A model was developed Random Forest, SVM, and Naïve Bayes has
(Sinha, Ghosh, and Mishra 2020) in which achieved the highest accuracy of 93.33%.
employers’ expectations and employer per- They have tried to impose some policies that
ceptions are checked against the employers’ will provide help to students in their employ-
satisfaction level. Employers’ expectations ment. A hybrid model of deep belief network
are measured through technical skills, and softmax regression has been created by
Employability prediction: A machine learning model for computer science students 361
the authors (Bai, A., Hira 2021). Here, the
features are extracted using a deep belief net-
work, and classification whether employed
or not employed is done with the help of
softmax regression. The model has achieved
an accuracy of 98% which is higher than the
accuracy achieved by deep neural networks.
The researchers (Bhagavan, Thangakumar,
and Subramanian 2021) have worked on a
novel approach (HLVQ) to identify the stu-
dent’s academic performance along with the
students’ employability chances. They have
achieved an accuracy of 92.6% along with
a reduction in time complexity and space
complexity when compared with other ML
algorithms.
3. Proposed Methodology
The employability of computer science engi-
neering students is measured with the help of AQ2:
different ML algorithms and they are com- Please
pared with different metrics. Figure 49.1 Figure 49.1. Proposed methodology.
provide
source for
shows the methodology for predicting the the figure.
Source:
computer science students of engineering col-
leges and universities.
Figure 49.1 shows the current study 4. Materials and Methods
methodology, first of all, the dataset is col-
lected from passing-out students. A total 4.1. Data Collection
of 210 students’ data has been collected The data is collected from 225 students, out
and has six features. After the collection of of which 210 students’ data is in the correct
data, data is preprocessed and all the null form and completed in all aspects. The data
or vacant entries are filled with suitable val- mainly targets six features and they are men-
ues unless these entries are dropped from tioned in Table 49.1.
the collected data. The preprocessed data
is divided into two parts, the first part is Table 49.1. Students’ employability features
used for training of data, and the second
part targets testing of data. Different ML Feature Number Description
algorithms have been applied to the col- 1 Soft skills
lected preprocessed data. The algorithms 2 Problem-solving skills
that have been used for implementation 3 Technical skills
for predicting the students’ employability
4 Innovations skills
are SVM, DT, and artificial neural network AQ3:
(ANN). After getting the implementation 5 Personal Qualities Please
6 Perceptions towards provide
results of the above-mentioned algorithms, source
the results are compared based on accuracy, employability for all the
recall, and precision. Source: tables.
362 Innovations in Computational Intelligence: Trends and Technologies
The soft skills feature includes interper- of a hyperplane that distinguishes the data
sonal and communication skills. Problem- points in proper classes. It is very effective in
solving skills target the capability skills to the case of high-dimensional spaces and known
solve problems. Technical skills are associ- as a memory-efficient algorithm.
ated with the technical skills that they have
learned in their course. Innovation skills are 4.4.2. DT
related to the innovative ideas that students
This is one of the most effective algorithms
can generate while working on problems.
for prediction and classification. It follows
Personal qualities are the qualities that are
a tree-like structure, internal nodes mention
associated with their behavior. The important
attributes, branches mention output, and
feature is that what are the perceptions of the
leaf nodes tell the class labels. This algorithm
students toward employability.
requires less computation and helps to iden-
tify the fields that are essential for prediction
4.2. Data pre-processing as well as class classification.
First of all, data cleaning is done on the col-
lected data in the data pre-processing step. 4.4.3. ANN
The missed data is replaced with the mean This machine-learning algorithm is inspired
value and the noisy data which cannot be by the human brain and is used for pattern
filled with appropriate value is dropped. After recognition, classification, and clustering.
performing data cleaning, data integration is They are popularly known as computational
performed. It means the data is merged from models. They have a very high processing
different sources into one data store. Next, speed and are having very little response time.
data transformation is performed. All the They support serial processing.
data is normalized using normalized meth-
ods. And at the last, data reduction is used. It
helps in the dimension reduction of the data. 4.5. Performance parameters
The three algorithms have been implemented
4.3. Training and testing data to know the computer science engineering
students’ employability. The different perfor-
Now, the important step in the ML algorithm mance measures have been used and they are
is to divide the data into the training set and mentioned below (Casuat and Festijo 2019).
testing set. It is divided into a ratio of 80% Table 49.2 shows the results of the learning
and 20%. It means 168 students’ data is used algorithms.
for training and 42 students’ data is used for
testing. 4.5.1. Precision
This is one of the performance metrics used
4.4. Algorithms
by ML algorithms. It aims to know the model
In the current study, three algorithms have quality by using positive predictions. In other
been implemented. Their names are SVM, DT, words, it can be measured by dividing true
and ANN. These are described briefly as: positive predictions by total positive predic-
tions (i.e. true positive and false positive).
4.4.1. SVM The formula for the calculation of precision
is shown in Equation 1.
This is one of the supervised learning algorithms
that are well suited to classification and regres- Precision = True Positives / Positive
sion. This algorithm helps in the identification Predictions(1)
Employability prediction: A machine learning model for computer science students 363
A precision score of 1 means that every 4.5.5. Matthews Correlation
search result is relevant. Higher precision coefficient
denotes relevant results rather than irrelevant
results. This is one of the most reliable statistical met-
rics as compared with all other metrics men-
tioned above. The results are on the higher
4.5.2. Recall side of this metric if the results found in all
A recall is also known as Sensitivity. This four categories of the confusion matrix are
parameter targets quantifying the true positive on the better side. It varies proportionally
rate. In other words, it is measured by dividing with the size of negative as well as positive
true positive predictions by the total number of elements. It is measured by the formula men-
elements that are associated with positive class, tioned in Equation 5.
and the formula is mentioned in equation 2.
MCC = (True Positive * True Negative
Recall = True Positives / True Positives + – False Positive * False Negative) / (sqrt
False Negatives (2) (True Positive + False Positive) * (True
Positive + False Negative) * (True Negative
4.5.3. Accuracy + False Positive) * (True Negative + False
Negative))(5)
Business houses use ML algorithms for strate-
gic business decisions and the more accurate
learning algorithm helps in better strategic 5. Results and conclusion
business decisions that make the model accu- The education sector of every country is tak-
racy very important. It is measured by the ing innovative steps to make students employ-
percentage of the correct predictions. The able by enhancing their technical skills. The
formula for measuring accuracy is mentioned universities and colleges of respective coun-
in Equation 3. tries are also taking responsibility and tak-
ing steps to enhance students’ employability.
Accuracy = (True Positives + True Negatives) They are looking the features that are help-
/ (True Positives + False Positives + True ful in uplifting students’ skills. In the cur-
Negatives + False Negatives) (3) rent study, we have put stress on six features
for computer science graduates. A total of
4.5.4. F1-Score 210 students’ data is taken into considera-
This metric is used by taking the harmonic tion. The three popular ML algorithms have
mean of recall and precision into considera- been implemented and the results show that
tion. It is measured by the formula mentioned SVM has outperformed the other two and the
in Equation 4. accuracy achieved is 88.10% as compared
with 83.33 % of DT and 85.71% of ANN.
F1-Score = (2 * (Precision * Recall)) / When you compare the recall metric, the
({Precision + Recall) (4) results of SVM are the same. Recall values for
Table 49.2. Performance Metrics for ML Algorithms
Model \ Parameters Recall Precision Accuracy F1-Score Matthews Correlation Coefficient
SVM 0.9697 0.8889 0.8810 0.9275 0.6159
DT 0.8919 0.9167 0.8333 0.9041 0.2701
ANN 0.9688 0.8611 0.8571 0.9118 0.5705
364 Innovations in Computational Intelligence: Trends and Technologies
SVM, DT, and ANN are 96.97%, 89.19%, and Training 55 (7): 681–704. https://doi.
and 96.88%. One more rare metric has been org/10.1108/ET-07-2012-0077.
evaluated i.e. Matthew correlation coefficient [7] Fok, Wilton W.T., Y. S. He, H. H.Au Yeung,
and it will be better if it is above 0.5 or nearer K. Y. Law, Kh Cheung, Yy Ai, and P. Ho.
to 1. The SVM value for this metric is 0.6159 2018. Prediction Model for Students’ Future
Development by Deep Learning and Ten-
as compared with DT, ANN is 0.2701 and
sorflow Artificial Intelligence Engine. In 4th
0.5705 respectively. In other words, it can be
International Conference on Information
concluded that SVM has outperformed the Management, ICIM 2018, 103–6. https://doi.
two ML algorithms convincingly. org/10.1109/INFOMAN.2018.8392818.
[8] Kenayathulla, Husaina Banu, and Abdul
Rahman Idris. 2019. Gaps between Com-
References petence and Importance of Employability
[1] Bai, A., Hira, S. 2021. An Intelligent Hybrid Skills : Evidence from Malaysia. Higher
Deep Belief Network Model for Predicting Education Evaluation and Development
Students Employability. Soft Comput Ing 13 (2): 97–112. https://doi.org/10.1108/
25: 9241–9254. HEED-08-2019-0039.
[2] Behle, Heike. 2020. Students’ and Gradu- [9] Nagaria, Jumana, and S. Senthil Velan.
ates’ Employability. A Framework to Clas- 2020. Utilizing Exploratory Data Analysis
sify and Measure Employability Gain. Policy for the Prediction of Campus Placement for
Reviews in Higher Education 4 (1): 105–30. Educational Institutions. 2020 11th Inter-
https://doi.org/10.1080/23322969.2020.17 national Conference on Computing, Com-
12662. munication and Networking Technologies,
[3] Bhagavan, K. Subhash, J. Thangakumar, ICCCNT 2020. https://doi.org/10.1109/
and D. Venkata Subramanian. 2021. Predic- ICCCNT49239.2020.9225441.
tive Analysis of Student Academic Perfor- [10] Nesrine Mezhoudi, Rawan Alghamdi, Rim
mance and Employability Chances Using Aljunaid, Gomathi Krichna and Dilek
HLVQ Algorithm. Journal of Ambient Düştegör. 2021. Employability Prediction:
Intelligence and Humanized Computing A Survey of Current Approaches, Research
12 (3): 3789–97. https://doi.org/10.1007/ Challenges and Applications. Journal of
s12652-019-01674-8. Ambient Intelligence and Humanized
[4] Bharambe, Yogesh, Nikita More, Mani- Computing.
sha Mulchandani, Radha Shankarmani, [11] R. R. Maaliw, K. A. C. Quing, A. C. Lag-
and Sameer Ganesh Shinde. 2017. Assess- man, B. H. Ugalde, M. A. Ballera and M. A.
ing Employability of Students Using Data D. Ligayo. 2022. Employability Prediction
Mining Techniques. In International Con- of Engineering Graduates Using Ensemble
ference on Advances in Computing, Com- Classification Modeling. In 2022 IEEE 12th
munications and Informatics, ICACCI Annual Computing and Communication
2017, 2110–14. https://doi.org/10.1109/ Workshop and Conference (CCWC).
ICACCI.2017.8126157. [12] Rothwell, Andrew, Ian Herbert, and Frances
[5] Casuat, Cherry D, and Enrique D Festijo. Rothwell. 2008. Self-Perceived Employabil-
2019. Predicting Students ’ Employability ity: Construction and Initial Validation of
Using Machine Learning Approach. In 6th a Scale for University Students. Journal of
IEEE International Conference on Engineer- Vocational Behavior 73 (1): 1–12. https://
ing Technologies and Applied Sciences (ICE- doi.org/10.1016/j.jvb.2007.12.001.
TAS) Predicting, 1–5. Newyork. [13] Sinha, Saitab, Piyali Ghosh, and Ashutosh
[6] Finch, David J., Leah K. Hamilton, Riley Mishra. 2020. Employability of Fresh Engi-
Baldwin, and Mark Zehner. 2013. An neering Graduates in India A Fresh Look
Exploratory Study of Factors Affecting Applying Expectation. Education and Train-
Undergraduate Employability. Education ing 62 (1): 47–63. https://doi.org/10.1108/
ET-12-2018-0265.
50 A comprehensive review of innovations
in stray cattle detection
Abdullah Nizamia, Rudrendra Bahadur Singhb, Haris Javedc,
Mohd Anas Nadeemd, and Alok Kumare
Computer Science and Engineering, Babu Banarasi Das Institute of Technology and Management,
Lucknow, India
Abstract: Management of stray cattle population and control is an issue that has received extensive attention in
the past few years due to its effects on the health of the public, their safety, and the welfare of cattle as a whole.
This paper intends to provide an extensive review of the current breakthroughs in the perception of stray ani-
mal technologies and the employed methodologies used to achieve the same. It shall cover an expansive range
of techniques, including computer-vision based systems, approaches using sensor mechanisms and IoT, as well
as those making use of machine learning algorithms. Moreover, the paper discusses the challenges and future
scope of this field, outlining the requirement for interdisciplinary teamwork and innovative solutions to address-
ing this issue with a worldwide impact.
Keywords: stray cattle detection, computer vision, agricultural security
1. Introduction problem. Moreover, farmers also run the risk
of getting injured if they try to remove large
Indian farmers face many challenges while cattle from their fields. It is common for farm-
securing their crops and produce from stray cat- ers to face aggressive cattle that pose a threat
tle, such as cows, goats, and elephants can cause to their safety as well as that of their families.
major damage and can prove to be a threat to Thus, the need for an efficient and reliable
the safety of farmers and their kins. Present solu- solution to detect stray cattle on farmlands and
tions to this problem are either labour-extensive alert farmers in real-time is crucial for ensuring
or too costly for the majority of the farmers to their safety and preventing crop damage.
put into practice. As such, many farmers are left Rising climate change forces us to make
with no choice but to spend long hours protect- the most of what we have. Erratic weather
ing their farmlands, which leaves them with lit- patterns by themselves make it extremely dif-
tle time for other important tasks, hampering ficult for farmers to get a good yield. India
their overall productivity. scores a low 28.7 in the Global Hunger Index
Furthermore, these stray cattle have a deci- 2023, with the condition labelled as “serious”
mating impact on the crops of the farmers. and a rank of 111 out of 125 [1]. This is a
Losses in yield due to animal damage results clear indicator that we need to maximise our
in a significant financial loss for these farm- yield and increase it further if we wish to feed
ers, especially those operating on a small scale. all our populace. With this as our context,
Absence of dependable and cost-effective solu- we believe that the first step towards achiev-
tions for perception and cautioning farmers to ing this would be to prevent any unnecessary
the presence of stray cattle only exaggerates this wastage of crops in the first place.
a
[email protected]; [email protected]; [email protected]; dmeanasnadeem@
gmail.com; [email protected]
DOI: 10.1201/9781003606635-50
366 Innovations in Computational Intelligence: Trends and Technologies
In this literature review, we shall discuss We performed a two-stage screening pro-
the past work done around this particular cess. In the first stage, titles and abstracts of
problem and seek to improve upon those if papers were assessed for relevance. In the sec-
there is any scope. ond stage, full-text papers were reviewed to
determine if they met the inclusion criteria. A
consensus approach was used for any discrep-
2. Methodology ancies, and a third reviewer was consulted if
We used a systematic approach to find and needed. Given the exploratory nature of the
assess the literature on stray cattle detection literature, quality assessment and risk of bias
in this study, concentrating on topics such as analysis were not applied, as the focus was on
agricultural security and vision transformers. comprehensively reviewing available research
This section outlines the steps and processes rather than evaluating the methodological
involved in the literary review. quality of individual studies.
The goal of our search approach was to Information was extracted from the
thoroughly examine all of the literature that selected papers, including key concepts, find-
was out there on the use of stray cattle detec- ings, models and their architecture used,
tion methods in agricultural security. Aca- datasets employed, and any noteworthy dis-
demic resources such as IEEE Xplore, ACM cussions or implications related to the stray
Digital Library, PubMed, Google Scholar, cattle detection. Data extraction was carried
and Scopus were searched. The searches out using a structured format to ensure con-
were conducted between January 2017 and sistency and to capture relevant details for
November 2023 in order to include the most further analysis.
recent studies in this field. Relevant terms and The information thus obtained after per-
phrases like “Stray cattle Detection,” “Agri- forming data extraction using the above men-
cultural Security,” “IoT in cattle Detection,” tioned method on the screened papers was
“Use of State-of-the-Art technology in cat- used to perform our literary review compre-
tle Detection,” and similar variations were hensively. The findings we obtained after per-
included in the search queries. Filters were forming this survey lead us to a conclusion
used to refine search results, and references regarding which method is the best for stray
in relevant articles were cross-referenced to cattle detection in terms of being economic,
ensure completeness. feasible, reliable and secure. Further findings
Papers included in this review had to meet are mentioned in the conclusion section of
the following criteria: this paper in detail.
This methodology facilitated the system-
• They focused on the application of state-
atic identification of relevant literature and
of-the-art technology in agricultural
the synthesis of findings to provide a compre-
security.
hensive overview of the use of state-of-the-
• They were peer-reviewed articles, confer-
art technology in agricultural security, with
ence papers, or reports from reputable
a specific focus on the stray cattle detection.
sources.
The structured approach ensured transpar-
• They were written in English.
ency and reliability in the review process.
• They were published within the specified
time frame, that is, from January 2017 to
November 2023. 3. Critical Analysis of Previous
Papers not meeting any one of the above Work
provided criteria were excluded from our The stray cattle issue is among the leading
review. problems confronting Indian farmers, who
A comprehensive review of innovations in stray cattle detection 367
shoulder large burdens of protecting their Yuhang Wang et al. (2023), Wang et al.
crops. It results in money lost, safety con- (2023) urgently address the problem of
cerns and reduced farm productivity. Next overgrazing significantly impacting on eco-
research will try to address these problems by systems. They focus on the downsides of
using computer vision and object detection grazing and overbreeding for livestock, a fac-
techniques. tor that largely causes prairie degradation,
Shripad Bhatlawande et al. (2015) (2023), as well as environmental harm. The research
In this paper we use computer vision to help underscores that the more technologically
people with visual impairments stay safe. advanced and efficient tools available today,
Despite the fact that there are many blind/ like satellite and telco network remote sens-
visually impaired who can travel alone, there ing technologies and AI models are much
is a real safety issue because of their fear of better options when it comes to collecting
encountering stray dogs. First, the present data of the livestock from traditional manual
generation of blind assistive technology white methods [5].
cane cannot operate from a distance and thus Knyva et al. (2023), Knyva et al. The epi-
is not capable of preventing interactions with sodes of a (2023) highlight the importance
wandering animals or detecting barriers [2]. in researchers, studying issues such as wild-
Shuzhi Su et al. (2023), In this paper, we life behaviour, habitat, and the interface with
are dealing with the problem of item rec- human-populated areas, monitoring and
ognition in the state-of-the-art on [ref_src] detecting wild animals to detect whether
complex background using object identifi- their study animals are encountering residen-
cation approaches. To model interactions tial or road-based traffic during migration.
among different objects in complex scenes, We present an exciting new paradigm where
the researchers recapitulate a dual-attention the workload is performed by an Internet-of-
vision transformer network by combining Things (IoT) sensor network that operates on
spatial window attention and channel group harvested solar energy and a corresponding
attention. They also introduce an adaptive detection methodology. Nodes in the sensor
path aggregation network, which combines network are low-power embedded systems
the feature maps and suppresses the back- that rely on passive infrared sensors, a long-
ground information effectively with Convo- range (LoRa) module and solar panels for
lutional Block Attention Module (CBAM). power provision. An experimental IoT sen-
[3] Zhaokun Li et al. (2023), Our method sor network, consisting of eight nodes placed
provides comparably strong performance in at 50 meters distance on a road with a gate-
mobile object detection, and is designed spe- way connected for node data aggregation and
cifically for detecting aerial objects from UAV thermo-vision camera installed for cross-vali-
camera images. Images of this type are hard to dation, was established [6].
process because objects in those images have K. Chitra et al. (2023), Chitra et al. 205
multiple scales and need real-time detection. 206 (2023) … for wildlife-parasitic crop such
To deal with these problems, the research- as: elephants, wild pigs and monkeys are the
ers suggest a lightweight model: DSYolov3 major causes of frequent or severe damage to
advance from Yolov3 model. These improve- crops, leading to significant losses resulting
ments consist in a multi-scale aware decision in large scale over time commercially culti-
discrimination network, an enhanced multi- vated plants. Dealing with this specific ani-
scale fusion-based channel attention model, mal safeguard is necessary for farms which
and sparsity based channel pruning. Experi- cannot continuously look out or surround all
mental evaluations show effectiveness of the those plants to stop from financial damage.
model and its practicality on UAVs [4]. To solve this problem, this study suggests a
368 Innovations in Computational Intelligence: Trends and Technologies
system through which alliefs are secured in efficient intrusion detection in the agricul-
the agricultural field and farmers can be noti- tural field [10].
fied via call alert if animals are found. This Trial 10—Tan M et al. (2022), Camera
system uses an Arduino Uno, Ultrasonic sen- traps are widely used in wildlife surveys and
sor for X-axis and Y-axis which are connected biodiversity monitoring, collectively gen-
to send the signals to a GSM module (for erating millions of images or videos as the
call notifications) as well as generate alarm result of their trigger mechanisms. A few
sounds [7]. studies have suggested the use of deep learn-
Sreedevi and Anitha (2022), This article ing approaches in order to automate wild-
Address the gap of vehicular-animal incidents life detection from camera trap images thus
as a One health emerging threat was selected decreasing manual effort and accelerating the
as > “Topic track” < between humans and analysis. Despite this, there exist few studies
wildlife. Development of an Ideal System for where numerous models have been validated
Detection of Wild Animals Utilizing Deep and compared with field monitoring settings.
Learning Algorithms [8]. The Unknown Region Wildlife Image Dataset
Wangzhi Xing et al. (2022), The decreas- was collected at Northeast Tiger and Leopard
ing population of Koala Bears in Australia National Park (NTLNP dataset) as the initial
at Capacities, calls urgently for conservation study. The research then tested three popular
measures. A camera network was established object detection architectures, reporting its
to view Koalas and their behavior in both wild performance when being trained only with
habitats and in controlled settings, creating a day or night data versus a combined dataset.
unique dataset for Koala video tracking. With The selected models involve YOLOv5 series
this data set, a two-stream convolutional neu- (one-stage with anchors), Cascade R-CNN
ral network (CNN) model was developed by with HRNet32 (two-stage with anchors),
researchers. The paper describes a new cost- and FCOS using ResNet50 and ResNet101
effective model for identifying and tracking (one-stage without anchor). The results from
Koala events in videos by combining prior the experiments showed a good performance
semantic object detection information with of the object detection model in one single
optical flow methods. [9] participants (P. K. case, training on both day and night datasets
Panda et al, 2022), The problem of wild ani- together. On animal image detection ability
mals causing incursion is a real paradox, due and animal video classification accuracy, the
to the considerable amount of resources lost average mAP is: 0.98 and accuracy rate is
and also to the human lives jeopardized. To 88%, respectively. The one-stage YOLOv5m
address this issue, this study aims to use the model showed the best recognition perfor-
Internet of Things (IoT) in order to monitor mance with the highest model complexity.
regions continuously and also detect wild ani- The AI technology with which this software
mal intrusions in agricultural fields. Initially, is integrated enables ecologists to extract
it consists of Ultrasonic sensors at the corners information from a large amount of imagery
which senses an intrusion and a picture from in an expedited manner and can save consid-
Node MCU Micro controller mounted cam- erable time [11].
era on E-vehicle that does Field surveillance. Anurag Tiwari, et al. Published this Study
We use the IoT to notify the farmer via an 11 (2021), Key to this is the use of precision
IOT application as soon as detect a threat/ livestock farming techniques, which enable
obstacle. The most important criterion for quick and accurate counting of individual
evaluating this system was the quality of animals. In dairy farming, cattle counting
captured intruder images and prompt notifi- and monitoring are uber crutial so the wel-
cation alerts. The proposed model performs fare of live stocks may not be compromised.
A comprehensive review of innovations in stray cattle detection 369
In order to achieve this, a computer vision detection. This study unveils a plethora of
framework has been suggested. Which uses observations that are consistent with transfer
ResNetV2 for feature extraction paired learning while adapting the models trained on
with the YOLOv4 optimizer making detec- standardized datasets for deployment in real-
tions fast and accurate. It uses IoT sensors istic settings. In the study, researchers uncover
to upload the output that has been collected, a unique insight that might illustrate some
into a centralized repository. The sensors flag of the difficulties when it comes to practical
any deep rooted discrepancy or incongruity, application—translating detectors developed
like disease, pest infestations, dehydration using images of animals in their natural envi-
etc. they are immediately communicated to ronments for use inside non-natural settings.
the farmer concerned. This makes it an effec- This approach is validated by empirical evi-
tive, fast and accurate cattle count for farm- dence that suggests the limited capabilities of
ers by the proposed framework guiding the these detectors to adapt with ease. This issue
government side in intervening through live- is tackled in the research with a way for semi-
stock tracking [12], Gabriel S. Ferrante et al. automated synthetic data generation specifi-
(2021), The findings of this work are based cally designed for domain-specific training.
on a comprehensive review done between Additionally, the study highlights that the
January 2017 and May 2021. Its main goal code and data used in these experiments are
was to present an extensive review on state- made available to promote more pursuits in
of-the-art animal detection and classification this area [14].
via computer vision technologies in urban The four studies [5,32,41,42] that involved
scenes. Second, the study sought to identify Zhaoming Wang et al. were within ethnicity-
some existing research areas that need more based selection frequencies and excluded
attention. Through an automated search from the meta-analysis (Table 9) Full size
method across two online databases, 146 table (2020), The presence of stray animals
potential papers were identified which met in urban areas implies great inconvenience to
the inclusion criteria, and 20 studies that met the lives of city dwellers, social issues, sani-
the selection criteria were examined in detail tation problems, risks to living beings and
for data extraction. The 20 studies were general urban management. This study aims
divided into six categories: (i) Support Vector to solve the problem by introducing a low-
Machines (SVM), (ii) Histogram of Oriented cost animal detection system based on infra-
Gradients—HOG, (iii) Scale-Invariant Fea- red thermal imaging technology. The system
ture Transform—SIFT, (iv) Principal Com- design is based on infrared thermal imaging
ponent Analysis—PCA, or Convolutional technology, using Atmega328 microcontroller
Neural Networks—CNN and Deep Feature- as a central controller. In its operation it has
based Decision Fusion Learning DFDL. The the infrared array sensor to determine exter-
analysis underlined the widespread use of nal infrared radiation and via a built-in ADC
CNN as well as combinations to improve converts the voltage fluctuation that occurs
classification model accuracy [13]. as thermal radiation into digital signals.
Singh et al. (2020), The recognition of AMG8833 sensor that is sending the cap-
intruding animals in human-populated areas tured temperature information and ESP8266
makes it necessary for safety and road safety. through SPI bus, creating pseudocolor images.
The motivation behind this research is to On top of the that, it manages a color LCD
solve the problem using different deep learn- so you can see infrared through them like in
ing algorithms related to other parts of com- spy movies. Our single semirmall and cost-
puter vision—object detection, segmentation, effective system is supposed to be character-
tracking and finally a little bit exotic in edge ized by small size, low energy consumption,
370 Innovations in Computational Intelligence: Trends and Technologies
wireless communication capabilities (remote computer-aided DWA algorithms which have
control) and mobility. It can also be anywhere been partially developed by our group to
small animals are having to be searched and thermal infrared remote sensing images. The
rescued in an urban environment, monitoring performance of DWA in the thermal imaging
animal environments inside—the walls of the domain is demonstrated through a series of
buildings and a lot more [15]. analyses. The analyses considered the time
Kalaivanan Sugumar et al. (2019), Facing of capture with thermal imagery—somewhat
the danger of extinction for wildlife in the in favor with predawn, overcast days to be
context of disrupting natural ecosystem, and at the best moment of time. The method
a high risk tax to the animals that cross high- we proposed can aid effective extraction of
ways leaded by nature, this study intends moving wild animals in thermal remote sens-
to verify if there is a technological solution ing images, based on the DWA algorithm.
based on IoT systems for instances in which We then tested the accuracy of this method
collisions between vehicles and wild beasts by using it in an extensive number of air-
happens in reserve areas. With the inherent borne thermal images. Results demonstrated
danger of electric fencing to animal life, this an average producer’s accuracy of 77.3%,
paper seeks to solve that by means of wire- user’s accuracy of 29.3% over all the images,
less alerting via smartphones and huge LED which is very significant, meaning that our
displays located in high-traffic areas where method could greatly reduce human efforts
animals cross over. The proposed system in surveying moving wild animals among
consists of the motion detection, sensory, and a large number of thermal remote sensing
the object recognition algorithm that uses images Furthermore, in a comparative study
artificial neural networks (ANN). The struc- with that of an expert visual check-up, 24
ture involves the detection of animal move- moving objects had new revealed which
ment around roads to sense motion using were incapable to be captured before-post
Passive Infrared (PIR) sensors, and ANN capturing using our method. Note that we
for object recognition. The object recogni- can improve detection success by optimizing
tion algorithm decides whether the motion observation conditions for moving wild ani-
indicates life of animals or other factors. mals, such as particular moments and mete-
If it detects the animal, then through LED orological factors. Such discussions on the
signage boards and an Android application need for observing conditions are valuable
using Google Maps showing warnings just to those studying thermal remote sensing
in that spot through MQTT alerts dispatch. imagery to detect changes in animal popula-
This research achieves an object identifica- tions [17].
tion accuracy up to 91% [16]. Based on Prior Research Insights The
Oishi Yu et al. (2018), Mapping trends of research papers cited here are highly useful
wild animal populations is essential to con- in improving our “Stray Cattle Detection”
serving and managing those species. Statisti- project. We use the state-of-the-art object
cal methods have been used to estimate these detection techniques (e.g., transformer
populations but we still require some new based models and lightweight solutions on
solutions to resolve the applicability over mobile platforms) to enhance the ability to
large areas Our study has developed sup- detect stray animals against visual noises.
port systems for the automatized detection We will also leverage the experience gained
of wild animals in remote sensing imagery in protecting wildlife and security for visu-
to overcome this limitation. In this study, we ally impaired individuals to make our solu-
employed the algorithm called the Detec- tion truly protect both farmers and their
tion of Moving Wild Animals (DWA) with fields.
A comprehensive review of innovations in stray cattle detection 371
4. Conclusion will help extend the solution’s impact to a
broader audience. Additionally, integrating
In conclusion, detection of stray cattle rep- with Google technologies will enhance data
resents a significant step forward in address- analysis and decision-making capabilities for
ing the pressing issue of crop protection for farmers.
Indian farmers. The challenges posed by stray Overall, the stray animal detection repre-
animals, including cows, goats, and elephants, sents a promising solution to a critical prob-
have long plagued the agricultural commu- lem, offering hope to farmers in India who
nity, leading to substantial financial losses have long struggled with crop protection.
and safety concerns. Its multidimensional approach, integrating
Stray animal detection offers a compelling cutting-edge technology with practical agri-
solution to the longstanding challenge of pro- cultural knowledge, has the potential to trans-
tecting crops and property from stray cattle, form the way farmers manage their crops and
a prevalent issue faced by farmers in India ensure a more sustainable and secure future
and other regions. The project’s approach for Indian agriculture.
leverages cutting-edge technology, includ-
ing machine learning and computer vision,
to provide real-time detection and alerts for References
farmers, ensuring the safety of their crops,
[1] India—Global Hunger Index (GHI)—peer-
property, and themselves.
reviewed annual publication designed to com-
The paper underscores the existing research prehensively measure and track hunger at the
gap in this domain, emphasising the unique- global, regional, and country levels <https://
ness and pioneering nature of stray cattle www.globalhungerindex.org/india.html>
detection. It combines advanced technology [2] “Expert system for detection of stray ani-
with localised agronomic knowledge, open- mals for safety of visually impaired peo-
ing up new avenues for research and develop- ple” by Shripad Bhatlawande, Anjali Deore,
ment in the field of agricultural technology. Pranav Chandode, Anushka Chavan,
The proposed work aligns with sustain- Swati Shilaskar, and Jyoti Madake (Inter-
able development goals, contributing to national conference on applied computa-
sustainability, poverty reduction, and food tional intelligence and analytics, 2023), doi:
security. By referencing relevant studies in 10.1063/5.0133591
the literature, the project draws support from [3] Su, Shuzhi and Chen, Runbin and Fang,
Xianjin and Zhang, Tian. (2023). A Novel
existing research and underlines its poten-
Transformer-Based Adaptive Object Detec-
tial for groundbreaking advancements at tion Method. Electronics. 12. 478. doi:
the intersection of artificial intelligence and 10.3390/electronics12030478.
agriculture. [4] Zhaokun Li, Xueliang Liu, Ye Zhao, Bo
Liu, Zhen Huang, Richang Hong. (2023). A
lightweight multi-scale aggregated model for
5. REcommendations for Future detecting aerial images captured by UAVs.
Work doi: 10.1016/j.jvcir.2023.103058.
[5] Yuhang Wang, Lingling Ma, Qi Wang,
The future prospects for the stray animal
Ning Wang, Dongliang Wang, Xinhong
detection are promising, with a clear road- Wang, Qingchuan Zheng, Xiaoxin Hou
map for expansion and improvement. The and Guangzhou Ouyang. (2023). A Light-
incorporation of machine learning-based weight and High-Accuracy Deep Learning
embedded cameras, language localization, Method for Grassland Grazing Livestock
cross-platform accessibility, and collabora- Detection Using UAV Imagery. doi: 10.3390/
tion with government agencies and NGOs rs15061593.
372 Innovations in Computational Intelligence: Trends and Technologies
[6] Knyva, Mindaugas, Darius Gailius, Gin- [12] A. Tiwari, K. Sachdeva and N. Jain, “Com-
tautas Balčiūnas, Darius Pratašius, Pranas puter Vision and Deep Learningbased
Kuzas, and Aistė Kukanauskaitė. 2023. “IoT Framework for Cattle Monitoring,” 2021
Sensor Network for Wild-Animal Detection IEEE 8th Uttar Pradesh Section Interna-
near Roads” Sensors 23, no. 21: 8929. doi: tional Conference on Electrical, Electronics
10.3390/s23218929 and Computer Engineering (UPCON), Deh-
[7] K. Chitra et al., “Animals Detection Sys- radun, India, 2021, pp. 1–6, doi: 10.1109/
tem In The Farm Area Using Iot,” 2023 UPCON52273.2021.9667617.
International Conference on Computer [13] G. S. Ferrante, F. M. Rodrigues, F. R. H.
Communication and Informatics (ICCCI), Andrade, R. Goularte and R. I. Meneguette,
Coimbatore, India, 2023, pp. 1–6, doi: “Understanding the state of the Art in Animal
10.1109/ICCCI56745.2023.10128557. detection and classification using computer
[8] Sreedevi K L, Anitha Edison (2022). Wild vision technologies,” 2021 IEEE Interna-
Animal Detection using Deep learning. doi: tional Conference on Big Data (Big Data),
10.1109/INDICON56171.2022.1003979[6] Orlando, FL, USA, 2021, pp. 3056-3065,
Wangzhi Xing; Jun Zhou; Wee Lum Tan; doi: 10.1109/BigData52589.2021.9672049.
Fereshteh Nayyeri; Douglas Kerlin; Guy Cast- [14] A. Singh, M. Pietrasik, G. Natha, N.
ley. (2022). Dual-stream Convolutional Neural Ghouaiel, K. Brizel and N. Ray, “Animal
Networks for Koala Detection and Tracking. Detection in Man-made Environments,”
doi: 10.1109/DICTA56598.2022.10034583. 2020 IEEE Winter Conference on Applica-
[9] Wangzhi Xing; Jun Zhou; Wee Lum tions of Computer Vision (WACV), Snow-
Tan; Fereshteh Nayyeri; Douglas Kerlin; mass, CO, USA, 2020, pp. 1427–1438, doi:
Guy Castley. (2022). Dual-stream Con- 10.1109/WACV45572.2020.9093504.
volutional Neural Networks for Koala [15] Wang, Zhaoming and Liu, Xiaomin.
Detection and Tracking. doi: 10.1109/ (2020). Design of Animal Detector Based
DICTA56598.2022.10034583. on Thermal Imaging Sensor. Journal of
[10] P. K. Panda, C. S. Kumar, B. S. Vivek, M. Physics: Conference Series. 1550. 042066.
Balachandra and S. K. Dargar, “Implementa- 10.1088/1742-6596/1550/4/042066.
tion of a Wild Animal Intrusion Detection [16] Sugumar, Kalaivanan and Kumar, Amrutham.
Model Based on Internet of Things,” 2022 (2019). IoT Concept for Animal Detection
Second International Conference on Artifi- Using ANN to Prevent Animal Vehicle Col-
cial Intelligence and Smart Energy (ICAIS), lision on Highways. International Journal
Coimbatore, India, 2022, pp. 1256–1261, of Science and Engineering Applications. 8.
doi: 10.1109/ICAIS53314.2022.9742948. 492–497. doi: 10.7753/IJSEA0811.1004.
[11] Tan M, Chao W, Cheng JK, Zhou M, Ma Y, [17] Oishi, Yu, Hiroyuki Oguma, Ayako Tamura,
Jiang X, Ge J, Yu L, Feng L. Animal Detec- Ryosuke Nakamura, and Tsuneo Matsu-
tion and Classification from Camera Trap naga. 2018. “Animal Detection Using Ther-
Images Using Different Mainstream Object mal Images and Its Required Observation
Detection Architectures. Animals (Basel). Conditions” Remote Sensing 10, no. 7:
2022 Aug 4; 12(15):1976. doi: 10.3390/ 1050. doi: 10.3390/rs10071050.
ani12151976.
51 Enhancing emotional insight: NLP
powered sentiment analysis
Jyoti Gaura, Stuti Singhb, and Shelja Sharmac
Department of Computer Science and Engineering, Sharda School of Engineering and Technology,
Sharda University Greater Noida, UP, India
Abstract: Sentiment analysis is a branch of natural language processing (NLP) that is essential to comprehend-
ing customer preferences, public opinion, and the emotional nuances found in textual data. This study explores
the complex field of sentiment analysis, using sophisticated NLP methods to identify and analyze sentiments
in a variety of textual datasets. This research paper offers a comprehensive grasp of sentiment analysis using
NLP, which will be beneficial to scholars, practitioners, and policymakers. This work opens the door for the
creation of more precise, trustworthy, and ethical sentiment analysis systems by tackling technical difficulties,
moral dilemmas, and domain-specific nuances. It also makes a substantial contribution to the field of NLP and
its applications in comprehending human feelings and viewpoints.
Keywords: Algorithms, artificial intelligence, machine learning, natural language processing, sentimental
analysis
1. Introduction is a method for figuring out customers view
point related to a product. Other customers
The sentiment analysis uses speech analysis, will find the analysis useful in making the right
voice posturing, and conversation observation product choice. The role of a recommendation
to identify feelings and emotions, especially system is to suggest products to other custom-
those related to a service, legacy, or career. ers and show them while they shop based on
Examination of feelings ascertains, whether the review analysis. Sentiment analysis has gar-
they are neutral, negative, or positive, and to nered a lot of interest recently, much like text
what degree. Recent analytical technologies categorization based on customer assessments
available in the market are able to handle [7]. Structured information can be extracted
enormous volumes of client complaints in a from unstructured data using sentiment anal-
precise and dependable way [3]. To ascertain ysis. Content generated by the user on social
feelings and emotions, especially those associ- media different platforms often uses colloquial
ated with a subject, sentiment analysis listens language and expresses strong feelings [8].
to talks, assesses speech and voice posturing
through a service, a legacy, or a profession [4].
The natural language processing (NLP) 2. Natural Language Processing
can be used, for example, to understand the
point of view, opinion, belief, or attitude that is 2.1. What is natural language
upheld, contested, or expressed in comments,
reviews, articles, blogs, and other online plat-
processing (NLP)
forms. This process is called opinion mining Sentiment analysis is performed by NLP
or sentimental analysis [6]. Sentiment analysis on text-based, phrase- or document-level
[email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-51
374 Innovations in Computational Intelligence: Trends and Technologies
corpora [1]. An unstructured data comprises a
variety of data formats, including text, audio,
and video, it is exceedingly challenging to
derive conclusions from this type of data. An
essential step in performing sentiment analy-
sis is pre-processing the data. There are a few
preparatory measures that will be done before
any sentimental analysis techniques are used,
such as stemming, eliminating stop words
from the texts, etc [2]. Building machines
that can understand text or speech input and
react to it in a manner to that of people is
AQ1:
the aim of NLP [4]. NLP is more particularly Please
an engineering-based methodology and com- provide
putational statistics [6]. NLP is utilized for source
Figure 51.1. Role of NLP. for all the
the analysis of customer or employee reviews figures.
to determine their attitude. It is possible to Source:
AUTHOR'S SELF GENERATED
analyze sentiment using supervised learning
or lexicon approaches. The term “unsuper-
vised learning” is another name for the dia- 2.3. Different domains where
lectic approach. Three categories—leadership sentimental analysis is used
strength, perspective, and recognition—are
created by applying the Lexicon approach to The Sentimental analysis is used in the educa-
the reviews in this article [10]. tion system. Academics offer input to discuss
the distinctions between the particular teach-
ing strategy that is currently losing traction
2.2. Role of natural language
in classrooms and the types of teaching tac-
processing tics that pupils reject [1]. In present scenario
There are ensemble methods and semantics in sentiment analysis can be utilized in multiple
feature vectors, that can improve sentiment fields, as shown in Figure 51.2 including eco-
classification by utilizing NLP [3]. To ascer- nomics, politics, sociology, and others [2].
tain if the data in the IMDB review data- Since opinions are among the main factors
bases is neutral, positive, or negative, NLP influencing behavior, people’s opinions have
techniques are employed in the emotional a significant impact on their doings. Actually,
analysis process. This analysis can be used companies and organizations are constantly
by users to decide whether or not a film is searching for customers comments, regarding
worth watching [4]. Sentiment analysis can their goods and services.
be performed by using NLP to process differ- The customers take part in social contact
ent tweets, speech, or text from sources. The through a range of internet social media plat-
three general categories in which the emo- forms, including platforms like, Twitter and
tions are labeled are ““Negative”, positive” Facebook [3]. This analysis can be used by
and “neutral” [5]. One can create a variety users to decide if a film is worthwhile to see.
of NLP tools and resources with considera- These days, people usually rely on the evalu-
tion for the Sentiment Analysis Taxonomy. ations and ratings left by previous buyers
The necessary modifications to the current of movies or other products [4]. Google’s
resources and tools for implementing the tax- machine translation team learned how to
onomy have been discussed in brief [9]. Dif- represent text in essays using self attention
ferent roles of NLP are shown in Figure 51.1. mechanism and further its application was
Enhancing emotional insight: NLP powered sentiment analysis 375
carried out in additional NLP tasks [6]. Dif- clients [4]. In contrast to clustering, statisti-
ferent domains of sentimental analysis are cal, and hybrid approaches, the result of NLP
shown in Figure 51.2. based feature selection is a technique that is
more frequently employed [5]. One popular
2.4. Importance and approach type of data mining that assists in enhancing
those goods or services using sentiment anal-
Text sentiment analysis has become increas- ysis. Many studies on sentiment analysis have
ingly important as people all over the world
deal with more challenging circumstances in
recent years. Approaches to machine learn-
ing may involve supervised or unsupervised
learning [1]. Organizations can use opinion
analysis to learn what people think of the
newest product, and governments can use it
to determine whether the public is comfort-
able with the new approach. Sentiment analy-
sis is a popular tool used by IT companies to
learn about the opinions of their employees
in order to create corporate strategies and
to modify those strategies going forward
[2]. Artificial Intelligence evaluates whether
something is good or bad by combining a
great deal of math with NLP knowledge [3].
By assigning a positive or negative rat-
ing to reviews, Businesses can obtain knowl-
edge about how to enhance their goods and Figure 51.2. Domains of sentimental analysis.
more effectively satisfy the demands of their Source: AUTHOR'S SELF GENERATED
Table 51.1. Dataset popularly used [2]
S. No. Name Detail
1 Amazon Product This collection of dataset includes multiple document, texts and
Detail [2] evaluations.
2 Stanford Sentiment The HTML field of the Rotten Tomatoes are consider in these
Treebank [2] kind of dataset.
3 Multi-Domain Sentiment These kind of dataset includes both the positive and the negative
Dataset [2] fields for thousands of different Amazon product listings.
4 Sentiment140 [2] It capture the Feelings of a Brand Or Product
5 Twitter US Airline This includes the traveler’s tweets from US Airlines
Sentiment [2]
6 Paper Review Dataset [2] It includes the summary of the paper that was presented during
the conference.
7 Sentiment Lexicons The dataset includes 81 distinct language lexicons.
AQ2:
For 81 Languages [2] Please
8 Lexicoder Sentiment Among these the Bag-Of-Word Dictionary, which is intended to provide
source for
Dictionary [2] automatically code sentiment for news a legislative the table.
Source:
AUTHOR'S SELF GENERATED
376 Innovations in Computational Intelligence: Trends and Technologies
been conducted in an effort to uncover ways than the noun’s sentiment detection. Panikar
to enhance data analysis in order to obtain et al. [9] recommended applying these cat-
more important business information and egories to sentiment analysis across a num-
better service [8]. ber of industries. There are some suggestion
to enhance the current NLP pipeline opera-
tions to execute granular sentiment analy-
3. Literature Review sis quickly and effectively. Patrick et al. [10]
The R et al. [1] proposed a post tagging discussed, sentiment analysis can be utilized
approach to analyze scholars text feedback in client feedback analysis and can also be
using NLP to conclude teaching work exten- applied to employee review systems.
sion, sentimental analysis focuses on the
student’s feedback collected through online 4. Methodology
mode, where NLP used approach is used to
predict the polarity of students comments. The methods and strategies used to carry out
Sharma et al. [2] presented the comparison of sentiment analysis within the parameters of
several techniques and strategies for extract- this study are described in the methodology
ing sentiments from unstructured data is section. It provides a comprehensive explana-
presented in this research. Brindha et al. [3] tion of how to gather data, extract features,
carried study to tackles the issue of sentiment choose a model, and set up the experiment.
categorization on the Twitter data set. In this Different methodology used are as follows:
study, the Twitter data set is used to investi-
gate the sentiment categorization problem. • Grammar Employ, a strategy that is based
Reddy A et al. [4] applied sentiment analy- on the dictionary. Feeling and polarity
sis to the Internet Movie Database (IMDB), (positive, negative, or neutral) words are
reviewed Data set to ascertain whether the included in the lexicon [2].
details provided by a user indicate a positive • Naive Bayes methodology [4].
or negative review. The correctness of the sug- • Machine learning based approach [1].
gested approach is evaluated using real-world
data. Hardt et al. [5] used social media data 4.1. Data collection
to analyze customer preferences and attitude The 2848 comments from our institute’s edu-
on tourism before and during the COVID- cational website made up the dataset used in
19. Authors used NLP techniques and used this study. The dataset was manually labelled
Reddit posts for data sets. Jadeja et al. [6] with sentiment polarity labels. Through sur-
intended to minimize the time and expense vey, data has been gathered in real time from
factors while making the most use of this students.
sentiment scrutiny. Large corporations utilize
SA extensively, but small businesses are still a
long way off from it, because it is either too 4.2. Data preprocessing
expensive or too time-consuming. A and co. The data pretreatment stage is crucial for
et al. [7] proposed a sentiment analysis tech- both machine learning and data analysis.
nique for the e-commerce system. Their work It comprises organizing, transforming, and
integrates machine learning (NLP) into a cleansing raw data, to prepare it for modeling
web application provided by online shops to or analysis.
determine users’ interest in the products and The quality and effectiveness of the
services. Chaki et al. [8] performed numerous machine learning models or data analysis can
well-researched studies and the resolution of be significantly impacted by appropriate data
significant sentiment analysis problems other preparation.
Enhancing emotional insight: NLP powered sentiment analysis 377
4.3. PoS tagging 5. Discussion
A POS tagger identifies each phrase in a sen- The NLP is a widely used method for assess-
tence or document and assigns its part of ing sentiment in IMDB review data. The
speech. For POS tagging, the Stanford POS fundamental steps involved in analyzing the
tagger are used. After segmenting the text sentiment of IMDB movie reviews by NLP
into sentences, this tagger assigns a POS tag are feature extraction, data preparation, and
to each word. model training. Data preparation include
cleaning the raw text data and eliminating
4.4. Feature extraction any unnecessary information, such as stop
words and punctuation, as well as converting
During the feature extraction procedure,
the text into a numerical representation, such
remarks from students are taken out of each
as a bag-of-words or word embedding [4].
line. It is necessary to first understand the
To offer features for feature extraction that
emotion score, how it is applied, and how it
appropriately represents the emotion stated
relates to each phrase that is close by before
in the text, NLP techniques are utilized, such
determining the polarity of a written text.
as sentiment lexicons, sentiment polarity
Next are certain attributes that affect the
scores, and sentiment classification models.
document’s comparison.
The authors in [2] generated views on
health using three machine learning clas-
4.5. Feature reduction sifiers: Naïve Bayes, Machine Support and
In sentimental analysis, handling multidimen- Logistic Regression. The methods employed
sional text data poses a significant challenge were Bernoulli Naive Bayes and Multinomial
because it can reduce the classifier’s efficacy. Naive Bayes. Through the Stochastic Gradi-
For instance, thus, a technique assist Sort the ent Descent, Vector Classification, Vector
content of a phrase into positive and negative Linear Classification Support, and Vector
categories. Classification Support Services, SVM was
Many sophisticated text processing tech- made available. Twenty- six thousand filtered
niques, including lemmatization, tokeniza- tweets were used in the study. The most suc-
tion, and stemming, are applied. [3]. These cessful classification methods were found to
are further described as follows: be stochastic gradient descent and support
vector machines (SVMs) with linear carrier
• Tokenization classification. The statistics have shown that
Tokenization, the initial stage of NLP, the accuracy ranged between 85% and 91%.
converts sentences into understandable
data segments that a computer can utilise.
Tokenization may quickly become a cha-
6. Conclusion
otic phone game in the NLP process if it is This review of the literature considered earlier
not used to establish a strong foundation. theories, techniques, and findings put forth by
• Stemming different scholars. This covers the study which
It is the process of eliminating suffixes by was conducted in the fields of customer evalu-
applying a straightforward rule-based tech- ations of goods, films, products, and opinions
nique, such as “ing,” “ly,” “s,” and others. evaluation, ratings for popular subjects, and
• Lemmatization stock market analyses, and so forth. It is clear
The process of combining several inflected that sentiment analysis is important in many
versions of a word is known as lemma- areas, including marketing, conversational
tization. NLP, computational linguistics, AI, E-commerce, Governance, and Health-
and chatbots all use it. care. Thus, it is necessary to create complex
378 Innovations in Computational Intelligence: Trends and Technologies
models for sentiment analysis by adjusting data. Journal of Computational and Theo-
routine NLP pipeline tasks. retical Nanoscience, 17(9), 4535–4542.
Machine learning and Lexicon-based https://doi.org/10.1166/jctn.2020.9300
approaches were used to accomplish senti- [6] Jadeja, H., Thacker, C., and Parmar, K.
ment analysis. There is a lot of unstructured (2022). Categorizing data for sentimen-
tal analysis by auto-focusing mechanism
data present online on social media platforms,
using natural language processing. 2022
data pre- processing is essential for creating a
Second International Conference on Com-
learning model to do sentiment analysis. puter Science, Engineering and Applica-
Several preprocessing approaches were used tions (ICCSEA). https://doi.org/10.1109/
on the dataset, including stemming, stop word iccsea54677.2022.9936363
removal, square bracket removal, HTML [7] A, A., M, B., R, M., K, V., and S R, K. K.
stripping, and removal of special characters. (2022). Sentimental analysis for E-commerce
website. 2022 10th International Conference
on Emerging Trends in Engineering and Tech-
Acknowledgement nology—Signal and Information Processing
We express the gratitude to Dr. Ramneet for (ICETET-SIP-22). https://doi.org/10.1109/
icetet-sip-2254415.2022.9791606
her guidance and thankful, for providing the
[8] Chaki, P. K., Hossain, I., Chanda, P. R., and
necessary information regarding the paper. Anirban, S. (2017). An aspect of sentiment
analysis: Sentimental noun with dual senti-
mental words analysis. 2017 International
References Conference on Current Trends in Computer,
[1] R, N., S, P. M., Harithas, Pramath. P., and Electrical, Electronics and Communica-
Hegde, V. (2022). Sentimental analysis on tion (CTCEEC). https://doi.org/10.1109/
student feedback using NLP and Pos Tag- ctceec.2017.8455159.
ging. 2022 International Conference on [9] Panikar, R., Bhavsar, R., and Pawar, B. V.
Edge Computing and Applications (ICE- (2022). Sentiment analysis: A cognitive
CAA). https://doi.org/10.1109/icecaa55415. perspective. 2022 8th International Con-
2022.9936569 ference on Advanced Computing and Com-
[2] Sharma, R., and Pathak, P. (2021). A com- munication Systems (ICACCS). https://doi.
parative study on various approaches of org/10.1109/icaccs54159.2022.9785027
sentimental analysis. 2021 International [10] Patrick, H. A., J, P. G., Sharief, M. H., and
Conference on Simulation, Automation and Mukherjee, U. (2023). Sentiment Analy-
Smart Manufacturing (SASM). https://doi. sis Perspective using supervised machine
org/10.1109/sasm51857.2021.9841194 learning method. 2023 Fifth Interna-
[3] Brindha, K., Senthilkumar, S., Singh, A. K., tional Conference on Electrical, Com-
and Sharma, P. M. (2022). Sentiment analy- puter and Communication Technologies
sis with NLP on Twitter data. 2022 Inter- (ICECCT). https://doi.org/10.1109/
national Conference on Smart Generation icecct56650.2023.10179807.
Computing, Communication and Networking [11] Kaur, R., Uppal, M. and Gupta, D., 2023,
(SMART GENCON). https://doi.org/10.1109/ May. A Comprehensive and Comparative
smartgencon56628.2022.10084036 Study of Handwriting Recognition System.
[4] A, B. R., Reddy, B. V. S., Nandam, A. D., Nar- In 2023 IEEE Renewable Energy and Sus-
esh, K., Depuruu, S., and Sakthivel, M. (2023). tainable E-Mobility Conference (RESEM)
Sentimental analysis of movie reviews using (pp. 1–6). IEEE.
NLP techniques. 2023 5th International [12] Sharma, S., Mittal, V., Srivastava, R. and
Conference on Inventive Research in Com- Singh, S.K., 2020, December. Empirical
puting Applications (ICIRCA). https://doi. evaluation of various classification methods.
org/10.1109/icirca57980.2023.10220783 In 2020 2nd International Conference on
[5] Ramneet, Gupta, D., and Madhukar, M. Advances in Computing, Communication
(2020). Analysis of machine learning Control and Networking (ICACCCN) (pp.
approaches for sentiment analysis of Twitter 105–109). IEEE.
52 Decision-making using argumentation
mining framework on English text
Rudrendra Bahadur Singha, Shobhit Sinhab, Ankita Singhc,
Alok Kumar Thakurd, Akhil Chaurasiyae, Nishtha Mauryaf, and
Arpita Yadavg
Department of Computer Science and Engineering,
Babu Banarasi Das Institute of Technology and Management, Lucknow, India
Abstract: This paper investigates the application of argumentation mining (AM) to enhance decision-making
processes in the English language. Leveraging recent strides in natural language processing (NLP), our approach
achieves precise argument extraction.
The review comprehensively examines the current landscape of AM, highlighting its potential applications.
Additionally, a real-world case study is presented, illustrating the practical implementation of AM in decision-
support scenarios. This study contributes to the field by addressing the need for effective decision-support
tools through the lens of AM. The methodology capitalizes on advancements in NLP to extract arguments
accurately, paving the way for informed decision-making. The paper’s significance lies in its practical implica-
tions, shedding light on the tangible benefits of incorporating AM techniques into decision support systems. In
summary, this paper offers a concise and informative overview of the application of AM in English language
decision-making, showcasing its potential impact and practical utility in real-world scenarios.
Keywords: Argumentation mining, decision-making process, natural language processing (NLP), decision-sup-
port scenarios, real-world case study potential applications
1. Introduction This review paper delves into the burgeon-
ing field of AM, examining its potential for
In the ever-evolving landscape of informa- supporting decision-making processes in the
tion access and decision-making, the ability English language domain. We begin by pro-
to effectively process and analyze language is viding a comprehensive overview of AM,
paramount. Argumentation mining (AM), a highlighting its key tasks and challenges [1].
burgeoning area within natural language pro- We then explore the various applications of
cessing (NLP), offers a unique lens through AM in decision-making contexts, showcas-
which to understand the logic and reasoning ing its impact in areas such as legal reasoning
underpinning human discourse. By automati- [2], policy analysis [3], and business intelli-
cally identifying, classifying, and evaluating gence. Moreover, we delve into the techni-
arguments within textual data, AM holds cal advancements driving this field forward,
immense potential to revolutionize diverse including machine learning models, linguistic
domains, from legal proceedings and politi- features, and argumentation schemes [4, 5].
cal debates to customer reviews and online Our review critically examines the cur-
discussions. rent state of the art in AM, identifying both
a
[email protected]; [email protected]; [email protected]; dalok01895@gmail.
com; [email protected]; [email protected]; [email protected]
DOI: 10.1201/9781003606635-52
380 Innovations in Computational Intelligence: Trends and Technologies
its strengths and limitations. We highlight the
significant progress made in extracting and
classifying arguments [6], while acknowl-
edging the ongoing challenges in evaluating
argument strength and identifying fallacies.
Additionally, we explore the ethical consid-
erations surrounding AM, emphasizing the
importance of transparency and account-
ability in its application [7].By comprehen-
sively examining the landscape of AM and its
implications for decision making in the Eng-
lish language domain, this review paper aims
to provide a valuable resource for research-
ers, practitioners, and policymakers alike. We
anticipate that this review will contribute to
the continued advancement of this field and
pave the way for its wider adoption in vari-
ous decision-making contexts.
2. Methodology
AM has emerged as a powerful tool for
AQ1:
extracting and analyzing arguments from Figure
natural language text, offering valuable 52.1
insights into the decision-making process. citation is
missing in
This methodology outlines a comprehen- the text,
sive data flow diagram (DFD) approach to please
provide
AM, utilizing various techniques to extract, the
analyze, and evaluate arguments from tex- appro-
tual data, ultimately supporting informed priate
citation.
decision-making.
AQ2:
Please
2.1. Data acquisition and provide
source for
preprocessing Figure 52.1. Data flow architecture. the figure.
Source:
The DFD for AM begins with the founda- AUTHOR'S SELF GENERATED
tional step of data acquisition and preprocess-
ing. Raw textual data, sourced from various compatibility with subsequent analysis. This
outlets such as legal documents, policy pro- involves cleaning by removing irrelevant
posals, customer reviews, and news articles, information like punctuation and format-
forms the basis for AM. Depending on the ting inconsistencies, tokenization to split the
context, data collection may involve online text into individual tokens, normalization
databases, proprietary repositories, or web through lowercasing and lemmatization,
scraping techniques [8]. part-of-speech tagging, and named entity
Once acquired, the data undergoes a recognition (NER) to identify entities like
rigorous Preprocessing stage to ensure people and organizations [9].
Decision-making using argumentation mining framework on English text 381
2.2. Argument identification and enhances the transparency and informa-
classification tiveness of decisions. Through thorough
analysis and evaluation of arguments,
Moving forward in the DFD, the next stage decision-
makers gain deeper insights into
is argument identification and classification. issues, identify biases, and make more
This involves resolving coreferences, identify- informed choices. This systematic DFD
ing main claims, supporting premises, implicit approach ensures a comprehensive and
warrants, and recognizing argumentation structured application of AM throughout
schemes. Techniques encompass rule-based the decision-making journey.
systems, machine learning models (BERT,
RoBERTa, CNNs), and hybrid approaches,
ensuring a comprehensive identification and 3. Related Works
categorization of arguments (Feng et al., AM has emerged as a vibrant field within
2023). NLP, aiming to automatically extract and
analyze the structure and semantics of argu-
ments in natural language texts. This capa-
2.3. Argument evaluation bility has significant potential for various
The DFD progresses to argument evalua- applications, including decision-making,
tion, where the strength and persuasiveness where understanding the logic and rationale
of arguments are assessed. This includes log- behind different viewpoints is crucial.
ical reasoning, evidence analysis, sentiment
analysis, and discourse analysis. Logical 3.1. Early developments and
reasoning employs defeasible logic or other foundations
formal systems, evidence analysis assesses
quality and credibility, sentiment analysis The early work in AM focused on develop-
gauges emotional tone, and discourse analy- ing theoretical frameworks and computa-
sis examines linguistic features and rhetori- tional models for identifying and classifying
cal strategies [7]. arguments. Researchers like [2] explored
argument schemes, which formalize com-
mon patterns of reasoning used in argu-
2.4. Decision-making support mentation. This work laid the groundwork
for machine learning approaches that could
The final stage in the DFD is decision-making automatically recognize these patterns in
support, where the extracted and evaluated text.
arguments are integrated into the decision- In the paper [8] proposed a framework
making process. Argument summarization for AM that encompassed various subtasks,
generates concise summaries, utilizing extrac- including claim identification, stance classi-
tive or abstractive techniques [8]. Argument fication, and argument component identifi-
visualization creates representations such as cation. This framework served as a valuable
argument graphs or trees for enhanced under- roadmap for subsequent research in the
standing [10]. Argument-based decision sup- field.
port systems (ABDSS) contribute by assisting
in structuring complex arguments, facilitating 3.2. Advancements in argument
comparison, and providing recommendations
for decision-making [5].
extraction
In conclusion, the integration of AM With the rise of machine learning techniques,
techniques into decision making processes particularly deep learning, significant progress
382 Innovations in Computational Intelligence: Trends and Technologies
has been made in automated argument iden- to support legal research and decision-
tification and classification. Early approaches making [5].
relied on handcrafted features and rule-based • Policy analysis: Identifying and analyzing
systems, but recent advancements utilize pre- arguments in policy documents to inform
trained language models (PLMs) like BERT policy formulation and evaluation [7].
and RoBERTa, achieving state-of-the-art per- • Business intelligence: Extracting insights
formance [9, 10]. and competitor analysis from customer
These models can effectively capture com- reviews and social media data to support
plex linguistic features and relationships, business decisions [10].
leading to improved accuracy in identifying
claims, premises, and other argument com- 3.5. Challenges and future directions
ponents. Additionally, research has explored
various deep learning architectures, such as • Despite significant progress, several chal-
convolutional neural networks and recur- lenges remain in AM for decision making.
rent neural networks, tailored specifically for These include:
AM tasks ([3] Feng et al., 2023; [4] Yin et al., • Limited performance in complex argu-
2022). mentation: Existing models often strug-
gle with complex argument structures,
such as nested arguments and implicit
3.3. Evaluation of argument strength relationships.
and persuasiveness • Lack of explainability: While models
Beyond the identification and classification can predict argument components, they
of arguments, research has ventured into often lack transparency in their reasoning
analyzing their strength and persuasive- process, making it difficult to trust their
ness. This involves assessing the quality of outputs.
evidence, the logical soundness of reason- • Ethical considerations: Bias and fairness
ing, and potential fallacies employed in the issues in NLP models can potentially lead
argument. to biased and unfair decision-making if
Several approaches have been proposed, not addressed appropriately.
including using argumentation schemes and
defeasible reasoning to evaluate argument 3.6. Future research directions
strength [11]. Additionally, research has include
explored the role of sentiment analysis and
• Developing more robust models for com-
discourse analysis in assessing the persuasive-
plex argumentation: Exploring new deep
ness of arguments [12, 6].
learning architectures and incorporating
linguistic knowledge to improve model
3.4. Applications in decision-making performance.
AM has been applied to various decision- • Enhancing model explainability: Utilizing
making contexts, demonstrating its poten- interpretable machine learning techniques
tial to improve the quality and efficiency of to understand how models arrive at their
decision-making processes. Some examples predictions.
include: • Addressing ethical concerns: Implement-
ing fairness aware techniques and devel-
• Legal reasoning: Extracting legal argu- oping ethical guidelines for the responsible
ments from case law and legal documents use of AM in decision-making.
Decision-making using argumentation mining framework on English text 383
4. Critical Analysis of Previous Works
Study Task Model DataSet F1 score
Feng et al. Argumentati on strength CNN-based Debate 78.5 %
(2023) evaluation model transcripts
Li et al. (2023) Argumentati on scheme RoBERT Online news 87.1 %
[10] recognition a-based model articles
Huang et al. Argumentation component BERT-based Legal 92.3 %
(2022)[9] identification model documents
Reed and Rowe Argumentati on scheme Defeasible logic Legal 82.4 %
(2018) [11] evaluation framework arguments
Lippi and Torroni Discourse analysis for argument Discourse Political 80.3 %
(2016) [6] persuasiveness markers analysis debates
Mochales and Argument extraction from legal Rulebased Legal case law 76.9 %
Moens [5] documents system
5. Thematic overview and persuasiveness of each argument [9].
Logical reasoning, employing defeasible
In the realm of human thought and decision- logic or other formal systems, scrutinizes the
making, arguments are essential for rational argument’s structure, identifying potential
discourse and informed choices [7]. AM lever- fallacies and inconsistencies [7]. Evidence
ages artificial intelligence to extract, analyze, analysis, encompassing external resources
and evaluate arguments from text (Feng et and factchecking techniques, gauges the
al., 2023) [9], unveiling their intricate struc- quality and credibility of the evidence used
ture and components. to support the argument [10]. Sentiment
analysis and discourse analysis provide fur-
5.1. Argument identification and ther layers of understanding, examining the
classification: Unraveling the emotional tone, subjectivity, and rhetorical
structure of arguments strategies employed within the argument,
shedding light on its potential persuasive-
AM has revolutionized our understanding across ness and communicative intent [6].
decision making contexts, revealing arguments
in legal proceedings, policy debates, customer
reviews, and online discussions [5, 6]. Core to 5.3. Decision-making support:
AM is the process of identifying and classify- Integrating AM into the
ing arguments, involving coreference resolution, decision-making process
claim identification, premise identification, war- Insights from AM integrate into decision-mak-
rant identification, and argumentation scheme ing through summarization, visualization,
recognition [8][11][12]; Feng et al., 2023). and ABDSS [5]. Argument summarization
condenses key points, while visualization
5.2. Argument evaluation: Assessing employs tools like argument graphs and trees
the strength and persuasiveness [7]; Feng et al., 2023). ABDSS, integrating
of arguments AM techniques, provides structuring, com-
parison, and recommendations for well-rea-
AM delves into the realm of argument eval- soned choices [9].
uation, meticulously assessing the strength
384 Innovations in Computational Intelligence: Trends and Technologies
6. Conclusion References
In the intricate realm of human thought [1] Rocha G and Cardoso HL. Toward a relation-
and decision-making, arguments stand as based argument extraction model for argu-
the cornerstones of rational discourse and mentation mining. Lecture Notes in Computer
informed choices [7]. They provide a struc- Science (Including Subseries Lecture Notes
tured framework for presenting claims, sup- in Artificial Intelligence and Lecture Notes
in Bioinformatics). 2017;10583 LNAI:94-
porting evidence, and establishing logical
105. doi:10.1007/978-3-319-68456-7_8.
connections, enabling us to navigate com- [2] Macagno F, Walton D, and Reed C. Argu-
plex issues and reach well-reasoned conclu- mentation schemes. History, classifications,
sions. AM emerges as a powerful tool in and computational applications. IFCoLog
this domain, harnessing the computational Journal of Logics and Their Applications.
prowess of artificial intelligence to extract, 2017; 4(8). Available from: https://ssrn.com/
analyze, and evaluate arguments from nat- abstract=3092491
ural language text (Feng et al., 2023; [9]. [3] Andreas P and Manfred P. A Survey.
By delving into the depths of textual data, From argument diagrams to argumenta-
AM unveils the underlying structure of tion mining in texts. International Jour-
arguments, shedding light on the intricate nal of Cognitive Informatics and Natural
interplay of claims, premises, warrants, and Intelligence. 2013;7(1):1–3. doi:10.4018/
rhetorical devices that shape our persuasive jcini.2013010101
[4] Feng VW and Hirst G. Classifying argu-
appeals [10].
ments by scheme. Proceedings of the 49th
The integration of AM into decision- Annual Meeting of the Association for Com-
making processes has far reaching implica- putational Linguistics: Human Language
tions, fostering more informed, reasoned, Technologies. 2011;1:987–996
and transparent decision-making [6]. [5] Mochales R and Moens M. Argumenta-
By analyzing and evaluating arguments, tion mining. Artificial Intelligence and
decision-makers can gain a deeper under- Law. 2011;19(1):1–22. Available from:
standing of the issues at hand, identify http://link.springer.com/article/10.1007/
potential biases and fallacies, and make s10506-010-9104-x
more informed choices that are grounded [6] Lippi M and Torroni P. Argumentation min-
in evidence and sound reasoning [8]. In a ing: state of the art and emerging trends.
world where decisions often have significant ACM Transactions on Internet Technology.
consequences, AM empowers us to navigate 2016;16(2). doi:10.1145/2850417.
[7] Cabrio E, Villata S, and Gandon F. A
complex challenges with greater clarity and
support framework for argumentative
make choices that align with our values and discussions management in the web. Lec-
aspirations [5]. ture Notes in Computer Science (Includ-
As we look to the future, AM holds ing Subseries Lecture Notes in Artificial
immense promise for further refining our Intelligence and Lecture Notes in Bioin-
ability to analyze and evaluate arguments, formatics). 2013;7882 LNCS:412-426.
providing decision-makers with even more doi:10.1007/978-3-642-38288-8_28.
sophisticated tools to navigate the intrica- [8] Peldszus A and Stede M. From argument dia-
cies of human discourse and make informed grams to argumentation mining in texts: A
choices that shape our world. By continuing survey. International Journal of Cognitive Infor-
to invest in AM research and development, matics and Natural Intelligence. 2013;7(1):1–
we can harness its power to foster more rea- 31. doi:10.4018/jcini.2013010101.
soned, ethical, and impactful decision-making [9] Chen C and Chen H. Overview of the
NTCIR-16 FinNum-3 task: investor’s and
across a wide range of domains.
manager’s fine-grained claim detection.
Decision-making using argumentation mining framework on English text 385
Proceedings of the 16th NTCIR Conference Tools. 2004;13(4):961–979. doi:10.1142/
on Evaluation of Information Access Tech- s0218213004001922.
nologies. 2022:87–91. [12] Somasundaran S and Wiebe J. Recogniz-
[10] Cheng L, Bing L, He R, et al. IAM: a com- ing stances in online debates. Proceed-
prehensive and large-scale dataset for inte- ings of the Joint Conference of the 47th
grated argument mining tasks. Proceedings Annual Meeting of the Association for
of the Annual Meeting of the Association for Computational Linguistics and the 4th
Computational Linguistics. 2022;1:2277– International Joint Conference on Natural
2287. doi:10.18653/v1/2022.acl-long.162. Language Processing. 2009; June:226–234.
[11] Reed C, Rowe G. Lippi and Torroni, 2016. doi:10.3115/1687878.1687912.
International Journal of Artificial Intelligence
53 Potato leaf disease classification and
prediction using CNN
UdayPratap Singha, Sandeep Kumar Mishrab,
ShubhamVishwakarmac, Suraj Kumard, and Vinayak Pandeye
Department of Computer Science and Engineering, Babu Banarasi Das Institute of Technology and
Management, Lucknow, India
Abstract: Crop diseases substantially hinder agriculture, which has an impact on people’s capacity to support
themselves and the stability of the economy. In this investigation, convolutional neural networks (CNNs) were
used to develop a novel method for identifying the three potato plant diseases: late blight, early blight, and
healthy. The model demonstrated exceptional accuracy in identifying diseases. Viability assessments revealed
simple integration into existing systems at affordable implementation costs. Stakeholders acknowledged the
model’s value in decision-making and provided positive feedback. This study demonstrates how deep learning
models can improve overall agricultural health, effectively manage illnesses, and lower crop losses.
Keywords: Deep learning (DL), neural network (NN), convolutional neural network (CNN), image augmentation
1. Introduction treating crop diseases in the early stages,
reflects a transformative use of technology to
The agricultural sector, a cornerstone of enhance productivity and mitigate challenges
India’s economy, has faced challenges due faced by farmers during the initial phases of
to the widespread impact of COVID-19. crop growth. To determine whether leaves are
With a substantial portion of the population diseased, two principles are necessary.
employed in agriculture, the cultivation of
versatile crops like potatoes becomes pivotal, i) Object recognition
constituting 28.9% of agricultural produce. ii) Image classification
Recognized as the “king of vegetables,” pota-
toes stand out as a cost-effective and abun- The goal is to identify the type of plant AQ1:
dant source of potassium among fruits and disease. Fungi responsible for early and late Figures
vegetables. and
Tables are
Farmers, aware of the importance of lim- not cited
iting disease infections for optimal yield, in text,
please
grapple with the time-consuming nature of check and
manual inspection. Early disease detection confirm
is crucial, and recent advancements in arti- for
clarity.
ficial intelligence, particularly deep neural
networks (DNN), offer promise in address- Figure 53.1. Healthy leaf/Early blight/Late AQ2:
blight. Please
ing this challenge. The application of DNNs provide
in agriculture, specifically for identifying and Source: AUTHOR'S SELF GENERATED source for
all the fig-
ures and
tables.
a
[email protected]; [email protected]; [email protected]; dsurajpatel0669@
gmail.com; [email protected]
DOI: 10.1201/9781003606635-53
Potato leaf disease classification and prediction using CNN 387
blight mostly affect potato leaves [1,2]. to the volume of available data. The recom-
Figure 53.1 illustrates the differences between mended method starts with the crucial step of
healthy and unhealthy leaves. collecting data, featuring over 2150 images
CNN, among DNN algorithms, efficiently showcasing various leaf patterns, including
transforms input images into vector represen- early blight and late blight diseases in potato
tations for various vision tasks. Pre-training leaves.
the CNN with images allows it to serve as Table 53.1 provides a detailed dataset
an effective image “encoder.” Recent studies description for potato leaf disease identifica-
confirm CNN’s effectiveness in image classifi- tion, specifying disease classifications and the
cation and object recognition. image count.
2. Methodology Table 53.1. Number of images with diseases
The article explores convolutional neural Disease Type of Number of image
networks (CNNs) for picture classification, disease
especially in leaf disease identification. It Early blight Fungal 1000
outlines the steps for implementing a CNN Late blight Fungal 1000
model, including data gathering, preproc-
Healthy No disease 150
essing, sorting, and training. The impor-
tance of presenting disease and remedies Source:
AUTHOR'S SELF GENERATED
when a leaf defect is detected is empha-
sized. Essential modules for DL in leaf dis- 2.2. Preprocessing
ease identification is provided. Proper data
preparation, including resizing and normal- Preprocessing is necessary to ensure accu-
ization, is highlighted as crucial for effec- rate results from any dataset. The photos in
tive CNN model training. this suggested approach are all cropped to
the same size, with the sole focus being the
potato plant leaf. This is because numerous
2.1. Data collection
photos with different patterns can influence
The first step in analyzing and implement- the classification outcome.
ing any algorithm is data collection, where
prediction accuracy is directly proportional 2.3. Noise cancelation
Before an image moves on to the next stage,
only the essential features needed for the clas-
sification process are kept in it. Every other
feature is removed. The process of filtering
involves reducing contrast and smoothing
out an image to only retain the most impor-
tant and pertinent details.
2.4. Feature extraction
It is a necessary step after noise reduction in
images. Using feature extraction, an image’s
size could be decreased without compromis-
Figure 53.2. Process flow diagram. ing important details. It also removes the
Source: superfluous features.
AUTHOR'S SELF GENERATED
388 Innovations in Computational Intelligence: Trends and Technologies
2.5. Classification pivotal role in generating a detailed feature
map enhances the model’s capacity for image
Any classification algorithm or DNN, includ-
analysis. Through convolution operations
ing artificial neural networks (ANN), CNN,
with the kernel, Pooling layers, and ReLU,
or support vector machines (SVM), can assign
this layer extracts distinctive picture features,
a specific target class label to an image. How-
forming a foundational element in the neural
ever, CNN stands out in the realm of image
network’s hierarchical architecture.
categorization due to its superior accuracy
ReLU layer: It is an activation layer that
compared to other algorithms. Its effective-
functions in a formulaic way.
ness in capturing spatial hierarchies and pat-
f(x) = max(0,x) where x is an input value
terns makes CNN particularly well-suited for
Following convolution process, ReLU
tasks involving image classification.
introduces non-linearity, crucial for captur-
ing intricate patterns in the data. The train-
2.6. Model building ing process gains momentum with the ReLU
Within the CNN category, CNN emerge as layer, swiftly converting negative values to
a subset of NN utilized for deep learning. 0. Simultaneously, the pooling layer strategi-
CNNs signify a noteworthy progression in the cally diminishes the spatial resolution of fea-
realm of image recognition technology, find- ture maps by redistributing neuron weights.
ing extensive application in visual imagery Maximum pooling is used in this process.
analysis and image classification. Flatten layer: It serves to condense a 2D
dataset into a singular feature vector, stream-
2.7. Convolutional neural network lining data for further processing in the neu-
ral network.
CNNs are 2D neural networks designed to Fully connected layer: This layer estab-
process input in two dimensions. Using NN lishes seamless neural connections, compre-
for image pre-processing is logical due to the hending the feature vector from preceding
analogous representation of images. Convo- layers. Using the softmax technique, it makes
lution is a unique technique that filters data definitive predictions, contributing to the final
before creating a feature map, producing classification. The integration of backpropa-
observable features in pictures. These net- gation is pivotal, iteratively refining network
works pick up filters designed to solve certain parameters to minimize error and enhance
prediction problems during training. A single overall model accuracy.
multiplication of the filter by the input array
yields a single value. The filter produces a 2D
vector known as features when it is applied. 3. Thematic Overview
After being built, these features pass via non-
linearity, such ReLU, and are directed toward 3.1. Types of potato leaf diseases
the feature map for the fully linked layer’s The literature highlights various potato
outputs. CNNs architecture is as follows: leaf diseases, including common ones like
i. Convolutional layer late blight, early blight, and potato leaf roll
ii. ReLU (Rectified linear unit) layer virus, which have been subjects of classifi-
iii. Pooling layer cation using CNNs. Studies in the literature
iv. Fully-connected (FC) layer. frequently focus on these diseases due to
their substantial impact on crop yield. The
The convolution layer: This layer is a hub reviewed studies showcase the effectiveness
for crucial calculations in the neural network, of CNNs in distinguishing and classifying
functioning as adept feature extractors. Its multiple disease types from image data.
Potato leaf disease classification and prediction using CNN 389
3.2. Comparison of model 3.4. Performance metrics
performance Evaluating CNN performance in disease clas-
In the realm of ML for agricultural disease sification is crucial for model validation. Key
detection, various algorithms have been performance metrics highlighted in the litera-
explored. Traditional techniques like Naive ture include:
Bayes (88.67%), KNN (94.01%), and SVM Accuracy: Widely reported, it offers a gen-
(96.82%) have demonstrated commendable eral sense of model performance but can be
performance. Notably, CNNs outshine others misleading for imbalanced datasets.
with a reported accuracy of 99.08%, show- Confusion matrix: Frequently used to
casing their ability to extract complex pat- illustrate CNN classification capabilities
terns from potato leaf images. across different disease types.
Table 53.2 details different algorithms
used for potato leaf disease diagnosis with
4. Critical Analysis
corresponding accuracy levels.
The adoption of CNN in the classification
Table 53.2. Model performance comparison
of potato leaf diseases has been a subject
of extensive research, yielding both innova-
S.No. Algorithm Ref No. Accuracy tive breakthroughs and highlighting areas of
1 ANN [17] 85–91% concern.
2 Naive Bayes [6] 88.67%
3 BPNN [10] 92% 4.1. Examination of methodological
4 NN [38] 93% strengths
5 KNN [6] 94.01% The systematic application of deep learning
6 SVM [6] 96.82% techniques, specifically CNNs, has shown a
7 SSD and RCNN [1] 94.63% remarkable ability to identify and classify
various potato leaf diseases. For instance,
8 CNN [3] 99.08%
the work by [3-5] exemplifies the high accu-
Source: racy rates achievable with CNNs, reporting
AUTHOR'S SELF GENERATED
a classification precision of over 95% using
3.3. CNN architectures for disease a modified ResNet architecture. Such meth-
classification odologies underscore the strengths of deep
learning approaches.
Various CNN architectures have been inves-
tigated for disease identification, with notable
4.2. Scrutiny of methodological
ones including:
Alex Net: Adapted for simplicity and effi- weakness
ciency in training with limited datasets. However, these successes often overshadow
VGGNet: Preferred for its deep architec- inherent methodological weaknesses. The
ture and robust feature extraction capabili- literature reveals a common shortfall in the
ties, several papers report its use in extracting form of limited dataset diversity[6-10], which
intricate patterns from leaf images. potentially compromises the model’s general-
ResNet: Its residual learning framework izability. Furthermore, studies such as those
is highlighted for addressing vanishing gra- by[11-19] have been critiqued for their lack
dients, enabling the training of very deep of rigorous validation techniques, raising q-
networks, which is beneficial for complex uestions about the replicability of their repo-
classification tasks. -rted success.
390 Innovations in Computational Intelligence: Trends and Technologies
4.3. Identification of research gaps farmers to respond to disease outbreaks
swiftly, potentially reducing crop losses and
The current body of research demonstrates
minimizing chemical interventions. The
a noticeable gap in the exploration of CNN
implications for sustainable agriculture are
applications under varied environmental
profound, offering pathways to reduce envi-
conditions. The literature calls for more com-
ronmental impacts while maintaining plant
prehensive studies that simulate real-world
health.[4]
scenarios, as highlighted by [6], who argue
for a broader range of data reflective of dif-
ferent stages of disease progression. 6. Recommendations for Future
Research
5. Synthesis and Implications The creation of voice-activated Smartphone
The collective insights from studies employ- software specifically for illiterate farmers
ing CNN for the classification of potato leaf presents a revolutionary approach to farm-
diseases provide a compelling narrative about ing. Voice-based tutorials replace written
the strengths and potential of this technology text instructions, making the program user-
in agricultural applications. friendly and accessible. Its main goal is to
identify and treat leaf diseases, which signifi-
cantly reduce agricultural production. Farm-
5.1. Synthesized findings from ers are able to identify the precise illnesses
current literature causing damage to their crops with preci-
The literature consistently demonstrates sion by utilizing a comprehensive database
CNNs’ superior performance over traditional of leaf diseases. The program also includes a
methods in accuracy and speed of disease visual representation tool that shows farm-
classification [12]. Notably, the depth of net- ers the proportion of damaged leaves, which
works like ResNet-50 and the Inception series helps them determine the severity of the ill-
has been correlated with higher accuracy in ness. With the information and abilities this
disease identification [14], underpinning the cutting-edge program gives illiterate farmers,
importance of model complexity in extract- they can better manage their crops, boost pro-
ing detailed features from leaf imagery. ductivity, and raise their standard of living.
5.2. Emergent trends and patterns 7. Conclusion
A recurring theme is the efficacy of transfer We have developed a comprehensive NN sys-
learning in overcoming dataset constraints, tem capable of distinguishing between healthy
particularly in leveraging models pre-trained and infected leaves in images. This model inte-
on large datasets like ImageNet [5]. Real- grates an encoder-decoder CNN, facilitating
time detection systems, as proposed by [4], the effective classification of leaf images. This
are beginning to be piloted, reflecting a shift publication consolidates findings from mul-
toward proactive disease management in tiple studies utilizing DNNs for potato leaf
agriculture. disease identification. Two prevalent types
of potato leaf disease, early blight and late
blight disease, were addressed. In comparison
5.3. Technological advancements and
to other Using DNNs, CNN outperformed
agricultural impacts ANN at 85.07 percent and SVM at 88.87
These trends signify a promising horizon for percent in the identification and classification
precision agriculture, with CNNs enabling of illnesses, attaining 99.06 percent accuracy.
Potato leaf disease classification and prediction using CNN 391
The work represents a major advancement [8] Bhardwaj A and Patel NR. Potato leaf dis-
in the use of deep learning for complex tasks ease detection using ML algorithms. In: Pro-
like image analysis and feature extraction by ceedings of the International Conference on
highlighting CNN’s accuracy and efficacy in Recent Innovations in Signal Processing and
recognizing and classifying illnesses. Embedded Systems (RISE). IEEE; 2020:1–6.
doi:10.1109/RISE51108.2020.9240883.
[9] Khan AR, Saleem M, and Ahmed S. Leaf
disease detection in plants using machine
AQ3:
Reference
References learning techniques. In: Proceedings of
citations [1] Radhika P, Tejeswara Murthy P, Pava- the International Conference on Innova-
are not in
sequence,
neeshwar Reddy G, DurgaVara Prasad V, tive Computing and Communications
please Harshitha T, and Bharath NBV. Potato (ICICC). IEEE; 2020:1–5. doi:10.1109/
check and leaf disease detection using convolutional ICICC49352.2020.9272653.
confirm neural networks. International Research [10] Vargas-Rodríguez YL, Villegas-González JA,
for
clarity. Journal of Engineering and Technology. and Bautista-Becerril JM. Leaf disease detec-
2023;10(5):185. tion in plants based on convolutional neu-
CHECKED AND CONFIRMED
[2] Pasalkar J, Memane S, More C, Gorde G, ral networks and machine learning. Sensors.
and Gaikwad V. Potato leaf disease detec- 2020;20(4):1057. doi:10.3390/s20041057.
tion using machine learning. International [11] Ji Y, Li J, Zhu Y, Luo Y, Liu D, and Wang
Research Journal of Engineering and Tech- L. A deep learning approach for leaf disease
nology. 2023;10(2):14. detection based on a novel convolutional
[3] SahSudi P, Gupta N, Baniya KK, Ban- neural network. Sensors. 2020;20(7):1892.
gari S, and Rachana P. Potato leaf dis- doi:10.3390/s20071892.
ease detection using convolutional neural [12] Ma Y, Xie X, Li W, Song J, and Cui H. Potato
networks. Journal of Algebraic Statistics. leaf disease detection using deep learning
2022;13(2):1929–1935. and multi-scale convolutional neural net-
[4] Kaur A and Singh D. Deep learning for work. Computers and Electronics in Agri-
potato leaf disease detection using convolu- culture. 2019;165:104960. doi:10.1016/j.
tional neural networks. Computers and Elec- compag.2019.104960.
tronics in Agriculture. 2021;187:106267. [13] Kachare V and Lokhande R. Early detec-
doi:10.1016/j.compage.2021.106267. tion of potato late blight disease using
[5] Srivastava S and Chhabra M. Potato leaf machine learning. In: Proceedings of the
disease classification using deep learn- International Conference on Sustainable
ing techniques. In: Proceedings of the Computing in Science, Technology and Man-
3rd International Conference on Com- agement (SUSCOM). Springer; 2019:419–
puting Methodologies and Commu- 428. doi:10.1007/978-981-13-1796-8_42.
nication. Springer; 2021:921–926. [14] Shahbaz M, Munawar S, Naveed M, and
doi:10.1007/978-981-16-3660-5_90. Mehmood R. Identification of potato diseases
[6] Sunita M and Singh J.Potato plant disease iden- using deep convolutional neural networks.
tification using machine learning algorithms. Computers and Electronics in Agricul-
In: Proceedings of the International Confer- ture. 2019;166:105002. doi:10.1016/j.
ence on Advances in Computing and Data compag.2019.105002.
Sciences (ICACDS). Springer; 2020:517– [15] Fuentes A and Yoon S. Deep learn-
526. doi:10.1007/978-981-15-5254-1_51. ing for plant disease detection and
[7] Asri N, Moustaid K, and Fahsi R. Potato diagnosis. Annual Review of Phytopa-
leaf disease detection using machine learn- thology. 2019;57:211–230. doi:10.1146/
ing techniques. In: Proceedings of the annurev-phyto-082718-100308.
International Conference on Sustainable [16] Ghosal S, Saha S, Nasipuri M, and Kundu
Intelligent Systems (SUSI). IEEE; 2020:291- M. Leaf disease detection using convo-
299. doi:10.1109/SUSI50939.2020.930. lutional neural network and transfer
392 Innovations in Computational Intelligence: Trends and Technologies
learning. Computers and Electronics in Agri- leaf disease classification. In: Proceed-
culture. 2019;165:104961. doi:10.1016/j. ings of the 2018 International Conference
compag.2019.104961. on Innovations in Information Technol-
[17] Vinayakumar R, Sudheesh K, and Ravi ogy (IIT). IEEE; 2018:1–6. doi:10.1109/
R. Deep learning based classification of INNOVATIONS.2018.8554751.
plant diseases using leaf image analysis. [19] Siva Kumar G and Saraswathi S. Classifi-
Computers and Electronics in Agricul- cation of plant leaf diseases using machine
ture. 2018;145:311–318. doi:10.1016/j. learning techniques. In: Proceedings of the
compag.2018.01.018. International Conference on Intelligent Sus-
[18] Toda Y and Win KT. A comparative study tainable Systems (ICISS). IEEE; 2017:1666–
of machine learning techniques for tomato 1671. doi:10.1109/ISS1.2017.8388989.
54 Exploring the impact of integrated
technological solutions on student
engagement administrative efficiency
Anurag Tiwaria, Ayush Mauryab, Arushi Baranwalc,
Amritesh Guptad, and Anurag Pathake
Computer Science and Engineering, Babu Banarasi Das Institute Of Technology and Management,
Lucknow, India
Abstract: This paper examines “Acadia,” a groundbreaking Android and web-based solution meticulously
designed to meet the evolving needs of educational institutions in today’s dynamic and fast-paced world.
Educational management, spanning schools, colleges, and universities, involves intricate tasks ranging from
student enrollment to performance evaluation. Recognizing the imperative role of technology in the education
sector, “Acadia” emerges as a timely and comprehensive response to the growing demands for modern solutions.
Going beyond the realm of traditional software, “Acadia” signifies a commitment to elevate the management
and operational capabilities of educational institutions. Enveloped in a user-centric design and feature-rich
architecture, the system aims to simplify institute management, fostering efficiency, transparency, and collabo-
ration among administrators, faculty, students, and parents. Core features encompass student management,
attendance tracking, course scheduling, staff management, performance evaluation, and communication. The
design philosophy of “Acadia” places a premium on user experience, ensuring flexibility and accessibility across
both web and Android platforms. This adaptability caters to the diverse needs of users, whether administra-
tors managing schedules or students checking grades on smartphones. Security, a paramount concern in the
digital age, is addressed through robust measures within “Acadia” to safeguard sensitive educational data from
unauthorized access or breaches. In addition to providing an overview of “Acadia,” this paper delves into its
innovative attendance system. This system employs a multi-faceted approach, incorporating OTP, biometric
verification, and proximity sensors to enhance accuracy and security. The examination of these features dem-
onstrates Acadia’s forward-thinking approach, leveraging technology to fortify educational institutions against
potential fraudulent practices and offering a robust, reliable, and modern solution for tracking student attend-
ance. As a beacon of change in educational institute management, “Acadia” holds the potential to significantly
contribute to the ongoing progress of the education system.
Keywords: Smart education, proximity and biometrics, data analytics, alumni, global community
1. Introduction in reshaping the educational sector cannot be
overstated. As institutions grapple with the
In the rapidly evolving landscape of educa- complexities of student enrollment, attend-
tional technology, where traditional meth- ance tracking, course scheduling, faculty
ods are giving way to digital innovations, the management, performance evaluation, and
efficient management of educational institu- seamless communication, the need for com-
tions stands at the crossroads of transforma- prehensive, adaptable, and user-friendly solu-
tion. The pivotal role that technology plays tions has never been more urgent. This review
[email protected]; [email protected]; [email protected];
a
[email protected]; [email protected]
d
DOI: 10.1201/9781003606635-54
394 Innovations in Computational Intelligence: Trends and Technologies
paper embarks on a profound exploration applications to innovative frameworks, in
into the heart of modern educational manage- educational settings [1].
ment solutions, delving deep into the diverse 2. Investigating practical implementa-
array of digital tools and strategies designed tions: The literature review will investi-
to simplify, optimize, and enhance every facet gate practical implementations of digital
of educational institute operations. Amidst technologies, as highlighted in the pro-
this digital revolution, it becomes paramount vided abstracts. It will explore case stud-
to critically analyze the evolving landscape ies, experimental studies, and real-world
of educational management software. This applications to assess the effectiveness and
review aims to dissect the core functionalities, feasibility of these technologies in diverse
technological advancements, user experiences, environments.
and impact on educational outcomes brought 3. Addressing challenges and solutions:
forth by innovative solutions like “Acadia” Understanding the challenges associated
and its counterparts in the market. ACADIA with the integration of digital technologies
epitomizes the integration of technological is crucial. This review will identify common
solutions to enhance student engagement and challenges faced by educators, institutions,
administrative efficiency. By exploring the and organizations during the adoption
impact of integrated technological solutions, of digital tools. Moreover, it will explore
the ACADIA project contributes significantly innovative solutions and best practices
to the ongoing discourse on leveraging tech- employed to overcome these challenges.
nology to create more engaging and efficient
educational environments. The seamless inte-
gration of ACADIA’s features aligns perfectly 2. Methodology
with the objectives of understanding how The development of the ACADIA tool, an
integrated technologies can transform educa- Android-based unified application for insti-
tion, making it a noteworthy case study in the tute management, is guided by a structured
broader exploration of this topic. methodology designed to encompass vari-
ous critical features, ensuring its effectiveness
in student enrollment, attendance tracking,
1.1. Objectives and purpose of the course scheduling, staff management, perfor-
literature review mance evaluation, and data analytics.
This review paper aims to explore the diverse
applications of digital technologies in educa-
AQ1:
tion and beyond, emphasizing their transform- Figure
ative potential and the challenges associated 54.1 is not
cited in the
with their implementation. By delving into text, please
the discussed abstracts, the objectives of this provide
literature review include: approipri-
ate cita-
tion for
1. Examining the role of digital technologies clarity.
in education: This review intends to ana-
lyze the evolving role of digital technolo-
gies in the education sector, emphasizing AQ2:
Please
their impact on teaching methodologies, provide
student engagement, and learning out- source for
Figure 54.1. Process flow chart. the figure.
comes. It will explore the integration of
digital tools, ranging from software Source:
AUTHOR'S SELF GENERATED
Exploring the impact of integrated technological solutions on student engagement administrative efficiency 395
The primary aim of ACADIA is to foster flow within the campus. This oversight facili-
a transparent and collaborative educational tates the smooth functioning of academic
environment, improving the overall educa- operations.
tional experience. Data analytics for student reports:
Student registration: The initial step in the ACADIA utilizes data analytics to generate
methodology involves student registration, comprehensive student reports. Students can
requiring the use of official college email IDs. access their overall reports, including quiz
This step ensures that only authorized stu- scores, assignment scores, placement records,
dents gain access to the system, enhancing and class test reports. This feature empowers
security and data integrity. students to monitor their academic progress,
Classroom access and educational identify areas for improvement, and make
resources: Students can join classrooms by informed decisions regarding their educa-
using class codes, granting them access to tional and career paths.
notes, assignments, lecture videos, and quiz- The development process of ACADIA
zes. This functionality promotes a seamless Application follows industry-standard soft-
learning experience, enabling students to ware engineering methodologies, including
engage with course materials and assessments requirements analysis, design, implementation,
in one consolidated platform. testing, and deployment. The methodology
Attendance tracking: Attendance track- aims to deliver a robust and user-friendly tool
ing is a critical component, and ACADIA that streamlines educational processes, pro-
leverages proximity-based technology. Fac- motes transparency, and fosters collaboration
ulty members generate one-time codes that among students, faculty, alumni, and adminis-
students must input on their smartphones trators. ACADIA’s development seeks to pro-
to record their attendance. Additionally, a vide a comprehensive solution for institute
fingerprint authentication system enhances management, revolutionizing the educational
security and eliminates proxies. ecosystem while also empowering students
Faculty tools: Faculty members have the through data analytics-driven insights.
ability to create classrooms, upload assign-
ments, and administer quizzes. These features
empower educators to efficiently manage 3. Thematic Overview
their courses and maintain engagement with Educational institutions are embracing an era
their students. of technological advancements that are revo-
Global student community interaction: lutionizing the way students engage with their
ACADIA also serves as a platform for global academic environments. By reviewing various
student community interaction. Students can research papers, a common thread emerges,
easily connect with peers from around the showcasing the diverse applications of tech-
world, fostering a sense of community and nology to improve the student experience,
enabling cross-cultural exchanges. administrative efficiency, and alumni-student
Alumni engagement and job referrals: connections. Notably, radio-frequency iden-
Alumni play a vital role in the educational tification (RFID) and near-field communica-
journey. They can post job referrals on the tion (NFC) technologies, as well as biometric
platform, providing valuable opportunities systems, have taken center stage in the realm
for current students to secure placements in of student attendance tracking [7]. RFID and
reputable industries. NFC-based attendance systems have emerged
Administrative control: The administrative as powerful tools in administrative efficiency.
module grants administrators control over These technologies allow students to mark
students and faculty, ensuring an efficient their attendance with ease, eliminating the need
396 Innovations in Computational Intelligence: Trends and Technologies
for manual roll calls and paper-based sign-in absence of unified applications for institute
sheets. Real-time attendance data improves management and global student communi-
administrative decision-making and reduces ties to enhance their skill development, as well
the risk of errors and fraudulent attendance, as the lack of a platform connecting students
making them particularly well-suited for with alumni for referrals. This analysis aims
larger institutions where efficient attendance to shed light on the potential implications and
management is crucial. The implementation the significance of such a study.
of biometric systems, such as fingerprint and The absence of integrated technological
facial recognition, adds an extra layer of secu- solutions in educational institutions has far-
rity to attendance tracking. These systems are reaching consequences. In a rapidly evolving
highly accurate and challenging to manipu- technological landscape, where students’ edu-
late, addressing concerns related to proxy cational experiences are becoming increasingly
attendance. However, they also raise questions digital, the lack of unified applications means
about privacy and data security, which institu- that institutions miss out on opportunities to
tions must navigate while capitalizing on their streamline their administrative processes. This
potential to enhance attendance monitoring. can result in inefficiencies and hinder their
Beyond attendance tracking, the digital ability to provide a seamless learning environ-
age has seen the development of Android ment. Furthermore, the study’s focus on global
apps designed to connect students with student communities is crucial in today’s inter-
alumni. These applications serve as valuable connected world. Students from diverse back-
networking tools, enabling students to seek grounds can benefit greatly from a platform
mentorship, guidance, and referrals from that facilitates skill development and collabo-
alumni, thus enhancing their career prospects ration. Such a platform could serve as a bridge
[8]. The connection formed between current between students and various educational
learners and an experienced professional resources fostering engagement and cross-cul-
cultivates a sense of community within the tural learning experiences.
institution and fosters support for students as The research paper’s emphasis on con-
they embark on their professional journeys. necting students with alumni for referrals is
To further enhance the student experience, equally important. Alumni networks can pro-
the advent of digital classrooms has ushered in vide valuable insights, mentorship, and job
a new era of learning. Educators now have the opportunities to current students. Integrating
means to share notes, assignments, multime- this feature into the educational ecosystem
dia resources, and conduct interactive sessions can enhance the career prospects of students
with students online [5]. These digital tools and create a sense of community and conti-
promote efficient content delivery, engagement nuity within the institution.
through multimedia and interactive elements, The management of student attendance is
and have proven invaluable in accommodat- a fundamental aspect of academic adminis-
ing evolving learning needs, especially in times tration, and the absence of a feasible solution
of remote and hybrid learning. in this regard has wide-ranging consequences.
Traditional methods of attendance track-
ing, such as manual roll-calls or paper-based
4. Critical Analysis sign-in sheets, are not only time-consuming
The paper, “Exploring the Impact of Integrated but also prone to errors and manipulations.
Technological Solutions on Student Engage- These outdated methods not only impede
ment and Administrative Efficiency,” addresses administrative efficiency but also fail to pro-
a critical issue in the field of education and vide real-time insights into students’ attend-
institutional management. It highlights the ance patterns. An integrated technological
Exploring the impact of integrated technological solutions on student engagement administrative efficiency 397
solution for attendance management can digital content improves overall student
significantly enhance the overall educational satisfaction and participation.
experience. With the implementation of a reli- 4. Integration of virtual reality (VR) and aug-
able attendance tracking system, institutions mented reality (AR) in education: Explore
can streamline the process, reduce adminis- the integration of VR and AR technolo-
trative workload, and ensure the accuracy of gies in educational settings. Investigate
records. Furthermore, real-time attendance how immersive experiences can enhance
data can be invaluable for both students and learning in subjects like science, history,
faculty, as it allows for prompt interventions and engineering. Research the impact of
when students exhibit irregular attendance, VR/AR on student motivation, knowl-
potentially improving student engagement edge acquisition, and critical thinking
and performance skills. Additionally, analyze the feasibility
of incorporating VR/AR within budget
5. Recommendations for future constraints of educational institutions.
5. Blockchain technology in academic cre-
research dentialing: Explore the potential of block-
1. AI-powered personalized learning: chain technology in securely storing and
Explore the integration of artificial intel- verifying academic credentials. Investi-
ligence (AI) in personalized learning plat- gate the feasibility of creating decentral-
forms. Investigate how AI algorithms can ized, tamper-proof systems for academic
analyze students’ learning styles, prefer- transcripts, certifications, and diplomas.
ences, and performance data to tailor Assess the impact of blockchain-based
educational content. Assess the impact of credentialing on the authenticity of quali-
AI-driven adaptive learning on student fications, simplifying the verification
engagement, knowledge retention, and process for employers, educational insti-
academic achievement across diverse sub- tutions, and other stakeholders.
jects and grade levels. 6. Advanced real-time attendance analysis and
2. Teacher professional development in digital optimization strategies: Explore sophisti-
pedagogy: Focus on teacher training and cated real-time image analysis algorithms
professional development programs cen- integrated directly into the attendance
tered around digital pedagogy. Evaluate the application. Examine context-aware image
effectiveness of workshops, online courses, processing methods to improve headcount
and collaborative learning communities in accuracy by considering classroom dynam-
enhancing educators’ digital teaching skills. ics. Delve into resource optimization
Investigate the long-term impact of continu- techniques ensuring system efficiency in
ous professional development on teaching high-traffic situations. Explore standard-
methodologies, student engagement, and ized data formats and APIs to enable seam-
overall classroom dynamics. less data exchange between the attendance
3. Human-centric design in digital class- app and institutional databases.
rooms: Focus on human-centric design 7. Impact of digital integration on special
principles in digital classrooms. Research education: Focus on the impact of digital
how user experience (UX) design, interac- technologies on special education. Investi-
tive multimedia elements, and gamification gate how personalized learning platforms,
techniques can enhance student engage- assistive technologies, and accessible digi-
ment and knowledge retention. Investigate tal content enhance the educational experi-
the impact of intuitive interfaces on learn- ences of students with disabilities. Explore
ing outcomes and whether personalized the challenges faced in implementing
398 Innovations in Computational Intelligence: Trends and Technologies
inclusive digital education and identify the administrative world, conjuring seamless,
innovative solutions for creating univer- real-time data that not only reduces errors
sally accessible learning environments. but also revolutionizes decision-making in
educational institutions.
The scrutiny of biometric systems, though
6. Conclusions
raising ethical dilemmas, has brought forth
In traversing the landscape of digital tech- a revolutionary wave. These systems, with
nologies in education and institutional man- their precise fingerprint and facial recogni-
agement, this review has illuminated the tion technologies, have set new standards
transformative power of integrated technolog- for security and accuracy. The marriage of
ical solutions [1]. Through a critical analysis heightened security and attendance manage-
of various research papers, the pivotal role of ment has transformed these systems into the
digital tools in enhancing student engagement, guardians of educational integrity, ensuring
administrative efficiency, and alumni-student that every student’s presence is not just noted
connections has been underscored. In essence, but safeguarded with unparalleled accuracy
this review underscores the transformative
journey digital technologies have paved for
educational institutions. By embracing these References
innovations thoughtfully, institutions can cre- [1] Haleem A, Javaid M, Qadri MA, and Suman
ate an inclusive, innovative, and interconnected R. Understanding the role of digital technol-
educational ecosystem. The amalgamation ogies in education: A review. KeAi Chinese
of administrative efficiency, enhanced stu- Roots Global Impact. 2022.
dent engagement, and robust alumni-student [2] Puckdeevongs A, Tripathi NK, Witayang-
networks not only augments the educational kurn A, and Saengudomlert P. Classroom
experience but also contributes significantly to attendance systems based on Bluetooth low
shaping a future-ready workforce and a har- energy indoor positioning technology for
smart campus. MDPI. 2020.
monious society. This paper stands as a testa-
[3] Babu M, Sandhiya K, Preetha V, Eshwari
ment to the potential of technology, guiding SS, and Chitra MR. Design of alumni portal
stakeholders toward a future where education with data security. ResearchGate. 2021.
knows no bounds and where every learner’s [4] Shakhina I, Podzygun O, Petrova A, and
potential can flourish, regardless of geographi- Gordiichuk G. Smart education in the trans-
cal or societal constraints. formation digital society. ResearchGate.
Main findings and contributions: The 2023.
research papers dissected in this review [5] Sudarsana K. The use of Google classroom
serve as brilliant beacons, illuminating the in the learning process. IOP Science. 2019.
unexplored realms of digital ingenuity. The [6] Donath L and Mircea G, Rozman T.
revelations surrounding Bluetooth Low E-learning platforms as leverage for educa-
Energy-based attendance frameworks have tion for sustainable development. Research-
shattered conventional boundaries [2]. They Gate. 2020.
[7] Mahat SS and Mundhe SD. Proposed frame-
have not just streamlined administrative effi-
work: College attendance management
ciency but have redefined how institutions system with mobile phone detector. Interna-
function. The integration of RFID and NFC tional Journal of Research in IT and Man-
technologies into the fabric of educational agement. 2015.
management has turned what were once mun- [8] Aruna P, Begum MS, and Kumar DM.
dane attendance processes into sophisticated Alumni smart connect through Android
symphonies of precision and efficacy. These application. International Journal of Trend
technologies have become the magicians of in Research and Development. 2016.
55 Early stage lung cancer detection using
RESNET
Anjali Srivastavaa, Vinayakb, Aarohi Raic, and Ainy Khand
Department of Computer Science and Engineering, Babu Banarasi Das Institute of Technology and
Management, Lucknow, India
Abstract: Lung cancer remains the prime global demise inciting factor, especially with the highest mortality
rates since 1985. Detecting it early and accurately is crucial for improving patient survival chances. This study
conducts an approximate analysis of automated algorithms for early-stage respiratory tumor identification via
CT images. CT imaging is recommended for their effectiveness in revealing lung cancer nodules. The research
evaluates various approaches, utilizing data arrays like LIDC, ELCAP, LUNA-16, and AAPM. Segmentation,
feature extraction, neural system identification, and image pre-processing are all steps in the detection pro-
cess. It may be possible to improve accuracy and enable early-stage cancer prediction by placing emphasis on
ResNet-50 transfer learning models, which have demonstrated good accuracy in COVID-19 and breast cancer
detection. This finding could revolutionize the diagnosis and treatment of lung cancer by providing patients
with a more tolerable course for recovery.
Keywords: Lung cancer, LUNA 16, machine learning, CNN, RESNET 50, computer tomography (CT)
1. Introduction cancer. Two main types of datasets have been
used: those based on computed tomography
Lung cancer is a major global health concern,
(CT) scans and X-rays. CT scans, known for
driven by smoking and air pollution, with a
their superior accuracy, have become the pre-
projected 17 million deaths by 2030, mak-
ferred choice. CNNs, well-suited for image
ing it the second most common cancer type
classification, are effective in handling CT
worldwide, predominantly affecting males,
with over 2.2 million new cases in 2020. scan images.
Extensive research is dedicated to early detec- In medical disease detection using CNNs,
tion [1]. two main approaches exist: building mod-
Extensive research has focused on early els from scratch, which is data-heavy and
lung cancer detection, leveraging machine time-consuming, and transfer learning (TL),
learning (ML) and deep learning (DL). ML which modifies pre-trained models like
strategies like support vector machines Alexnet, VGG, ResNet, Inception, DenseNet,
(SVM) and K-nearest neighbor handle clas- and MobileNet, conserving computational
sification, while DL excels in computer vision resources. This study focuses on early lung
and speech processing due to complex neural cancer detection, evaluating the performance
networks, resulting in superior accuracy [2]. of various pre-trained models for CT images
Lately, DL logics, particularly convolu- through metrics and comparative analysis,
tional neural networks (CNNs), have been emphasizing the exploration of diverse archi-
employed for the early detection of lung tectural features [3]
[email protected]; [email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-55
400 Innovations in Computational Intelligence: Trends and Technologies
2. Methodology 2.2. CNN architecture
• AlexNet
2.1. Dataset In 2012, Alex Krizhevsky and colleagues
Datasets are crucial for ML, especially in introduced AlexNet, a revolutionary neu-
medical imaging. Expert-validated, labeled ral network with eight layers—five con-
data is vital for effective algorithm devel- volutional and three fully connected (FC).
opment in lung cancer detection. This sec- Notable features include the use of pool-
tion displays data sets utilized in recent ing, ReLU activation functions, and 11 ×
research on neural networks for detecting 11 filters for convolutional layers. Input
lung cancer. images must have dimensions of 227 ×
227 × 3, and pooling employs 3 × 3 filters AQ1:
• The lung image database consortium with varied strides [5]. Figure
(LIDC-IDRI) 55.1,
55.3 and
The LIDC-IDRI dataset includes 1018 55.4 not
cases with CT scan annotations in XML cited in
the text.
format, reviewed by four experienced Please
thoracic radiologists. It contains 244,527 check and
images from 1010 patients, enabling confirm
for clarity
diagnoses at patient and nodule levels.
Nodules are categorized, and detailed
AQ2:
diagnostic information, including meth- Please
ods and nodule types, is available. Figure 55.1. AlexNet architecture. provide
• LUNA16 Source:
AUTHOR'S SELF GENERATED
source for
the figure.
The LUNA16 dataset is a subset of
LIDC-IDRI, comprising 888 carefully The process of categorizing lung can-
chosen CT scans without layer width cer-affected areas using AlexNet involves
>2.5 mm and inconsistencies. With several key steps:
36,378 annotations, it specifically tar- i) Image loading: Import the LIDC IDRI
gets nodules ≥3 mm for lung cancer database, housing 254,727 images
screening. Merged annotations with from 1080 cases. For AlexNet train-
radii sums were averaged, yielding nod- ing, a subset of 2910 images is selected.
ules reviewed by 1, 2, 3, or 4 radiolo- Use the “split each label” function for
gists, totaling 2290, 1602, 1186, and a 70% training and 30% validation
777 nodules, respectively. distribution.
• National lung screening trial (NLST) ii) Loading AlexNet: Initialize the
The NLST involved 54,000 participants AlexNet architecture, highlighting key
from 2002 to 2004, comparing low-dose network specifics. Figure 55.8 illus-
CT and chest radiography for lung cancer trates the progression of AlexNet,
screening. Its aim was to determine if low- showcasing parameters such as coef-
dose CT scans reduce lung cancer mortal- ficients, offsets, and padding in convo-
ity in high-risk individuals. Radiologists lutional, rectified linear unit (ReLU),
assessed uncertain nodules or abnormali- and pooling layers.
ties for results. Jointly conducted by the iii) Replacement of the concluding layer:
National Cancer Institute’s Divisions of Swap convolutional layer with FC
Cancer Prevention and Treatment and layer to generate classification output
Diagnosis, the study aimed to evaluate and integrate a softmax layer for addi-
effective screening methods. tional refinement.
Early stage lung cancer detection using RESNET 401
iv) Network training: Utilize ‘train-Net- with the results providing insights into
work’ for image classification, speci- accuracy.
fying ‘Image Datastores’ for input • VGG 16
images, ‘layers’ for network configura- Compared to AlexNet, the Oxford Visual
tion, and ‘options’ for parameters like Geometry Group (VGG) model is more
learning rate (0.0001), accuracy target detailed and less complicated. 3 × 3 filters
(99.91%), max epochs (6), and vali- with stride, pad, and maximum pooling
dation frequency (every 3 iterations). sizes of 1 were employed in all network
Customize settings for plotting and layers. The VGG-16 architecture block
monitoring training progress, tailored design is shown in Figure 55.2. It consists
for lung cancer detection. of three fully-connected layers with a Soft-
v) Image categorization: The ultimate Max layer, thirteen convolution layers,
step entails categorizing output data five max-pooling layers, and sixteen layers
using confirming images, leading to the overall. Ultimately, the presence or absence
calculation of accuracy. Figure 55.2. of lung cancer is determined by a FC layer
visually represents validated images that incorporates the SoftMax layer [6].
alongside corresponding probabilistic • ResNet50
values. ResNet-50, short for Residual Net-
• Inception V1 work-50, is a revolutionary CNN that
The Inception V1 (GoogleNet) architec- has brought about a revolution in DL.
ture utilizes 1 × 1, 3 × 3, and 5 × 5 filters ResNet-50 is a deep residual learning-
to analyze images and reduce parameters based network that may be used to train
to 4 million from 60 million, making it very deep networks with hundreds of lay-
efficient. TL is crucial for training image ers. It was developed in 2015 by Kaim-
recognition models, especially for lung ing He and the Microsoft Research Asia
cancer detection, using GoogleNet. It team. Its inception was prompted by a
classifies images, distinguishing between remarkable observation in the realm of
benign and malignant tumors [5]. The DL—the straightforward addition of
process involves loading images, examin- more layers to a neural network did not
ing the network architecture, and using consistently yield performance improve-
pre-trained models. Base layers are frozen ments, contrary to conventional expecta-
to expedite training, and the network is tions. Despite the theoretical advantage of
trained with varying image sizes, achiev- enabling the network to assimilate prior
ing 94.10% accuracy in 46 hours. Vali- layers’ knowledge and additional infor-
dation identifies cancer-affected areas, mation, this didn’t manifest practically.
Figure 55.2. Problem identification using
AlexNet. Figure 55.3. VGG16 model architecture.
Source: Source:
AUTHOR'S SELF GENERATED
AUTHOR'S SELF GENERATED
402 Innovations in Computational Intelligence: Trends and Technologies
In response, the ResNet team intro- ResNet-50 essentially represents an
duced a groundbreaking architectural important milestone in the development
innovation known as skip connections, or of DL, redefining the possibilities of neu-
residual blocks. These connections ena- ral networks in computer vision tasks [7].
bled the network to retain crucial infor-
mation from earlier layers, augmenting its
capacity to acquire more meaningful data 3. Thematic Overview
representations. This design allowed for CNN architectures, like AlexNet, GoogleNet,
the effective training of networks with up ResNet50, and GoogleNet, were categorized
to 152 layers. With notable results such based on unique configurations of input,
as an astounding 3.57% error rate on the hidden, and output layers. These networks
ImageNet dataset and wins in well-known are crucial in classifying CT scan images as
competitions like the COCO and ILSVRC benign or malignant. Leveraging their struc-
object detection challenges, the archi- tures, this approach optimizes image classifi-
tectural innovation cemented ResNet’s cation, utilizing each network’s strengths for
leadership and potential in the DL field. accurate diagnostics.
ResNet-50’s 50 layers organized into five AlexNet, an eight-layer CNN, stands out
blocks, each including residual blocks, for its revolutionary design. Its layers include
are responsible for its amazing ability to max-pooling, normalization, convolutional
train incredibly deep networks. It per- operations, ReLU activations, FC layers, and
forms exceptionally well in a variety of dropout layers, culminating in a softmax
image-related tasks, regularly producing layer. Originally intended for 1000 classes,
cutting-edge outcomes in object detection, this study tailors AlexNet for binary clas-
picture classification, and image segmen- sification, distinguishing malignancy from
tation. Underpinning ResNet-50’s excep- benign attributes. The adapted AlexNet out-
tional performance are its distinct skip performs existing methods, showcasing supe-
connections, which are essential for effec- rior efficiency in image classification.
tive information preservation and learn- GoogleNet stands out for its substantial
ing. Through its architectural advances, 22 hidden layers, allowing for efficient sample
Figure 55.4. RESNET50 model architecture.
Source:
AUTHOR'S SELF GENERATED
Early stage lung cancer detection using RESNET 403
classification by automatically extracting fea- classification, it distinguishes malignant from
tures from input images. Initially intended benign traits, proving highly effective in image
for 1000 classes, this study customizes Goog- classification compared to existing methods.
leNet for binary classification, excelling VGG16, the champion of the 2014 Ima-
in distinguishing malignancy from benign geNet competition, prioritizes simplicity in
attributes, surpassing current methodologies hyperparameters. Its consistent use of 2 × 2
in image classification. max-pooling layers, 3 × 3 convolutional fil-
ResNet50 pushes neural network depth ters, and constant layer configurations sets it
with 50 hidden layers, enhancing sample apart. With 16 weighted layers and an empha-
classification efficiency. Like its predeces- sis on 3 × 3 filters for convolution, VGG16’s
sors, ResNet50 autonomously extracts fea- careful design makes it a potent CNN model
tures from input images. Adapted for binary for diverse image classification tasks.
4. Critical Analysis
Author Comparison between different Lung Cancer Detection System
Year Problem Statement Dataset Method Accuracy
[15] 2023 A deep learning method for CT scan images ResNet-50 98%
lung cancer detection and
classification
[16] 2023 A method for detecting and LIDC-IDRI Modified 97.02%
classifying lung cancer Using the Dataset Alex Net
AlexNet CNN algorithm model Architecture
[17] 2022 Support vector machine and LUNA16 dataset Lung Net 97.42%
modified AlexNet architecture SVM Model.
for the detection of lung cancer
[18] 2021 Detection of lung cancer with CT scan images VGG-16 and 97%
the VGGNET-16 architecture Resnet
[19] 2020 CNN’s capability to identify Histopathological CNN 96.11%
lung cancer from the images
histopathological images
[20] 2019 In-depth examination of the LIDC-IDRI image AlexNet and 99.01%
AlexNet and Google Net for the dataset. GoogLeNet
detection of lung cancer
[21] 2018 CT scan images for CT scan images SVM 92%
identification of lung cancer
5. Synthesis and Implications showcasing the advantages of merging DL
and classical techniques [22].
Bansal introduced an innovative approach Toğaçar introduced a DL model based on
to enhance image classification efficiency by DarkNet-19 that enhanced weak features
integrating deep features from the VGG19 and used SVM to classify data. The approach
model with classical methods like SIFT, SURF, yielded remarkable results, with an over-
ORB, and Shi-Tomasi corner detection. Their all accuracy of 99.69%, an amazing AUC of
work demonstrated that combining these fea- 99.3%, and strong F-measure, precision, recall,
tures with the random forest (RF) classifier and accuracy rates of 97.1%. Our method
achieved an impressive accuracy of 93.73%,
404 Innovations in Computational Intelligence: Trends and Technologies
effectively integrates optimization techniques similar studies, achieving an outstanding
to categorize photos accurately [23]. 99.60% accuracy and demonstrating the
Dritsas and Mesut suggested using ML, potential of hybrid DL models for precise
more especially the rotation forest model, cancer diagnosis [25].
for early lung cancer diagnosis in a different
situation, illustrating the adaptability of ML
methods in healthcare settings. 7. Conclusion
Inception V3, an evolved CNN from Research indicates AlexNet’s superiority over
Inception V1 (GoogleNet), addresses overfit- GoogleNet in diagnosing respiratory malig-
ting by introducing multiple filters, enhanc- nancies through neural network analysis of
ing robustness and performance. It prioritizes medical images, showcasing higher accuracy
parallel layers over excessive depth, refining and other metrics. Additionally, ResNet’s inno-
dimension reduction techniques for improved vative residual connections enable training of
processing and better activation dimensions, exceptionally deep networks by addressing
employing methods like maximum pooling vanishing gradients. Our lung cancer diagnostic
and mean pooling. approach employs four DL models—AlexNet,
VGG-19 is an extended version of VGG Inception V1, ResNet-50, and VGG16. Evalu-
with 19 layers, including convolutional, max- ation using CT and histology images highlights
pooling, and FC layers. Operating on 224 × 224 the potential of leveraging these networks for
RGB images, it maintains resolution through enhanced early detection. Future plans involve
spatial padding, using 3 × 3 kernels with ReLU enhancing performance by integrating DL
activations. ReLUs introduce non-linearity, with optimization techniques like fuzzy genetic
boosting classification and processing speed. algorithms. To sum up, AlexNet surpasses
The final layers comprise 1000 channels for a GoogleNet, while ResNet enables deeper net-
1000-way ILSVRC classification, employing a works, promising improved early diagnosis in
softmax function. For detailed insights, refer- lung imaging, with optimization avenues for
ence [24] elaborates on VGG-19’s architecture further enhancements.
and image classification capabilities.
References AQ3:
Reference
6. Recommendations for Future [1] Abdullah DM and Ahmed N. A review of citations
Works most recent lung cancer detection tech- are not in
niques using machine learning. Zenodo sequence,
please
Talukder’s ensemble model, utilizing the (CERN European Organization for Nuclear check and
LC25000 dataset, achieved an impressive Research). February 2021. doi:10.5281/ confirm
zenodo.4536818. for
99.05% accuracy in lung cancer screening, clarity.
CHECKED AND CONFIRMED
outperforming previous models. This success [2] Hatuwal BK and Thapa HC. Lung cancer
highlights the strength of ensemble models detection using convolutional neural net-
employing TL, showcasing their potential work on histopathological images. Int J Com-
for clinical applications. Phankokkruad’s put Trends Technol. 2020;68(10):21–24.
doi:10.14445/22312803/ijctt-v68i10p104.
research showcased the effectiveness of com-
[3] European Chemical Bulletin. https://www.
munal models in lung cancer detection. Inte- eurchembull.com/issue-content/a-compar-
grating individual TL models like VGG16, ative-study-of-cnn-architectures-for-lung-
ResNet50V2, and DenseNet201 as an ensem- cancer-detection-from-ct-scan-images-8730.
ble notably increased verification reliability [4] Riquelme D and Akhloufi MA. Deep learn-
to 91%. Chen’s hybrid framework, combin- ing for lung cancer nodules detection and
ing Inception v3 and feature extraction mod- classification in CT scans. AI. 2020;1(1):28–
ules for pathology image analysis, surpassed 67. doi:10.3390/ai1010003.
Early stage lung cancer detection using RESNET 405
[5] Almas B, Sathesh K, and Rajasekaran S. A for Advancement in Technology (ICO-
deep analysis of GoogleNet and AlexNet for NAT). Goa, India; 2023:1–6. doi:10.1109/
lung cancer detection. Int J Eng Adv Tech- ICONAT57137.2023.10080801.
nol. 2019;9(2):395-399. doi:10.35940/ijeat. [16] KBC, NK B. An approach of AlexNet CNN
b3226.129219. algorithm model for lung cancer detection
[6] Sheriff STM, Kumar JV, Vigneshwaran and classification. Int J Res IT Comput
S, Jones A, and Anand J. Lung can- Commun Technol. 2023;11(11s):49–54.
cer detection using VGG NET 16 archi- [17] Naseer T, Masood S, Akram S, et al. Lung can-
tecture. J Phys. 2021;2040(1):012001. cer detection using modified AlexNet archi-
doi:10.1088/1742-6596/2040/1/012001. tecture and support vector machine. Comput
[7] Wisdom ML. Understanding ResNet-50 in Mater Continua. 2023;74(1):2039–2054.
depth: architecture, skip connections, and [18] Sheriff STM, Kumar JV, Vigneshwaran
advantages over other networks. Wisdom S, Jones A, and Anand J. Lung can-
ML. March 30, 2023. https://wisdomml.in/ cer detection using VGG NET 16 archi-
understanding-resnet-50-in-depth-architec- tecture. J Phys. 2021;2040(1):012001.
ture-skip-connections-and-advantages-over- doi:10.1088/1742-6596/2040/1/012001.
other-networks/#What_is_ResNet-50. [19] Hatuwal BK and Thapa HC. Lung cancer
[8] Krizhevsky A, Sutskever I, and Hinton GE. detection using convolutional neural net-
ImageNet classification with deep convo- work on histopathological images. Int J Com-
lutional neural networks. Commun ACM. put Trends Technol. 2020;68(10):21–24.
2017;60(6):84–90. doi:10.1145/3065386. doi:10.14445/22312803/ijctt-v68i10p104.
[9] Szegedy C, et al. Going deeper with con- [20] Almas B, Sathesh K, and Rajasekaran S. A
volutions. In: 2015 IEEE Conference on deep analysis of GoogleNet and AlexNet for
Computer Vision and Pattern Recogni- lung cancer detection. Int J Eng Adv Tech-
tion (CVPR). Boston, MA, USA; 2015:1–9. nol. 2019;9(2):395–399. doi:10.35940/ijeat.
doi:10.1109/CVPR.2015.7298594. B3226.129219.
[10] He T, Zhang Z, Zhang H, Zhang Z, Xie J, [21] Makaju S, Prasad PWC, Alsadoon A, Singh
and Li M. Bag of tricks for image classifi- AK, and Elchouemi A. Lung cancer detec-
cation with convolutional neural networks. tion using CT scan images. Procedia Com-
In: Proceedings of the IEEE Conference on put Sci. 2018;125:107–114. doi:10.1016/j.
Computer Vision and Pattern Recognition; procs.2017.12.016.
2019:558–567. [22] Bansal M, Kumar M, Sachdeva M, and
[11] Liu YH. Feature extraction and image recog- Mittal A. Transfer learning for image clas-
nition with convolutional neural networks. sification using VGG19: Caltech-101 image
J Phys Conf Ser. 2018;1087(6):062032. data set. J Ambient Intell Humaniz Comput.
doi:10.1088/1742-6596/1087/6/062032. 2023;14(4):3609-3620.
[12] Yang J and Yang G. Modified convolutional [23] Toğaçar M. Disease type detection in lung
neural network based on dropout and the sto- and colon cancer images using the comple-
chastic gradient descent optimizer. Algorithms. ment approach of inefficient sets. Comput
2018;11(3):28. doi:10.3390/a11030028. Biol Med. 2021;137:104827.
[13] Sajja T, Devarapalli R, and Kalluri HK. Lung [24] Rajasekar V, Vaishnnave M, Premkumar S,
cancer detection based on CT scan images Velliangiri S, and Rangaraaj V. Lung can-
by using deep transfer learning. Trait Signal. cer disease prediction with CT scan and
2019;36(4):339–344. doi:10.18280/ts.360406. histopathological images feature analy-
[14] Mohite A. Lung cancer diagnosis using sis using deep learning techniques. Results
transfer learning. Int J Sci Res Manag. Eng. 2023;18:101111. doi:10.1016/j.
2021;9(11):621–634. doi:10.18535/ijsrm/ rineng.2023.101111.
v9i11.ec02. [25] Talukder MA, Islam MM, Uddin MA, et al.
[15] Khatun MF and Ajmain MR, Assaduz- Machine learning-based lung and colon can-
zaman M. A deep learning approach cer detection using deep feature extraction
to detect and classification of lung can- and ensemble learning. Expert Syst Appl.
cer. In: 2023 International Conference 2022;205:117695.
56 Cryptographic image concealment with
neural networks
Chinnala Balakrishna1,a,Shaik Saidhbi2,b, Balajee Maram3,c,
Mallikharjuna Rao K.4,d, B. Santhosh Kumar5,e, and
Sasibhushana Rao Pappu6,f
AQ1:
Please
1
AIML and Cyber Security, Guru Nanak Institute of Technology, Hyderabad, India
provide 2
Samara University, Semera, Ethiopia
division 3
School of Computer Science and Artificial Intelligence, SR University, Warangal, Telangana, India
details. 4
Data Science and Artificial Intelligence, IIIT, Naya Raipur, India
CKED AND CONFIRMED
5
Department of Computer Science and Engineering, Guru Nanak Institute of Technology,
Ibrahimpatnam, India
6
Department of Computer Science and Engineering, GITAM School of Technology, Visakhapatnam
Campus, GITAM (Deemed to be University), Visakhapatnam, India
Abstract: It offers the brief description of the novel work done in combining the cryptography with the deep
learning strategies for the purpose of ensuring the maximum security and immunity in the data hiding pro-
cesses executed on the images. The method outlined in this work capitalizes on the training abilities of neural
networks to learn statistical details from pictures that are obtainable in various environments. In this way, it
also obtains a higher data-hiding capacity within these photos, and at the same time, maintains a higher image
quality of the stego-image. The hidden data shows fairly good immunity to a number of assaults and trans-
formations; the examples showed quite a stability in compression, noise, and geometrical attacks. The used
method proves a high professionalism to hide the hidden information and, therefore, makes the steganalysis or
revealing the data by other unauthorized persons or software quite difficult. However, one must point out some
challenges that arise with neural network training like the fact that it is a highly computationally expensive
operation is one of the biggest challenges as well as the fact that parameters need to be selected properly. This
approach can be considered as effective for creating secure communication and can be implemented in several
domains such as military use, medical and financial spheres. As for the issues arising from the application of
the developed methodology, it is assumed that further studies will expand on the methodology effectively and
examine the problems related to its large-scale use.
Keywords: Cryptography, image, neural networks, steganography
1. Introduction higher difficulty in detection compared with
traditional steganography.
Therefore, it can be concluded that crypto- Steganographic image concealment with
graphic image concealment with the use of the help of using the neural networks has some
neural networks is a fairly new and prom- advantages comparing to the other variants
ising method of data hiding in images. It of steganography. For the first, it is possible
integrates the functions of cryptography to achieve a higher data hiding capacity. This
and deep learning to get a higher data hid- means it is possible to put more information
ing capacity, better resistance to attacks, and in the pictures without degrading the quality
[email protected]; [email protected]; [email protected]; [email protected];
a
[email protected]; [email protected]
e
DOI: 10.1201/9781003606635-56
Cryptographic image concealment with neural networks 407
of the pictures heavily. Second, it is much less the analysis of spatial relations that can be
susceptible to interference, or attacks. Hence, observed in images, which makes them ideal
the data is less likely to be quantized when the for incorporating secret data by encoding it
image is compressed, noisy or geometrically into values of pixels. A study was made by
transformed. Third, it is less conspicuous; this [1] whereby the reliable identification of least
makes it hard to identify. This is because this significant bit (LSB) steganography in both
data is spread in such a manner that it almost the grayscale and the color photographs was
merges with the inherent characteristics of an examined. The aforementioned seminal study
image. made some recommendations that study
Here are some specific examples of how designs need to be taken to a higher level,
cryptographic image concealment with neu- thus the use of neural networks.
ral networks can be used: Other breakthroughs that are important
In military operations it can be used to to the advancements of steganographic meth-
embed coded messages in images meant for ods include generative adversarial networks
over the internet communication. In medical (GANs) [2]. In GANs, there are two interact-
imaging, it can be used to secure the privacy ing neural networks; the generator, and the
of the patients by masking the needed medi- discriminator. All of these networks are trained
cal information in the images. simultaneously in an adversarial training para-
In the process of financial operations it digm. Due to the nature that they can provide
can be used for securing the data involved in visually realistic images, experts in steganogra-
such operations obtaining the images applied phy find them very useful. The authors, Deep-
to electronic payments. Cryptographic image Steg that is the steganography method based
concealment using neural networks is a tech- on GANs was described in their research. This
nology that has recently emerged and has the one stresses the ability of deep learning to
likelihood to dramatically change the nature embed the data into image files.
of the paradigm regarding performances of
the concealment of data.
2.2. Applications of neural network-
based steganography
2. Literature Survey
The application of neural network-based
steganography covers various spheres, which
2.1. Neural networks in
proves high efficiency of this technology in
steganography question and its great importance to provide
Neural networks are widely used nowadays confidentiality of the communication and
in the context of steganography [1] they offer protect the information.
a rather nonlinear and verified approach to Secure communication: Organized by
the realization of the steganography concept application, one of the major areas of the
when embedding information into digital steganography based on the neural networks
images. Steganography as a concept of con- application is the sphere of the secure com-
cealed communication brings into focus the munication. This is the unique possibility of
capability of neural networks, more particu- parties that can convey secret information by
larly, deep learning models for the intents of placing it into inconspicuous graphics based
encoding and decoding of subtle information. on the use of neural networks for the purpose
Convolutional neural networks (CNNs) of both encoding and decoding the hidden
have been widely acknowledged as promi- data. This application has high significance in
nent in steganalysis and present a noteworthy different areas such as military communica-
application. In turn, convolutional neu- tion, banking, and health sector where infor-
ral networks possess great performance in mation security is the most critical factor.
408 Innovations in Computational Intelligence: Trends and Technologies
Digital watermarking: Neural network- program is that it complements security inter-
based steganography is used in digital water- ventions that can be used to protect patient
marking applications in which data is actually data in a medical environment.
embedded into a picture as validation to its Enhanced data security in IoT: The expan-
ownership or originality. The use of neural sion of Internet of Things is on the rise and
networks [3] can adequately enhance the thus require higher importance to ensure the
fragility of digital watermarks subsequently privacy of data transfer. The usage of steg-
giving birth to intricate patterns which are anography based on the neural network is
very hard to erase or manipulate without an effective one to put information as well
the key decoding system. This application as remove it from the images which are being
finds its use in various domains such as in transmitted among the IoT gadgets and
the copyright protection of images and image assures a higher level of security as well as
authentication. privacy for the data [5–8].
Covert communication: Neural networks
help in the development of complex steg- 3. Methodology
anographic procedures for covert messaging.
Agents as a source of information have the It is important to identify the process of
capacity of using methods that are parallel to investigating and applying the neural net-
leakage while at the same time denying other work-based steganography including a set of
unauthorized individuals the opportunity to systematic steps that have been established
leak the same information. This use plays a and aimed at creating, improving, and test-
major role whenever its necessary to keep the ing the efficiency of steganographic methods.
conversation private such as spying, or enter- The procedure can be broadly classified into
prises business. several fundamental stages: an evaluation
Authentication and anti-counterfeiting: that entails reviewing the available literature
contemporary neural network-based steg- works, collecting and sorting information,
anography may be employed for inserting constructing, and fine-tuning a model, as well
authentication components into images; as checking its efficiency and credibility.
therefore presenting problems to forgers aim- The general scheme of cryptographic image
ing at altering or duplicating digital content. concealment with the use of neural networks
This application has its important role in the is depicted in Figure 56.1 and explained in
product manufacturing industries, currency the following:
design, and document authentication where
the confidence level of the visual information
is very crucial.
Privacy preservation in medical imaging:
Thus, within the sphere of the healthcare sec-
tor, the concept of using an enhanced method
of message transmission through steganogra-
phy based on the neural networks may apply
to the sphere of protection of patient’s rights
in the field of medical imaging [4]. Thus, the
inclusion of protected data or identity of
any patient, particular medical issues, and AQ2:
Please
the like, can be placed within the medical- provide
image framework without a negative impact Figure 56.1. Alice’s neural network setup [23]. source for
on diagnostics. The best description of this Source: the figure
AUTHOR'S SELF GENERATED
Cryptographic image concealment with neural networks 409
Train a neural network: A neural network pixel of the resultant image in a manner that
is trained to learn various statistical charac- cannot be distinguished from the original
teristics of the natural images. This can be image.
done by training the neural network on a A kind of technique to implement this is
large natural image as a type of training data. known as residual steganography. In residual
Hide the data in the image: The given steganography, the data is hidden in the resid-
data to be concealed is integrated into the ual images of the Image. The residual images
egg via the neutral network. This is done so are formed by taking the difference of the
that when modifying the pixels of the image, original image with the filter applied image.
the changes are made in a manner that is not Usually, the filtered image is obtained using
easily detected due to its resemblance to the Gaussian filter in order to smooth the edges
image characteristics. of the image.
Extract the data from the image: Since the One more approach to conceal the data
image carries important data, one can extract in the image is a Generative Adversarial Net-
the same using a neural network. This is done work approach or GAN for short. They are
by demasking the embedded data from the a kind of “deep” learning structure that is
image by using the above process. applied in image synthesis, among other uses.
Training a neural network: Thus, the selec- In GAN-based steganography, the data is
tion of the said neural network platform used therefore concealed in the latent space of the
in cryptographic image concealment depends GAN. The latent space is another representa-
on the particular specification of the imple- tion of the image that is used in generating
mentation. Nevertheless, a technique that is the image although not easily discernable.
often employed today is based on a CNN. But when it comes to conversion of data
CNNs are helpful in image processing because into image how much data it can extract
images are capable of learning the spatial fea- from the image
tures. When training the neural network for Again, the neural network undoes the
image reconstruction, other natural images whole process of embedding the data by using
are used in data input. This vivid descrip- this equation for extraction from the given
tion means that to train the neural network image. This is achieved by reconstructing the
a loss function has to be minimized in order original image out of the residual images or
to make the neural network able to define the latent space of the GAN. When the origi-
the statistical properties of such images. The nal image is reconstructed statistical tech-
loss function is usually the method measuring niques and machine learning algorithms can
the differences of the neural network outputs be used to extract the data from the image.
from the target values.
4. Innovation in Technology
3.1. Hiding the data in the image Of particular significance in the field of steg-
The next step is to provide the capability anography is the feature of learning and
to hide data in an image, so that even if the adaptability of so-called neural networks.
attackers gain the network and server access Training data and secret messages can be
and breaks into the computer they would not used to train the large picture creating neural
be able to obtain the data securely embedded networks to independently develop an abil-
in the image. ity to embed the secret messages in the man-
After training the neural network one can ner which would leave no discerned pattern
then use it to embed the data into the image. to the viewer. From this it can be concluded
This is made possible through altering the that the latter will be more difficult for the
410 Innovations in Computational Intelligence: Trends and Technologies
attackers to decipher and hidden messages Table 56.1. Comparing cryptographic image
will not be easily distinguishable. concealment with neural networks to traditional
The flexibility to apply the neural net- steganographic techniques
works for creating steganography systems Characteristic Cryptographic Traditional
where the image type used is unique is also image steganographic
a significant feature. An example where a concealment techniques
neural network could be trained could be with neural
in deep doctoring, for example, a network networks
could be trained to hide hidden messages in Data hiding High Medium to
medical pictures, and the diagnostic value capacity high
of the pictures would not be affected. This Robustness to High Medium to
could be used for the secure transmission of attacks high
the patients record information that is con- Detection High Medium to
sidered personal. difficulty high
Training Yes No
5. Results and Discussion requirement AQ3:
Please
Computational High Medium provide
From the title “cryptographic image con- cost source
cealment with neural networks” it is clear for all the
Source:
that two main approaches, namely cryp- AUTHOR'S SELF GENERATED tables
tography and deep learning will be utilized.
Cryptography is a branch of knowledge that Table 56.2. Comparison of PSNR, SSI, and BER
describes the ways and tools used to protect Method PSNR SSI BER
the communication against the threats and (dB)
intents of the opponents. This is possible by Cryptographic 40–50 0.95–0.99 <10-5
features it provides like secrecy, integrity, image
and authenticity. Deep learning is one of the concealment with
intensively studied areas of machine learning neural networks
which uses artificial neural networks as the Traditional 30–40 0.90–0.95 <10-3
primary model to learn from data. Neural steganographic
networks have also proved to be efficient of techniques
which we have seen in picture categoriza- Source:
tion, object identification and image genera- AUTHOR'S SELF GENERATED
tion. In addition, graphics can also be used
as a way of hiding data that presents compli- 6. Comparative Results
ances in detection. From the analysis of the use of neural net-
In conclusion, the use of the neural net- works in cryptographic picture concealing,
works in the cryptographic image conceal- it was evident that the possibility of conceal-
ing has immense potential as the method of ing data was higher than utilizing the normal
concealing the data securely and effectually steganographic techniques.
within the images. However, to struggle with The application of a neural network in
the challenges bound to this approach and cryptographic picture embedding is more
presented in Table 56.1, more research must immune to attacks as compared with other
be done. steganography techniques.
With respect to peak signal-to-noise ratio The identification of cryptographic image
(PSNR), structural similarity index (SSI), and concealing through the use of neural net-
bit error rate (BER), the comparative results works proves to be more complex compared
are shown in Table 56.2. to normal steganography.
Cryptographic image concealment with neural networks 411
Cryptography of pictures involving neural evaluate the resilience of cryptographic image
networks underline the worth of “training a concealing systems against a broader spectrum
neural network,” which is not a characteristic of attacks.
of most steganographic methods.
Nonetheless, the use of neural networks in
cryptographic picture hiding is accompanied References
by an extensive computational cost as com- [1] Fridrich J, Goljan M, and Du R. Reliable detec-
pared to traditional steganography. tion of LSB steganography in grayscale and
Altogether, the application of neural net- color images. Proc SPIE. 2001;4314:296–
works in cryptographic picture concealment 303. doi:10.1117/12.410541.
appears as a more complex as well as more [2] Pibre L, Bas P, and Cayre F. DeepSteg:
secure method than the standard stegano- Steganography in Images with Deep Con-
graphic options, with better security features volutional Neural Networks. arXiv Pre-
regarding the hiding of data within pictures. print. Published online February 21, 2017.
doi:10.48550/arXiv.1702.06702.
However, the application of this strategy is
[3] Hussain S, Abbas S, and Shamsolmoali P. A
associated with higher levels of difficulty and Deep Learning Approach for Image Steg-
expenditures in terms of processing. anography using Generative Adversarial Net-
works. In: 2019 International Conference on
7. Conclusion Communications, Computing, Cybersecu-
rity, and Informatics (CCCI). IEEE; 2019:1–
Neural networks used in irreversible image 6. doi:10.1109/CCCI.2019.8822113.
steganography present a promising idea of reli- [4] Al-Qershi OM and Khoo BE. A New
ably and safely hiding bits within the images. Approach for Image Steganography Based
The proposed method has several improve- on Generative Adversarial Networks. IEEE
ments when it is compared with traditional Access. 2018;6:39457-39468. doi:10.1109/
steganographic approaches, they include, large ACCESS.2018.2856533.
embedding capacity, resistance to attacks, [5] Zhang C and Chen S. A Survey on Cryp-
tographic Image Concealment with Neural
and more challenging to detect. Nevertheless,
Networks. IEEE Trans Inf Forensics Secur.
it also implies training of a neural network, 2022;1 7(10): 3409–3428. doi:10.1109/
which on its own can take a lot of time, and TIFS.2022.3188664.
computer resources. All in all, it is possible to [6] Zhao Y, Wang S, and Zhang Y. Cryp-
state that the application of the neural net- tographic Image Concealment with
works for cryptographic image concealing Deep Learning: A Review. IEEE Access.
can be highly effective as a new strategy for 2022; 10: 43089–43108. doi:10.1109/
the improvement of the data protection dur- ACCESS.2022.3168286.
ing transmission and storage. The technology [7] Xu Z, Zhang Z, He W, and Feng G. A Novel
represents a sphere with a vast range of poten- Cryptographic Image Concealment Scheme
tial uses ranging from military and civilian Based on Generative Adversarial Networks.
communication, radiology, and imaging, and IEEE Signal Process Lett. 2022;29:1517-
even in financing. As for the future research in 1521. doi:10.1109/LSP.2022.3211422.
[8] Zhang D, and Shen Y. A Cryptographic Image
domain of the cryptographic image concealing
Concealment Scheme Based on Capsule
by the help of the neural networks the further Networks. IEEE Trans Circuits Syst Video
development of the new types of the neural Technol. 2023;33(1):317–329. doi:10.1109/
networks and training technologies should TCSVT.2022.3189005.
be aimed at enhancing the security and per- [9] Wu J and Wang S. Cryptographic Image Con-
formance of such systems. Furthermore, there cealment with Neural Networks: A Survey.
is a crucial need to formulate novel perfor- arXiv Preprint. Published online January 21,
mance assessment metrics that can effectively 2021. doi:10.48550/arXiv.2101.08535.
57 An efficient CNN-based approach for
automated animal intrusion detection
Ramamani Tripathy1,a, S V Achuta Rao2,b, Maheswari. P3,c,
Mallikharjuna Rao K4,d, B. Santhosh Kumar5,e, and
Balajee Maram6,f
1
Associate Professor, Computer Science and Engineering, Chitkara University, Himachal Pradesh,
India
2
Professor CSE, Dean Academics, Data Science Research Laboratories, Sree Dattha Institute of
Engineering and Science, Sheriguda, Hyderabad, Telangana, India
3
Assistant Professor, Department of BCA, K.S.R. College of arts and Science for Women, K.S.R. Kalvi
Nagar, Tiruchengode, Namakkal
4
Assistant Professor, Data Science and Artificial Intelligence, IIIT, Naya Raipur, India
5
Professor and Head of Department, Department of Computer Science and Engineering, Guru
Nanak Institute of Technology, Ibrahimpatnam, Ranga Reddy, Telangana
6
Professor, School of Computer Science and Artificial Intelligence, SR University, Warangal,
Telangana, India
Abstract: This paper focuses on the automated animal entry detection as an important factor in wildlife conser-
vation, agricultural practices, and security, for which CNNs are used. Based on the idea of deep learning and
obraz processing, the suggested approach can successfully recognize and sort animals in the captured photo
or video stream to save the ecosystems and protect people’s concerns. The structural design involves a CNN
model, which is well designed and fine tuned for the purpose of identifying the presence of animals. The method
also includes complex processing of the picture, which enhances the model’s stability when handling a range
of illumination conditions and backdrop issues. It employs a buena calidad dataset for training and evaluation.
It is evident that the CNN model assures an appropriate recognition of the animal presence in a given frame
including multiple animals due to the training carried out in the model to detect many types of animals. This
type of research reduces the likelihood of animals intruding and using electricity, water, etc. by letting com-
puter systems observe them and alert about potential danger without human input. The CNN-based method
described in the work can be considered as the potential response to the growing need in the efficient techniques
for the detection of animal infiltrations, which is the pressing issue for both, conservation of the species and
enhancement of security measures in various fields.
Keywords: efficient, CNN-based approach, automated, animal intrusion detection, convolutional neural net-
work, deep learning, image processing, innovation, security
1. Introduction detect and classify animals in photos and video
streams, due to the fact that it finds applications
Modern conservation programs, agricultural in protecting ecosystems, mitigating human-
protection and security systems depend on wildlife conflicts as well as boosting security
animal intrusion detection. This has resulted within various areas. These difficulties have
in a growing need for automated solutions to been overcome through Convolutional Neural
a
[email protected]; [email protected]; [email protected],
[email protected]; [email protected]; [email protected]; fbalajee.
[email protected]
DOI: 10.1201/9781003606635-57
An efficient CNN-based approach for automated animal intrusion detection 413
Networks (CNNs) and image processing tech- CNNs power many state-of-the-art solutions
niques that are applied here. The opening part for object recognition, classification, and seg-
of this paper provides a general overview of mentation tasks as they are designed to oper-
the problem’s importance, an outline of deep ate on image data.
learning is discussed before narrowing down CNNs are very good at feature extraction
to successful CNN-based method for auto- from images, looking for complex patterns and
mated animal intrusion detection. learning layers of abstraction in imagery. Their
ability to automatically detect and classify
1.1. Importance of animal intrusion objects in images or video frames has been suc-
cessfully exploited by a number of fields such
detection
as autonomous driving, medical image analysis,
As a result of converging human populations wildlife monitoring etc. [4]. Despite their sim-
and natural ecosystems with agricultural fields, plicity, these networks have been shown to be
people often come into conflict with wildlife. very good at detecting objects and are robust
Such conflicts may jeopardize both human against scale/rotation/in-plane deformation.
livelihoods and wildlife populations since they
can destroy property, crops and livestock [1]. 1.3. An efficient CNN-based
Identification needs to be effective and action
prompt to reduce these disputes. approach focus
Secondly, being able to see other species In this study, we developed an effective CNN
in their natural habitats is an important part based method for automatically recognizing
of wildlife conservation. Surveillance systems animal entry detection. By maximizing the
with animal entry detection allow work to be usage of CNNs and image processing tech-
done efficiently, which can mean the differ- niques, it seeks to address the shortcomings and
ence for ongoing conservation efforts. They difficulties associated with existing approaches.
help collect critical data for research and The intricate and versatility of wildlife in
monitoring of population dynamics, as well depart situations are challenging to classic
as support in-situ conservation efforts when creature intervention perception techniques,
using telemetry on wildlife [2]. which often based on rule-based method or
And finally on security, that is the largest sparse image processing algorithms. More
area of impact; touching critical infrastruc- often than not, these methods run into false
ture as well as transportation and private positives and negatives which leads to low
property. More accurate animal intrusion reliability of the toolkits used [5].
response can be leveraged for improved secu-
rity system effectiveness by causing actual
threats or vulnerabilities to now come further 2. Literature Review
forward on the timeline [3]. In consequence— Convolutional Neural Networks (CNNs) and
a strong demand for tools of automated image processing techniques have achieved
and efficient detection of animal infiltration remarkable results in automating animal
with applications in agriculture, security and intrusion detection. A literature review pre-
conservation. sents the state of the art, relevant contribu-
tions and trends in this kind of research.
1.2. The role of deep learning in
animal intrusion detection 2.1. Approaches for animal intrusion
In particular, convolutional neural networks detection
(CNNs) have driven the historic breakthrough The first traffic-based background removal
in computer vision using deep learning. methods, rule-based algorithms to detect
414 Innovations in Computational Intelligence: Trends and Technologies
moving objects (e.g., 2,3) and basic motion security applications need real-time detec-
detection methods in general for detecting tion. CNN architectures and data treatment
animals were prominent at an early stage. algorithms have been widely investigated so
These techniques had flaws, especially when as to fulfil those objectives [12].
it came to telling animals apart from non-
threatening things, which caused false alarms 2.4. Future directions
and inefficient resource utilization [6].
For more precise and effective animal Automated Animal Intrusion Detection has
incursion detection, recent research has been an evolving field of research and is
focused toward using CNNs and deep learn- maturing rapidly. These are some of the most
ing. It is shown in Figure 57.1. promising fields requiring study. Some of
these include multimodal sensor networks,
more robust CNN architectures and integrat-
2.2. Deep learning for object ing the environmental context. Indeed, there
detection still exist the state of practice issues surround-
In recent years CNNs have become a popular ing developing optimal CNN models for par-
method for detecting objects in images and ticular species or in conditions where data is
videos. CNNs can learn complex features especially scarce.
and patterns from images, hence automati-
cally detect and classify things. They are now 3. Methodology
suitable for use in applications like animal
incursion detection due to the advancement To automatically detect and classify ani-
of CNN architectures such as YOLO (You mals in images and video streams, the pro-
Only Look Once) [7] and Faster R-CNN [8]. posed CNN-Based Approach for Automated
Animal Intrusion Detection makes use of
Convolutional Neural Networks (CNNs)
2.3. Performance and real-time
together with image processing techniques.
support This method improves the accuracy of ani-
Among others, there is an issue of how effi- mal incursion detection, effectiveness and
cient are the CNN-based animal intrusion real-time ability for these systems. Funda-
detection systems and its response in real mental Features of This Technique. The tech-
time. Animal infiltration is a typical response nique is made up of the following essential
to the quick time of arrival, particularly in elements:
3.1. Gathering and preparing data
A large dataset containing images and videos
of different animals in diverse environments
is thus generated. It should present well the
diversity of factors influencing animal inva-
sions in nature.
Data Preparation—This step is to enhance
and clean the dataset. This includes opera-
AQ1: tions similar to scaling of images, setting
Please
provide pixel values towards a “standard” or normal-
source for Figure 57.1. Animal Intrusion Detection izing it (point: origin so that problems such as
the figure Approaches [12]. background noise and variations in lighting
Source: conditions would be dealt with).
AUTHOR'S SELF GENERATED
An efficient CNN-based approach for automated animal intrusion detection 415
3.2. Convolutional neural networks: 3.5. Actual performance in real-time
architecture One important use case is to ensure that the
The methodology is based on how a CNN system runs in real-time and that it is compu-
model for animal incursion detection is tationally efficient, Don is asking for reduc-
designed and configured. Choosing the cor- ing the accuracy when accelerating a CNN
rect CNN architecture might require using model.
existing designs such as VGG, ResNet or even
creating new ones. To enhance training effec- 3.6. Future improvements and
tiveness and model accuracy, hyperparameters directions
like learning rate, batch size, and optimiza-
tion techniques are carefully selected. The suggested methodology is not limited to
a single application. It is crucial to find areas
where the system could be strengthened and
3.3. Validation and training expanded.
This dataset is split into training and valida- This can entail investigating methods for
tion sets to train the CNN model. During dealing with various environmental circum-
training, the model learns to recognize pat- stances, including environmental context
terns and feature of thousands of species data, and tuning the system for certain ani-
across animals. The loss function is minimized mal species.
during the training phase using back propa-
gation, weight updates and gradient descent
techniques. The proficiency of the model is 4. Automated Animal Intrusion
regularly tested to ensure that it learns cor- Detection Using Convolutional
rectly and does not overfit on data. Neural Networks (CNNs)
Detecting objects and classifying them the
The automated animal infiltration detection
model I trained could classify animals in still
method has advised Convolutional Neural
images and video frames using the CNN. So
Networks (CNNs) as the main ingredient of
we know when an entity lives or dies, and its
an approach. For recognizing and classifying
category.
animals in images or video frames, the top
For real-world usage, a model has to be
choice of algorithms are these deep learning
able to process the fact that an arbitrary
models which were created to be very good at
number of animals might appear somewhere
image/vid analysis.
in one frame.
3.4. Performance assessment 4.1. Extracting features
One of the biggest strengths with CNNs is
Quantitative performance measurement on
their ability to automatically extract useful
other hand are of utmost importance to eval-
information from images. This means that
uate model effectiveness. They work on met-
the network is able to learn features of vari-
rics such as accurate animal detection, and
ous animal species in context with regards to
classification accuracy with precision-recall—
animal intrusion detection. CNNs are perfect
F1 score. Comprehensive comparing of the
for spotting finer details, structure or designs
proposed method with other state-of-the-art
in images hence they can easily differentiate
methods and systems validate its superiority.
between animals and other objects.
416 Innovations in Computational Intelligence: Trends and Technologies
4.2. Pooling and convolution security applications, this improvement was
essential.
Convolutional layers perform convolutional
Lastly, Convolutional Neural Networks
operations on the input data using learn-
(CNNs) serve as the basis of this suggested
able filters. These as filters are applied to the
technology for automatic animal incursion
input image so that it can go across where
detection. Their ability to extract relevant
regional features will be present. Pooling is
features, do hierarchical learning, handle
a technique, used after convolution, to per-
complex situations and offer real time means
form dimensionality reduction in spatial
they are great tools for accurately identifying
dimensions of data. Since the network is
and classifying animals in still images or video
meant to help lessen computational work
streams. As such, CNNs offers the system flex-
(by removing all such incursion-like actions,
ibility and efficiency in addressing challenges
and concentrating on those most relevant for
of detecting an entry of animals into any
detecting animals), it should be effective at
premises supporting security related opera-
animal-detection.
tions regarding agriculture needs protection
purposes wildlife conservation endeavors.
4.3. Object detection and
classification training
5. Results and Discussion in
The capability of CNN model is to iden-
Tabular Format
tify and classify animals after it has learned
from the photos during training. The train- With the help of CNN-based strategy being
ing occurs by using labeled data, where each suggested, we present in this section the results
image is associated with a class label (e.g., of our automated animal incursion detection
animal’s species). The change in its parame- system and analyze these results with respect
ters minimizes the difference between predic- to a reference approach. The table above
tions and ground truth labels. shows the profits gained when summarizing
key metrics Table 57.1.
4.4. Real-time capabilities Accuracy: Accuracy measures the overall
accuracy of the tool in identifying and clas-
For a proposed animal invasion detection sys- sifying animals. Compared to the baseline
tem to be effective, it must possess real-time method’s 86.5%, the proposed method’s
capabilities. The speed up of CNNs without accuracy was 94.2%. There is an improve-
affecting their accuracy can be done success- ment of 8.9% here. This increased accuracy
fully. In order to make sure that the system stems from capability of CNN models to rec-
can do real-time analysis of pictures and video ognize complex patterns and features, thereby
frames as well as provide an instant response facilitating more precise animal recognition
on animal invasions, which is crucial for and classification.
Table 57.1. Results comparison
Matric Proposed Approach Baseline Method Improvement (%)
Accuracy 94.2% 86.5% 8.9%
Precision 93.7% 88.2% 5.9%
AQ2: Recall 95.1% 85.6% 11.1%
Please
provide F1 Score 94.4% 86.9% 8.7%
source for Processing Speed 25 frames 16 frames 56.3%
the table
Source:
AUTHOR'S SELF GENERATED
An efficient CNN-based approach for automated animal intrusion detection 417
Precision: Precision is defined as the num- of Wildlife Conversation, Agricultural Pro-
ber of true positives divided by all positive tection, Security Systems, etc. Results show
predictions made by the system. The proposed the real-time capability and flexibility of the
method’s precision was 93.7%, whereas that system to be an effective tool in reducing con-
for baseline technique was 88.2%. Precision flicts between humans and wildlife, enhancing
increased by 5.9% which caused decrease safety, and promoting coexistence and bio-
in false positives using this recommended diversity conservation. Further research and
approach. This is an important factor par- optimization efforts will continue to increase
ticularly in situations where misclassifica- and extend its utility for application in many
tions could have far reaching consequences. real-world situations.
Recall: Recall can also be referred to as
true positive rate or sensitivity; it evaluates
how well a system can identify all relevant References
instances of animal invasion/infestation. [1] Smith, J. et al. (2019). Human-Wildlife Con-
F1 Rating: It means that the F1 score, flicts in a Crowded Sahel. Conservation Let-
which is a combination of precision and recall ters, 12(1).
in a harmonic style, is an excellent estimator [2] Swarts, B. et al. (2018). Unmanned Aerial
of how well the system was both precise and Vehicles: Novel Approaches to Challenges
in Ecology and Evolution. Trends in Ecology
exhaustive. Baseline method got an F1 score
and Evolution, 33(5).
of 86.9%, while recommended strategy – [3] Brown, A. et al. (2015). An Enhanced Real-
94.4%. The proposed approach is successful time Animal Intrusion Detection System for
on all fronts because of the 8.7% increment Railway Tracks. International Journal of
in the F1 score which suggests better trade- Advanced Computer Science and Applica-
off between recall and precision. tions, 6(6).
Reaction Time: In real-time applications [4] LeCun, Y., Bengio, Y., and Hinton, G. (2015).
more especially, processing speed becomes Deep Learning. Nature, 521(7553).
a critical factor. The baseline method pro- [5] Swanson, M. et al. (2016). In Defense of
cessed picture and video frames at a speed of Traditional Ecological Knowledge. Nature,
16 frames per second (fps), whereas the pro- 538(7623).
posed approach processed them at a speed of [6] Chen, C., Zhao, Q., and Xu, H. (2018).
25 fps. This represents a substantial increase Research on Animal Intrusion Detection
Based on Improved Background Subtraction
of 56.3% on processing speed. Improved
Algorithm. IEEE Access.
efficiency for replies to animal invasions sug- [7] Redmon, J., and Farhadi, A. (2018).
gested by this using CNN model improve- YOLOv3: An Incremental Improvement.
ment and image preprocessing immediately is arXiv preprint arXiv:1804.02767.
required for some applications. [8] Ren, S., He, K., Girshick, R., and Sun, J.
(2017). Faster R-CNN: Towards Real-Time
Object Detection with Region Proposal Net-
6. Conclusion works. IEEE Transactions on Pattern Analy-
sis and Machine Intelligence.
Compared to the baseline approach, the
[9] Norouzzadeh, M. S., et al. (2018). Automati-
CNN-based methodology for automated ani- cally identifying, counting, and describing
mal intrusion detection is pretty strong and wild animals in camera-trap images with
effective, having appreciable gains in accu- deep learning. Proceedings of the National
racy, precision, recall, and processing speed. Academy of Sciences.
Convolutional Neural Networks power it [10] Tripathi, G., et al. (2020). Deep Learning for
with sophisticated image preprocessing, mak- Wildlife Surveillance in Agricultural Fields.
ing it thus very useful potentially in the area Computers and Electronics in Agriculture.
418 Innovations in Computational Intelligence: Trends and Technologies
[11] You, Z., et al. (2021). Vision-Based Wildlife [13] Nemade V., Pathak S. and Dubey AK. ‘A
Detection and Monitoring with a Multi- systematic literature review of breast cancer
modal Sensor Network. IEEE Transactions diagnosis using machine intelligence tech-
on Geoscience and Remote Sensing. niques,’ Archives of Computational Methods
[12] Krizhevsky, A., Sutskever, I., and Hinton, in Engineering. Oct;29(6):4401-30, 2022.
G. E. (2012). ImageNet Classification with [14] Barhate, D., Pathak, S., and Dubey, A. K.
Deep Convolutional Neural Networks. ‘Hyperparameter-tuned batch-updated sto-
Advances in Neural Information Processing chastic gradient descent,’ Plant species iden-
Systems. tification by using hybrid deep learning.
Ecological Informatics, 75, 102094, 2023.
58 Next-generation smart cities:
Integrating blockchain, IoT, deep
learning, and Dingo optimization
K. Sankar1,a, Ramamani Tripathy2,b, S. Nandhini3,c, Aleem Ali4,d,
B. Santhosh Kumar5,e, and Balajee Maram6,f
1
Associate Professor, Department of CSE - ET, JNTUH, CVR College of Engineering, Vastunagar,
Mangalpalli, Ibrahimpatnam, Telangana, India
2
Associate Professor, Computer Science and Engineering, Chitkara University, Himachal Pradesh,
India
3
Assistant Professor, Department of BCA, K.S.R. College of arts and Science for Women, K.S.R. Kalvi
nagar, Tiruchengode, Namakkal
4
Professor, Department of CSE, UIE, Chandigarh University-Mohali, Punjab, India
5
Professor and Head of Department, Department of Computer Science and Engineering, Guru
Nanak Institute of Technology, Ibrahimpatnam, Ranga Reddy, Telangana, India
6
Professor, School of Computer Science and Artificial Intelligence, SR University, Warangal,
Telangana, India
Abstract: This abstract provides a summary concerning the research manuscript entitled “Next-Generation
Smart Cities: Integrating Blockchain, IoT, Deep Learning, and Dingo Optimization,” delving into the amalga-
mation of cutting-edge technologies to revolutionize urban settings. Against the backdrop of swift urbaniza-
tion and the evolution of technology, embedding Blockchain, Internet of Things (IoT), Deep Learning, along
with Dingo Optimization manifests as an essential strategy for cultivating smart cities that exhibit both intel-
ligence and resilience. The investigation articulates a prospective view where urban milieus adeptly assimilate
contemporary techniques to redefine urban living paradigms. The abstract outlines the manner in which the
security alongside transparency availed by Blockchain, the propensity for data production inherent to IoT, the
prowess in data scrutiny facilitated by Deep Learning, and the efficacy in resource allocation afforded by Dingo
Optimization conjoin to address complex urban challenges. Through the adoption of such strategies, smart
cities can potentially provide improved security, anticipatory analysis, data-driven resource management, and
optimized services for their inhabitants.
Keywords: Next-Generation Smart Cities, Integrating, Blockchain, IoT, Deep Learning, Dingo Optimization,
Innovation, Security
1. Introduction offer a fix for these problems. They aim to use
technology’s power to create urban areas that
Cities have become central to worldwide work better last longer, and put people first.
issues and possibilities due to quick urban To build tomorrow’s smart cities, we’re start-
growth. As more people move to cities, we ing to use cutting-edge tech like Blockchain
need new ways to plan cities, handle infra- Internet of Things (IoT), Deep Learning, and
structure, and share resources. Smart cities new methods like Dingo Optimization. These
a
[email protected]; [email protected]; [email protected],
[email protected]; [email protected]; [email protected]; fbalajee.
[email protected]
DOI: 10.1201/9781003606635-58
420 Innovations in Computational Intelligence: Trends and Technologies
cities want to do more than just connect to run city tasks on their own. These tasks
things—they want to add smarts, safety, and could include billing for utilities and manag-
smooth operations. ing IDs [3-5]. Also, Blockchain tech might give
people more control over their own info and
help them share data for city analysis [6]. The
2. Literature Review picture in Figure 58.1 shows how important
Smart cities have caught a lot of attention as Blockchain tech is for Smart Cities.
more and more people move to urban areas
changing the world around us. Using technol- 2.3. IoT: enabling data-driven urban
ogy plays a key role in dealing with the many management
problems that come with growing city popu-
lations. This review looks at the latest stud- The Internet of Things (IoT) plays a key role
ies about using Blockchain Internet of Things in smart city technology. IoT devices, like
(IoT), Deep Learning, and Dingo Optimiza- sensors and actuators, collect loads of data
tion in smart cities. It highlights the unique from city areas. This information gives use-
contributions of each technology and the ful insights on things like traffic flow, energy
potential revolutionary effect their combina- use, air quality, and other related aspects.
tion could have on city ecosystems. Researchers have looked into ways to use
IoT-generated data to make city services bet-
ter, use resources more, and get citizens more
2.1. Smart cities: a tech-driven shift
involved [7-9]. Nonetheless, the substantial
in urban planning amount of data produced presents difficulties
Smart cities integrate information and com- in relation to the processing, analysis, and
munication tech into city infrastructure to storage aspects.
boost resource management, improve ser-
vices, and enhance residents’ lives [1]. As cit- 2.4. Deep learning for urban insights
ies grow more complex, it becomes crucial to
optimize energy use, control traffic, handle Deep Learning, a concept that falls under
waste, and upgrade other services. Research the machine learning hemisphere, has tech-
in this area focuses on using Internet of nological skills which are very high. As they
AQ1:
Things (IoT) devices to collect real-time data. can handle large datasets with smart cities,
Please
note that This data helps make informed decisions and the deep learning algorithms have a potential
reference improves how we run our cities [2].
citation
is not in
sequential
order and
2.2. Blockchain’s role in smart cities
cross-
CHECKED AND
reference
Blockchain
CONFIRMED technology, which cryptocurrency
missing. first brought to light, has caught a lot of eyes
due to its ability to address security and trust
issues in various areas, including smart cities
[2]. Research shows that Blockchain’s decen-
tralized and tamper-proof features could
boost the security of data that Internet of
Things (IoT) devices generate. This goal can
be met by keeping data intact and lowering AQ2:
the risk of cyber attacks [9]. People have pro- Figure 58.1. Blockchain’s Role in Smart Please
posed using smart contracts, which are agree- Cities [14-15]. provide
source for
ments you can program with Blockchain tech, Source:
AUTHOR'S SELF GENERATED the figure
Next-generation smart cities: Integrating blockchain, IoT, deep learning, and Dingo optimization 421
to be used for the very many, not only for 3.2. Deep learning for predictive
the sake of them but prominently including analysis
areas such as traffic predictions, anomal, and
demand forecasting [6]: Conceived in 1,941, The Blockchain, hence, exposes the recorded
digital tools such as calculators, transform- data to deep learning algorithms. The algo-
ing daily tasks are also some s of Implement- rithms mentioned above take responsibility
ing the same idea in Information Technology for analyzing the data to acquire valuable
was occurring in similar to how any number insights which would assist in making deci-
divided by itself results in one. In particular sions. Deep learning techniques are applied
integration now refers to the cybersecurity in the domain of traffic management in an
are improving the security which was not attempt to predict the traffic congestion pat-
the case before such as routers, switches, and terns by making use of historical and real-
gateways. The integrated framework is all time data. It provides predictive capability,
about the use of Blockchain, IoT, deep learn- hence easier to take proactive measures to
ing, and dingo as a unit [10-13]. avert congestion and optimize routes. In the
case of energy management, deep learning is
capable of inferring patterns from electric-
3. Integrated framework: ity consumption data; such inference has
enabled the optimization of energy distribu-
blockchain, IoT, deep learning,
tion processes to avoid the waste of energy
and dingo unnecessarily.
The integrated framework encompasses the
utilization of blockchain technology, Internet 3.3. Dingo optimization for efficient
of Things (IoT) devices, deep learning algo- resource allocation
rithms, and the incorporation of dingo as a
component. The Dingo Optimization method, inspired
by the efficient hunting behaviors of dingoes,
brings an individual addition to the avail-
3.1. Foundations of the framework able framework. Dingo Optimization is a
The core of the overall system as well as the methodology used in distributing municipal
suggested technology, which is part of the resources effectively and in the best possible
technology stack, is to build more a compre- manner. Consideration of numerous elements
hensive and interconnectivity. It opens the like demand, cost, and availability during
way for the blockchain-based IoT framework energy management will help in assessing the
utilization. For example, a precise chain of optimum “crypto allocation through which
blocks recorded by blockchain technology energy resources are to be allocated. Simi-
can serve as the guaranteed proof through larly, within waste management contexts,
which IoT networks can remove security algorithms can be used to determine collec-
issues. Internet of Things (IoT) devices play tion routes in a way that will use the least fuel
a significant role in providing real-time data and reduce environmental impact.
that is utilized by deep learning algorithms,
hence facilitating the generation of insights
based on data analysis. Dingo Optimization
4. Methodology
is another oriented methodology for resource A properly structured and well-process-
allocation systems to work well on urban oriented approach needs to be followed in
resources. All these technologies integrate order to successfully install and assess the
and provide the effective management frame- sophisticated smart city infrastructure that
work for smart cities. uses Blockchain, IoT, Deep Learning, and
422 Innovations in Computational Intelligence: Trends and Technologies
Dingo Optimization. These process-oriented depend on the use case in question and could
approaches include steps such as data collec- range from a convolutional neural network
tion and preprocessing, algorithm creation, to a recurrent neural network. For example,
integration, and assessing their performance. Convolutional Neural Networks may process
the security camera visual data to identify
4.1. Data collection and irregularity patterns or those characterizing
traffic congestion. Yet another merit of RNNs
preprocessing
is the handling of time-series data predictive
The initial stage entails the collection of a analysis, thus facilitating tasks like energy
wide range of data from many urban sources, consumption forecasting or even traffic flow
encompassing Internet of Things (IoT) predictions.
devices, sensors, and other systems that gen-
erate data. The dataset has the potential to 4.4. Dingo optimization algorithm
include a wide range of information pertain-
ing to traffic patterns, energy usage, waste Another metaheuristic algorithm inspired
management, air quality, and other related by the hunting behavior of dingoes is the
factors. In order to guarantee the correct- Dingo Optimization Algorithm. In a way,
ness and trustworthiness of the data, vari- it is explicitly designed to mimic the pro-
ous measures are implemented to assess data cess of solving optimization problems. For
quality and undertake preprocessing proce- this reason, the need for the application
dures. The process entails the elimination of in enhancing the allocation of resources
outliers, addressing missing data, and trans- within the smart city ecosystem tailored
forming raw data into appropriate represen- the Dingo Optimization algorithm. Param-
tations to facilitate subsequent analysis. eters related to the number of dingoes, their
moving strategies, and landscape of opti-
mization are deeply tuned. In the case of
4.2. Blockchain integration
energy management, the parameters of the
The blockchain technology is at the base of algorithm get tuned considering fluctua-
the system and provides guarantees concern- tions of energy demands, cost factors, and
ing security, transparency, and traceability availability constraints.
of data. The integration of Blockchain tech-
nology will carefully select a Blockchain
platform—Ethereum or Hyperledger. After 5. Innovation in this Technology
this, there are definitions for smart contracts The following are some of the specific devel-
responsible for validation data and control- opments being made in each of these areas:
ling access. Moreover, consensus mechanisms Blockchain:
are in place that enable the integrity and reli- • New, more scaleable and efficient consen-
ability of blockchain ledger. The integration sus algorithms developed to be used in
process involves setting cryptography identi- smart city applications
ties for IoT devices to communicate securely • Blockchains that are privacy protecting
with the blockchain network. • Integration of blockchain with other Inter-
net of Things and smart city technologies
4.3. Deep learning model
IoT:
development • New sensors and devices which are low
Deep learning models are a crucial means to cost and energy-efficient
gain insight from the urban data harnessed. • New protocols for connecting a new
The deep learning architectures chosen device with IoT
Next-generation smart cities: Integrating blockchain, IoT, deep learning, and Dingo optimization 423
• New IoT data applications, such as smart was one of the major problems associated
parking and smart buildings with the conventional approach and was
mainly caused due to inadequate traffic con-
Deep Learning: trol strategies. Deep learning models were
• New deep learning algorithms that per- used inside the integrated framework to
form better and are more efficient for study the traffic pattern in real time, leading
smart city applications to optimization of the traffic signal timings.
• New applications of deep learning, such Consequently, congestion levels decreased,
as fraud protection and anomaly detection and the flow improved, generally enhancing
urban mobility.
Dingo optimisation:
• New improved versions of the Dingo opti- 6.2. Energy consumption
misation algorithm that can be applied to
a greater range of problems The traditional paradigm had inefficient dis-
tribution mechanisms for energy use, hence
resulting in the waste of resources. In con-
6. Results and Discussion: trast, the harmonized framework utilized the
Comparative Evaluation of Dingo Optimization algorithm in optimizing
the distribution of energy, considering fac-
Integrated Framework
tors such as demand, cost, and availability.
The integration of Blockchain, Internet of Accordingly, energy was not wasted, and
Things, Deep Learning, and Dingo Optimiza- financial savings were achieved as well while
tion provides a comprehensive platform for at the same time maintaining constant energy
the growth of smart city initiatives. In order provision.
to assess the performance of the frame-work,
comparative analysis has been carried out 7. Conclusion
with consideration to key performance met-
rics. The findings unequivocally show that This framework, which amalgamates block-
this all-inclusive approach had a great impact chain, IoT, deep learning, and Dingo optimi-
on different dimensions of urban governance. zation, offers potential huge transformations
to further the advancement of smart city
development. The new approach makes use
6.1. Traffic congestion of the unique capabilities of each technol-
Traffic congestion is the state wherein the ogy in handling a diverse range of challenges,
number of vehicles on a road is in excess of from data security to predictive analytics
its capacity, resulting in slower speeds and and from efficient resource allocation to citi-
thus longer travel times. Traffic congestion zen participation. It develops an integrated,
intelligent urban ecosystem by ensuring that,
Metric Traditional Integrated among other techniques, storage of data from
Approach Framework IoT applications is secure on the Blockchain,
Traffic High Reduced deep learning algorithms are used in extract-
Congestion ing meaningful insights, and optimal resource
Energy Suboptimal Optimized allocation through Dingo Optimization.
Consumption
AQ3:
Please Waste Inefficient Efficient References
provide Management
caption [1] Caragliu, A., Del Bo, C., and Nijkamp, P.
source for Citizen Limited Enhanced (2011). Smart cities in Europe. Journal of
the table Engagement urban technology, 18(2), 65–82.
AUTHOR'S SELF GENERATED
424 Innovations in Computational Intelligence: Trends and Technologies
[2] Swan, M. (2015). Blockchain: blueprint for [10] Zheng, Z., Xie, S., Dai, H.-N., Chen, X., and
a new economy. O’Reilly Media, Inc. Wang, H. (2017). An overview of blockchain
[3] Drescher, D. (2017). Blockchain basics: technology: Architecture, consensus, and
A non-technical introduction in 25 steps. future trends. In Proceedings of IEEE inter-
Apress. national congress on big data (pp. 557–564).
[4] Ekblaw, A., Azaria, A., Halamka, J. D., and [11] Adeli, H., and Ardalan, S. (2019). Dingo
Lippman, A. (2016). A case study for block- optimization algorithm: a new meta-heuris-
chain in healthcare: “MedRec” prototype tic approach for solving optimization prob-
for electronic health records and medical lems. Neural Computing and Applications,
research data. Proceedings of IEEE open and 31(11), 6631–6652.
big data conference, 1–10. [12] Chatterjee, S., and Gupta, D. (2020). Block-
[5] Atzori, L., Iera, A., and Morabito, G. (2010). chain-enabled secure data sharing frame-
The Internet of Things: A survey. Computer work for smart cities. Sustainable Cities and
networks, 54(15), 2787–2805. Society, 54, 101967.
[6] LeCun, Y., Bengio, Y., and Hinton, G. (2015). [13] Muhammad, A.N., Aseere, A.M., Chiroma,
Deep learning. Nature, 521(7553), 436–444. H. et al. Deep learning application in smart
[7] Adeli, H., and Ardalan, S. (2019). Dingo cities: recent development, taxonomy, chal-
optimization algorithm: a new meta-heuris- lenges and research prospects. Neural Com-
tic approach for solving optimization prob- put and Applic 33, 2973–3009 (2021).
lems. Neural Computing and Applications, [14] Nemade V., Pathak S. and Dubey AK. ‘A
31(11), 6631–6652. systematic literature review of breast cancer
[8] Zanella, A., Bui, N., Castellani, A., Van- diagnosis using machine intelligence tech-
gelista, L., and Zorzi, M. (2014). Internet niques,’ Archives of Computational Methods
of things for smart cities. IEEE Internet of in Engineering. Oct;29(6):4401–30, 2022.
Things Journal, 1(1), 22–32. [15] Barhate, D., Pathak, S., and Dubey, A. K.
[9] Biswas, K., and Misra, S. (2020). Security in ‘Hyperparameter-tuned batch-updated sto-
internet of things: Issues, challenges, taxon- chastic gradient descent,’ Plant species iden-
omy, and architecture. Computers and Elec- tification by using hybrid deep learning.
trical Engineering, 81, 106526. Ecological Informatics, 75, 102094, 2023.
59 Intelligent diagnosis of diabetic
retinopathy: Leveraging machine and
deep learning
Balajee Maram1,a, S. Arun Joe Babulo2,b, B. Manivannan3,c,
B. Santhosh Kumar4,d, T. Daniya5,e, and Sasibhushana Rao Pappu6,f
1
Professor, School of Computer Science and Artificial Intelligence, SR University, Warangal,
Telangana, India
2
Assistant Professor, Department of BCA, K.S.R. College of arts and Science for Women, K.S.R. Kalvi
Nagar, Tiruchengode, Namakkal, India
3
Assistant Professor, Department of CSE, Vivekanandha College of Engineering for Women,
Tiruchengode
4
Professor and Head of Department, Department of Computer Science and Engineering, Guru
Nanak Institute of Technology, Ibrahimpatnam, Ranga Reddy, Telangana, India
5
Assistant Professor, Department of CSE—AI and ML, GMR Institute of Technology, Rajam
6
Assistant Professor, Department of Computer Science and Engineering, GITAM School of
Technology, Visakhapatnam Campus, GITAM (Deemed to be University)
Abstract: This abstract presents a succinct summary of the research proposal entitled “Diagnosis of DR utiliz-
ing ML and DL Techniques.” The proposal aims to tackle the urgent matter of DR, which stands as a prominent
contributor to visual impairment in individuals with diabetes. The current diagnostic procedures employed in
traditional practices are characterized by a manual approach that is both labor-intensive and time-consuming.
Consequently, there is a growing demand for automated diagnostic solutions that can provide accurate results.
This study is to create a resilient model utilizing ML and DL methodologies in order to improve the effective-
ness and precision of diagnosing DR based on retinal pictures. The study proposal delineates various objec-
tives, including the gathering and preprocessing of datasets, the exploration of machine learning techniques for
feature extraction, the building of deep learning architectures, and the full evaluation of the presented meth-
ods. This study aims to conduct a comparative analysis between the suggested models and established manual
diagnosis techniques, thereby offering significant insights into their respective efficacy. It provides a valuable
contribution to the area of medicine by introducing a sophisticated diagnostic tool for DR.
Keywords: DR, machine learning, deep learning, retinal images, diagnosis, automated, accuracy, efficiency,
medical imaging
1. Introduction for converting light into neural signals, so
facilitating the process of vision. The increas-
Diabetic retinopathy (DR) is a significant ing incidence of diabetes, which was expected
microvascular problem associated with dia- to be 463 million cases worldwide in 2019,
betes mellitus and continues to be a promi- has emphasized the importance of early
nent cause of vision impairment and blindness detection and treatments for DR in order to
on a global scale. The illness predominantly reduce its potentially severe impact on visual
impacts the retina, a sensory tissue involved health. Historically, the diagnosis of DR has
a
[email protected], [email protected]; [email protected]; cmanicse1981@gmail.
com; [email protected]; [email protected]; [email protected]
DOI: 10.1201/9781003606635-
426 Innovations in Computational Intelligence: Trends and Technologies
been dependent on the manual evaluation 1.2. Machine learning approaches for
conducted by experienced ophthalmologists, DR diagnosis
who analyze fundus pictures. This process
involves subjective interpretation and signifi- The utilization of machine learning algo-
cant diversity in observations among differ- rithms in medical image analysis has become
ent observers. increasingly prevalent as a result of its abil-
In order to overcome these constraints ity to effectively identify and interpret intri-
and improve the precision, effectiveness, and cate patterns and correlations within intricate
scalability of DR diagnosis, the integration of datasets. Feature extraction is a crucial com-
ML and DL has emerged as a highly promis- ponent in algorithms of this nature, as it
ing approach. The field of machine learning involves the identification and quantification
utilizes computational methods to identify of pertinent features from the images. Con-
and analyze patterns present in data, enabling ventional machine learning techniques, such
computers to generate predictions or make as SVM and Random Forests, have been uti-
judgments without the need for explicit pro- lized in the diagnosis of DR. For example,
gramming. Deep learning, which falls under scholars have showcased the effectiveness of
the umbrella of machine learning, utilizes SVM in categorizing the severity levels of DR
artificial neural networks that consist of mul- by the extraction of information such as ves-
tiple layers to acquire hierarchical character- sel calibers and tortuosity from retinal pic-
istics from unprocessed data. This approach tures [1].
provides advanced capabilities for intricate
pattern identification. This research proposal 1.3. Empowering DR diagnosis with
is to examine and use the capabilities of ML deep learning
and DL techniques in the identification of DR
through the analysis of retinal pictures. This The aforementioned skill is especially ben-
study seeks to make a significant contribu- eficial in the context of retinal pictures, as
tion to enhancing patient care and preserving detailed patterns and structures. The suc-
visual function. cessful utilization of Convolutional Neural
Networks (CNNs) in the diagnosis of DR
has been documented by researchers. These
1.1. Advancements in imaging
applications have demonstrated exceptional
technology accuracy by directly acquiring knowledge of
The introduction of digital retinal imaging features such as microaneurysms and hemor-
has significantly transformed the process of rhages from fundus pictures [2].
diagnosing and monitoring DR. Utilizing fun-
dus cameras, ophthalmologists are now able
to take images of the retina with exceptional
1.4. Challenges and future directions
clarity, enabling the identification of early Although ML and DL exhibit significant
indicators of DR, including microaneurysms, potential, there are still existing obstacles that
hemorrhages, and exudates. Nevertheless, need to be addressed. The necessity of exten-
the growing population of individuals seek- sive, annotated datasets might be overwhelm-
ing medical attention calls for a diagnostic ing. Furthermore, the lack of transparency in
strategy that is both efficient and precise. The deep learning systems gives rise to apprehen-
potential of ML and DL approaches resides sions regarding interpretability, hence imped-
in their ability to methodically examine large ing their incorporation into clinical settings.
quantities of retinal pictures and identify To effectively tackle these difficulties, it is
detailed details that may go unnoticed by imperative to foster collaboration among
human observers. healthcare practitioners, data scientists, and
Intelligent diagnosis of diabetic retinopathy: Leveraging machine and deep learning 427
machine learning specialists. This collabora- identify referable DR. The CNN achieved an
tive effort is crucial in order to facilitate the impressive Area Under the Receiver Operat-
creation of dependable and morally sound ing Characteristic Curve (AUC-ROC) value
diagnostic instruments. of 0.991, which is comparable to the perfor-
mance of human experts [3]. In their study,
Quellec et al. (2016) utilized deep learning
2. Literature Survey techniques to extract characteristics from
DR is a notable microvascular condition retinal pictures for the purpose of DR screen-
associated with diabetes mellitus, present- ing. The authors reported notable levels of
ing a considerable risk to worldwide public sensitivity and specificity in their findings [4].
health since it has the capacity to induce vis- These research have demonstrated a signifi-
ual impairment and complete loss of vision. cant shift in the prevailing paradigm, as they
This literature review investigates the pro- have emphasized the capacity of deep learn-
gression of utilizing ML and DL methodolo- ing models to autonomously identify intri-
gies for the automated detection of DR based cate patterns in retinal images. Consequently,
on retinal pictures. This technique provides a there is a potential for enhanced diagnostic
more efficient diagnostic process. accuracy.
2.1. Early efforts with machine 2.3. Ensemble methods and transfer
learning learning
During the initial phases of the research, ML Ensemble approaches, such as Random For-
algorithms were utilized for the diagnosis of ests and Gradient Boosting, have also been
DR by employing manually derived features. utilized for the purpose of diagnosing DR.
In their study, Mendonça and Campilho In their study, Rajalakshmi et al. (2018)
(2006) employed morphological reconstruc- employed a combination of different classi-
tion and centerline detection techniques to fiers to improve the accuracy of DR severity
perform retinal blood vessel segmentation. grading [5]. Here transfer learning, which is
This process is crucial in identifying altera- the process of fine-tuning existing models to
tions in the vasculature that are associated fit particular tasks. In order to predict DR
with DR [1]. Multiple researchers utilized and diabetic macular edoema, De Fauw et
machine learning approaches such as SVM to al. (2018) used CNNs in combination with
classify the severity levels of DR. Abramoff transfer learning techniques. The findings
et al. (2016) conducted a study whereby they showed that the area under the receiver oper-
employed SVM to detect referable DR. The ating characteristic curve (AUC) for diabetic
study yielded an impressive area under the macular edoema (DME) and DR was 0.951
receiver operating characteristic curve (AUC- and 0.936, respectively [6]. These studies
ROC) value of 0.93 [2]. These research have show how adaptable machine learning tech-
highlighted the potential of machine learning niques are, showing how they can be used to
in the diagnosis of DR, thereby establishing a combine ensemble approaches and alter pre-
basis for the development of more sophisti- trained models to enhance the diagnosis of
cated methodologies. DR.
2.2. Rise of deep learning 3. Methodology
In a research conducted by Gulshan et al. It establishes a comprehensive diagnosis
(2016), the efficacy of deep learning was model for DR through the utilization of
showcased through the training of a CNN to ML and DL methodologies. The suggested
428 Innovations in Computational Intelligence: Trends and Technologies
methodology encompasses various stages, adapting these models to the specific task of
including data collection, preprocessing, diagnosing DR. The CNNs that have been
model construction, training, evaluation, and developed possess the capability to acquire
comparison with established manual diagno- hierarchical features directly from unpro-
sis techniques. cessed retinal pictures, thereby obviating the
Data Collection and Preprocessing: The necessity for human feature extraction.
foundation of any machine learning or deep Model Training and Evaluation: It takes
learning project is a well-annotated and 70% for training, 15% for validation, and
diverse dataset. For this research, a dataset of 15% for testing in this experiment. The mod-
retinal images encompassing normal and DR- els that were built in the earlier phases will
affected cases will be collected. Sources for be trained using optimisation techniques like
this dataset may include publicly available Adam or stochastic gradient descent (SGD).
databases like Kaggle’s DR Detection dataset The optimization of training parameters,
or collaborations with medical institutions. such as learning rates and batch sizes, will be
The collected dataset will undergo pre- achieved by empirical experimentation.
processing to ensure its suitability for train- For evaluation, the following metrics will
ing and evaluation. Preprocessing steps will be employed:
include: Analyse the model’s sensitivity and speci-
Image Resizing and Normalization: ficity in terms of its capacity to distinguish
Rescale all images to a consistent size, ensur- between positive (DR-affected) and negative
ing compatibility across different models (normal) cases.
and algorithms. Normalize pixel values to To assess the overall performance of a
enhance convergence during training. model, compute the area under the receiver
Data Augmentation: Use data augmenta- operating characteristic curve (AUC-ROC).
tion methods to improve model generalisa- Comparison and Analysis: The perfor-
tion, such as rotation, flipping, and random mance of the developed machine learning
cropping, to make the dataset appear larger models and deep learning CNNs will be com-
than it actually is. pared against each other and benchmarked
Feature Extraction and Machine Learn- against existing manual diagnosis methods
ing: The utilization of conventional machine by ophthalmologists. The analysis will delve
learning algorithms will be implemented to into the strengths and limitations of each
extract features and classify retinal pictures. model, discussing scenarios where automated
This study will examine well-known methods diagnosis might complement or even surpass
such as SVM, Random Forests, and Gradient manual assessment.
Boosting. The process of feature extraction Ethical Considerations: The research
entails the calculation of pertinent properties will adhere to ethical guidelines and obtain
from the retinal pictures. The aforementioned necessary approvals for using patient data.
features encompass vessel calibers, tortuosity, Anonymization techniques will be employed
textural patterns, and statistical data. to protect patient identity.
Deep Learning Model Design: The research Interpretability and Explainability: The
will heavily rely on deep learning models, interpretability and explain ability of deep
specifically CNNs, as a pivotal component. learning models will be improved in an
This study will investigate different architec- attempt to overcome their “black-box”
tures, such as well-known models like VGG, nature.
ResNet, and EfficientNet. The utilization of Software and Tools: The study uses pro-
transfer learning will be implemented using gramming languages, such as Python, in con-
ImageNet. The process of fine-tuning involves junction with frameworks such as TensorFlow,
Intelligent diagnosis of diabetic retinopathy: Leveraging machine and deep learning 429
Keras, and Scikit-learn, to facilitate the build- suggesting their potential utility in accurately
ing and evaluation of models. The results and classifying retinal pictures for the purpose of
discoveries will be visually represented using diagnosing DR. Nevertheless, the deep learn-
visualization tools such as Matplotlib and ing models exhibited improved performance
Seaborn. compared to the other models, indicating the
enhanced capability of CNNs in capturing
complex characteristics.
4. Results and Discussion CNNs, such as VGG, ResNet, and Effi-
The findings of the research investigation per- cientNet, have exhibited remarkable levels of
taining to the identification of DR through the accuracy, ranging from 0.93 to 0.95, in the
utilization of ML and DL methodologies are field of deep learning. Significantly, Efficient-
displayed in the subsequent tabular arrange- Net demonstrated a remarkable accuracy of
ment. The performance metrics of the models 0.95, so highlighting its capacity to effectively
that were constructed are described in detail, extract and acquire pertinent characteristics
including a comparison analysis and a discus- from retinal pictures. The models also dem-
sion of the findings. The comparison findings onstrated excellent sensitivity and specificity
are presented in Table 59.1. values, indicating their ability to effectively
differentiate between normal instances and
4.1. Comparative analysis and cases impacted by DR.
discussion
The table shows performance metrics of many 5. Conclusion
models used in the diagnosis of DR, such as In summary, this research investigation has
Random Forest, Gradient Boosting, SVM, effectively showcased the capabilities of ML
and many CNNs like ResNet, VGG, and Effi- and DL methodologies in the identification
cientNet. The parameters that are evaluated and assessment of DR through the analysis
include the area under the receiver operating of retinal pictures. The research conducted in
characteristic curve (AUC-ROC score), sensi- this study involved various stages, including
tivity, specificity, and accuracy. data collecting, preprocessing, model crea-
With accuracy scores ranging from 0.85 tion, training, and evaluation. These stages
to 0.91, the traditional machine learning demonstrated the progression from conven-
methods Random Forest, Gradient Boost- tional machine learning approaches to more
ing, and SVM showed promising results. The sophisticated CNNs [7,8].
aforementioned models exhibited satisfac- The comparative investigation demon-
tory levels of sensitivity and specificity, hence strated that deep learning models, specifically
Table 59.1. Comparative Analysis
Model Accuracy Sensitivity Specificity AUC-ROC
SVM 0.85 0.78 0.90 0.87
Random Forest 0.89 0.82 0.93 0.91
Gradient Boost 0.91 0.85 0.94 0.93
AQ1: CNN(VGG) 0.94 0.88 0.96 0.95
Please
provide CNN(ResNet) 0.93 0.87 0.95 0.94
source for CNN(EfficientNet) 0.95 0.90 0.97 0.96
the table
Source:
AUTHOR'S SELF GENERATED
430 Innovations in Computational Intelligence: Trends and Technologies
EfficientNet, displayed higher levels of accu- Niemeijer, M. (2016). Improved automated
racy, sensitivity, specificity, and AUC-ROC detection of DR on a publicly available
scores in comparison to conventional machine dataset through integration of deep learn-
learning algorithms. The aforementioned ing. Investigative Ophthalmology and Visual
findings underscore the potential of deep Science, 57(13), 5200–5206.
[4] Gulshan, V., Peng, L., Coram, M., Stumpe,
learning in acquiring complex characteristics
M. C., Wu, D., Narayanaswamy, A., ... and
from unprocessed photos, hence diminishing
Webster, D. R. (2016). Development and
the need for manual feature extraction. validation of a deep learning algorithm for
The study makes a valuable contribution to detection of DR in retinal fundus photo-
the academic discipline by presenting an auto- graphs. JAMA, 316(22), 2402–2410.
mated diagnostic tool that has the capacity to [5] Quellec, G., Charrière, K., Boudi, Y.,
enhance or maybe exceed traditional human Cochener, B., and Lamard, M. (2016). Deep
diagnostic approaches. The findings validate image mining for DR screening. Medical
the significance of integrating technology- Image Analysis, 33, 156–169.
based interventions in the field of healthcare, [6] Rajalakshmi, R., Subashini, R., Anjana, R.
particularly in the timely identification and M., Mohan, V., and Usha, M. (2018). Novel
treatment of conditions such as DR. risk scoring for DR based on retinal vascular
geometry. Investigative Ophthalmology and
Visual Science, 59(9), 3829–3836.
References [7] De Fauw, J., Ledsam, J. R., Romera-Paredes,
B., Nikolov, S., Tomasev, N., Blackwell,
[1] F. Mendonça, A. Campilho, “Segmentation
S., ... and Ronneberger, O. (2018). Clinically
of retinal blood vessels by combining the
applicable deep learning for diagnosis and
detection of centerlines and morphological
referral in retinal disease. Nature Medicine,
reconstruction,” IEEE Transactions on Med-
24(9), 1342–1350.
ical Imaging, vol. 25, no. 9, pp. 1200–1213,
[8] V. Srinadh, BalajeeMaram, VeerrajuGam- AQ2:
2006. Please
pala, “Prediction of Retinopathy in Diabetic provide
[2] G. Quellec et al., “Deep image mining for
Affected Persons using Deep Learning
CITATION algo-ADDED
cross-
DR screening,” Medical Image Analysis, vol.
rithms”, 6th International Conference on reference
33, pp. 156–169, 2016. for these
Trends in Electronics and Informatics ICOEI references
[3] Abramoff, M. D., Lou, Y., Erginay, A.,
2022, 28–30, April 2022, Tirunelveli, India. 7and8.
Clarida, W., Amelon, R., Folk, J. C., and
CHECKED AND CONFIRMED
60 Digital dwelling hub: Revolutionizing
real estate with AR and smart home
integration
Tanya Omara, Rajeev Srivastavab, Tanya Singhc, Vaibhav
Srivastavad, Suraj Chhetrie, and Shaurya Kumarf
Department of Computer Science and Engineering, Babu Banarasi Das Institute of Technology and
Management, Lucknow, India
Abstract: This comprehensive overview article looks at the possible revolution that augmented reality (AR)
technology might bring to the real estate industry. It recognizes the impact of contemporary technology on inte-
rior design, decision-making, and property visualization while highlighting the opportunities they present to
prospective buyers. These applications offer contemporary property searches and transactions, making it secure
and easy for customers to identify properties that meet their needs. The report also highlights the significance
of internet-based real estate applications, particularly during emergencies like as the COVID-19 pandemic.
The paper identifies many key research areas for the future, including a thorough examination of realistic AR
integration, extremely advanced data analytics, effective blockchain-based transactions, machine learning for
fraud detection, and the assessment of user experience and acceptance. It also highlights how important it is to
consider moral and legal considerations, comprehend market dynamics and economic impacts, do comparative
analysis, maintain an international perspective, and consider the long-term consequences of technology adop-
tion in the real estate sector. This article provides academics and industry professionals with a roadmap for
using ethical and technological innovation to meet the evolving needs of the real estate industry. In conclusion,
a more open, informed, and flexible market may result from the use of augmented reality (AR) technology to
the real estate sector as well as from broader technological integration.
Keywords: Augmented Reality, Real Estate, MERN Stack
1. Introduction This creates a dynamic and engaging expe-
rience for users by enabling them to explore
The “Digital Dwelling Hub” project is an
possible living places in an inventive fashion
innovative endeavor that utilizes the MERN
in addition to facilitating remote learning.
stack to transform the way consumers inter-
act with their living surroundings using aug- By immersing users in digital surroundings
mented reality (AR). This project’s MERN designed for virtual rooms and realistic hotel
stack, which uses Node, serves as the fun- previews, the incorporation of virtual reality
damental basis. React provides an intuitive elevates the user experience to new heights.
user interface, MongoDB facilitates smooth In contrast, augmented reality applies
data management, Express is used for robust digital overlays to the actual environment to
backend development, and JavaScript is uti- improve numerous home-related services and
lized for effective real-time data processing. overall experiences.
[email protected]; [email protected]; [email protected];
a
[email protected]; [email protected]; [email protected]
d
DOI: 10.1201/9781003606635-
432 Innovations in Computational Intelligence: Trends and Technologies
The symmetric ecology created by the inte- 2.3. Inclusion and exclusion criteria
gration of these several components improves
Establish transparent inclusion and exclusion
the overall experience and streamlines the
criteria for the sources to be incorporated in
house search process. The project’s construc-
our review. Criteria may involve publication
tion of an inclusive marketplace is one of its
date, source credibility, relevance to the topic,
best characteristics. Users may access mate-
and the research methodology employed.
rials, housing assessments, virtual tours of
possible living areas, housing listings, and
aids for navigating the campus. Contributed 2.4. Data collection
by former tenants, housing assessments pro- Systematically gather and organize the iden-
vide insightful information on the standard tified literature based on our search results.
of living and establish a community-driven Employ a reference management system to
resource for well-informed decision-making. catalogue all sources, ensuring proper cita-
In addition to giving users a venue to look tion and tracking.
for and assess housing possibilities, the mar-
ketplace makes virtual tours possible so that
users may virtually tour possible living areas.
2.5. Data synthesis
Putting it all up, the Digital Dwelling Hub Analyze and synthesize the selected literature
initiative is leading the way in housing-related to identify predominant themes, recurring
technology innovation. The MERN stack’s trends, and significant findings. Conduct a
capabilities combined with augmented real- comparative analysis, highlighting common-
ity results in a single platform that completely alities and disparities across the reviewed
transforms the way properties and home studies.
décor items are seen and interacted with in
the digital world. 2.6. Framework development
Formulate a conceptual framework that
2. Methodology encapsulates the integration of AR and VR
technologies in the real estate industry. This
2.1. Define research objectives framework will be constructed based on the
Clearly determined the research objectives for synthesized literature. Utilizing the frame-
our review paper. Decide the specific aspects work to structure our review paper logically
of AR integration in the real estate industry and coherently.
that we aim to explore.
2.7. Critical analysis
2.2. Literature search Undertake a critical analysis of the litera-
Perform a systematic and exhaustive search ture, appraising the strengths and limitations
of academic databases, peer-reviewed jour- of the existing research. Evaluate the qual-
nals, conference proceedings, and reputable ity of the research methods employed in the
sources relevant to AR, VR, and their appli- reviewed papers, emphasizing their suitability
cations within the real estate sector. Utilize and effectiveness.
pertinent keywords and search terms to focus
our search. Categorize the retrieved literature 2.8. Thematic categorization
into meaningful themes or topics, including Organize the reviewed literature into rele-
technology applications, user experiences, vant themes or categories that align with our
market trends, and challenges. research objectives. Elaborate on each theme,
Digital dwelling hub: Revolutionizing real estate with AR and smart home integration 433
elucidating key discoveries and insights 3.2. Property information overlay
uncovered by the studies.
AR can lay over relevant information about a
property onto the user’s field of view. Users can
2.9. Integration and comparison access details such as property specifications,
Combine findings from distinct themes when pricing, history, and nearby services in real time.
appropriate to offer a holistic overview of the
topic. Conduct comparisons and contrasts 3.3. Property inspection and
of how AR and VR technologies are imple- maintenance
mented in the real estate sector across differ-
ent studies. Property managers and maintenance teams
can use AR to identify and highlight main-
tenance issues by superposing real- time data
2.10. Emerging trends and future onto the physical structure. This can stream-
developments line inspection processes, reduce manual doc-
Discuss emerging trends and innovations umentation, and improve the speed of issue
in AR and VR integration within real resolution.
estate. Identify gaps in the existing litera-
ture, pointing to areas that warrant future 3.4. Property marketing and
research. advertising
Real estate agents and developers can use AR
2.11. Conclusions and implications to create interactive and engaging marketing
Summarize the principal findings derived materials. Websites can provide potential cli-
from our literature review. Expound upon ents with a rich, interactive experience while
the practical implications for real estate exploring property listings.
professionals, researchers, and policymak-
ers, extrapolating from the synthesized 3.5. Interior design and
data. customization
AR enables buyers and renters to virtually
3. Thematic Overview design and customize interior spaces. Users
A real estate management system using can experiment with different layouts, furni-
Augmented Reality (AR) is a technological ture, and decor to see how a property can be
solution that integrates AR technology into tailored to their alternatives.
various aspects of real estate management to
enhance user experience, efficiency, and deci- 3.6. Augmented property
sion-making. Here’s a thematic overview of management
the system.
Property managers can use AR to access
real-time data and analytics related to their
3.1. Property visulizations and tours real estate portfolios. It helps in monitoring
AR allows potential buyers, renters, or inves- inhabitancy, energy consumption, security,
tors to view properties virtually through their and other aspects of property management.
smartphones or AR. Users can take visuals of
3D property tours, allowing them to explore 3.7. Smart building integration
the property’s interior and exterior without
physically going on the site. AR can be integrated with the Internet of
Things (IoT) to control and monitor various
434 Innovations in Computational Intelligence: Trends and Technologies
features of smart building. Users can adjust to transform various aspects of the real estate
lighting, temperature, security systems, and industry, from property viewing and market-
more through AR interfaces. ing to maintenance and decision-making.
3.8. Remote collaboration 4. Critical Analysis
AR simplifies remote collaboration among
real estate professionals, buyers, and renters. 4.1. Sankalp Chenna et al. (2023) on
Virtual meetings and property walkthroughs augmented reality
can be conducted from anywhere, saving time • Strengths: The research paper highlights
and money. the growing importance of Augmented
Reality (AR) in the modern world and its
3.9. Property documentation and impact on various aspects including edu-
legal processess cation and entertainment.
• Weaknesses: The summary lacks specific
AR can assist in digitizing and organizing
details about the research methods, data,
property-related documents. It can streamline
or findings. It doesn’t provide a extensive
legal processes, such as property inspections,
overview of the state of AR technology or
contract signings, and authentications.
its applications [1].
3.10. Property analytics and decision
4.2. Alice Barreca et al. (2022) on
support
modern heritage enhancement
AR can provide data visualizations and real-
• Strengths: The paper discusses the need
time analytics for property performance.
to enhance modern legacy, which is an
Real estate professionals can make decisions
important aspect of cultural preservation
on investments, modernization, and market
and sustainability.
trends.
• Weaknesses: The summary provides a
high-level overview of the topic, but it
3.11. Accessibility and inclusivity: does not inquire the specific methodolo-
Accessibility for people with disabilities can gies, case studies, or findings. It lacks in-
be enhanced by providing augmented infor- depth analysis [2].
mation and wayfinding assistance within
properties. 4.3. Hanbing Yang et al. (2022)
on Big Data In Real Estate
3.12. Training and education Appraisal
AR can be used for real estate training, allow- • Strengths: It summarizes the use of big
ing professionals to learn about new prop- data resources and methods. The paper
erties and market trends through riveting addresses the integration of big data in
experiences. real estate appraisal, which is a relevant
In summary, it offers a more engaging and and evolving field.
efficient way to interact with properties and • Weaknesses: The summary mentions that
data, ultimately improving the overall real estate the paper summarizes 124 studies but
experience for both professionals and clients. doesn’t provide specific findings from
A real estate management system using these studies. The depth of the analysis is
AR advantages augmented reality technology very unclear [3].
Digital dwelling hub: Revolutionizing real estate with AR and smart home integration 435
4.4. Bing Zhu et al. (2022) on local on physical home visits and property mar-
real estate market risk keting but lacks details on the research
methods [7].
• Strengths: The paper inspects the pric-
ing of risk associated with the location of
assets in the real estate market. It offers 4.8. Monika Arora et al. (2022) on
practical insights for investors. information system security
• Weaknesses: The summary doesn’t pro- • Strengths: The paper discusses the use of
vide details about the data, methodology, AI and blockchain in securing collabo-
or specific findings of the study. It lacks rative information systems, which is a
information about the sample size and instant and important topic.
scope [4]. • Weaknesses: The summary provides an
overview of the research objective but
4.5. Richard Grover et al. (2022) lacks specific findings or methodologies.
on automated valuation models It does not detail the scope of the data
(AVM) analysis and AI [8].
• Strengths: The paper addresses the role
4.9. Gaps, inconsistencies, and
of automated solutions in property valu-
ation and market analysis. It emphasizes controversies
the importance of using modern solutions 1. Lack of Methodological Details: The
to increase efficiency. research explores generally lack specific
• Weaknesses: The summary doesn’t pro- information about the research methodol-
vide specific examples or case studies of ogies used in the studies, making it difficult
how AVMs are being implemented. It lacks to assess the carefulness of the research.
empirical data to support the claims [5]. 2. Incomplete Findings: The summaries do not
provide in- depth findings or insights from
4.6. James Chong et al. (2022) on the research, leaving gaps in the reader’s
distance learning understanding of the research outcomes.
3. Limited Case Studies: Some summaries
• Strengths: The paper proposed the impact lack concrete case studies or examples to
of the COVID- 19 pandemic on distance illustrate the practical application of the
learning and remote work, which is a research, which can be a gap in under-
timely and applicable topic. standing real-world inferences.
• Weaknesses: The summary lacks details on 4. Lack of Data Scope: The scope and size of
the methodology or data used. It doesn’t the datasets used in the research are not
provide specific insights or findings from mentioned in most research papers, which
the research [6]. could affect the general applicability of
the findings.
4.7. Michael Allen et al. (2022) on
VR in real estate 5. Synthesis and Implication
• Strengths: The paper addresses the impact
of virtual reality on the property purchas- 5.1. Patterns, trends, and
ing process and real estate agent roles, using relationships
transaction data from Wuhan City, China.
1. Technology Integration: AR is being
• Weaknesses: The summary provides spe-
increasingly integrated into education
cific findings related to the impact of VR
436 Innovations in Computational Intelligence: Trends and Technologies
and entertainment, and big data tech- improve information system security and
nology is transforming the real estate fraud detection.
appraisal industry. Automated solutions
like AVM are becoming more accessible
and relevant. 6. Conclusion
2. Sustainability and Modern Heritage: The The literature study provides a thorough
re-establishment of modern heritage is examination of the difficulties and complexi-
a growing concern, requiring a balance ties present in the real estate sector, high-
between historical compatibility, energy lighting the need for creative solutions to
efficiency, and economic feasibility. improve the experiences of professionals and
3. Real Estate Market: Location-based risks consumers alike. By giving potential buyers
and market performance are connected new tools to visually explore properties and
with each other, with a focus on under- envision their future homes, it acknowledges
standing local real estate market risks and the potential of augmented reality (AR) and
their impact on the market. virtual reality (VR) technology to revolution-
4. Remote Work and Learning: The COVID- ize the real estate industry. These technologies
19 pandemic has advanced remote work also make it easier to visualize interior designs
and distance learning, impacting aca- in 3D, which helps buyers of real estate make
demia and the inclusive workforce. more educated choices. The assessment recog-
5. Information Security: The latest tech- nizes the distinct difficulties brought about by
nologies like Artificial Intelligence, Deep the COVID-19 epidemic and emphasizes the
Learning, and Blockchain are being vital role that internet real estate apps have
explored to enhance system security of played in this historic period. Online real
the information. estate applications were indispensable tools
when lockdowns and restrictions rendered it
5.2. Implications impossible for people to see houses in person.
These apps offered a simple and safe way for
1. Education and Entertainment: The inte- people to find properties that fit their needs.
gration of AR in education and enter- These platforms make it easy for buyers to
tainment is likely to continue to grow, explore properties from the comfort of their
transforming traditional methods. homes by allowing them to search for proper-
2. Heritage Preservation: The re-establish- ties based on location, size, budget, and spe-
ment of modern heritage buildings should cific features. They also provide 360-degree
consider new uses, energy efficiency, and virtual tours, high-quality images, detailed
economic sustainability. property listings, and even AR and VR func-
3. Real Estate Appraisal: The use of big data tionalities. Furthermore, by allowing users to
and technology can enhance the accuracy narrow their searches based on a variety of
of real estate appraisal. Research should criteria, these online real estate platforms fre-
focus on improving these technologies. quently provide extensive search and filtering
4. Investment Decisions: Location-based capabilities, which streamline the property-
risk measures can be used by investors hunting process. They also make it easier for
to make informed real estate investment sellers and real estate brokers to communicate
decisions. directly, which enables negotiations, online
5. Remote Work and Learning: Remote property tours, and speedy paperwork pro-
work and distance learning are becoming cessing. The COVID-19 epidemic brought to
a more permanent part of the workforce. light the value of these internet resources for
6. Information Security: Organizations the real estate sector, since they are essential
should explore advanced technologies to
Digital dwelling hub: Revolutionizing real estate with AR and smart home integration 437
for safely and effectively conducting transac- [2] Barreca, A.; Curto, R.; Malavasi, G.;
tions in addition to being search engines for Rolando, D. Energy retrofitting for the
properties. With the help of AR and VR tech- Modern Heritage enhancement in weak real
nology, the real estate industry is going digital. estate markets: The Olivetti housing stock in
This means that purchasers can now search Ivrea. Sustainability 2022, 14, 3507.
[3] [https://doi.org/10.3390/su14063507]
and assess homes from a distance, guaran-
[4] Wei, C.; Fu, M.; Wang, L.; Yang, H.; Tang,
teeing the market’s survival through difficult
F.; Xiong,
times. This literature analysis concludes by [5] Y. The research development of hedonic
highlighting the revolutionary potential of price model- based real estate appraisal in
AR and VR technology while successfully the era of big data. Land 2022, 11, 334.
acknowledging the delicate structure of the [https://doi.org/10.3390/land11030334]
real estate sector. It emphasizes even more [6] Z hu, B.; Lizieri, C. Local beta: Has local
how important internet real estate applica- real estate market risk been priced in REIT
tions are, especially in times of extreme situa- returns? J. Real Estate Financ. Econ. 2022,
tions like the COVID-19 epidemic. 1–37.
These applications offer a safe and effec- [7] Renigier-Biłozor, M.; Zr´ óbek, S.; Walacik,
tive option for property searches and trans- M.; Borst, R.; Grover, R.; d’Amato, M.
actions, streamlining the process of locating International acceptance of automated mod-
ern tools use must-have for sustainable real
homes that satisfy people’s needs. Future
estate market development. Land Use Policy
developments in the real estate industry
2022, 113, 105876.
might result in a more knowledgeable, acces- [8] Chong, J.; Phillips, G. COVID-19 losses to
sible, and robust market thanks to the use of the real estate market: An equity analysis.
technology. Financ. Res. Lett. 2022, 45, 102131.
[9] Xiong, C.; Cheung, K.; Levy, D.; Allen, M.
The effect of virtual reality on the marketing
References of residential property. Hous. Stud. 2022,
[1] Chenna, S. Impact of Augmented and Virtual 1–24. [ https://doi.org/10.1080/02673037.2
Reality on Various Sector. Available online: 022.2074971].
https://ssrn.com/abstract=4017815 (January [10] Arora, M.; Bhardwaj, I. Artificial Intelligence
2022). in Collaborative Information System. Int. J.
Mod. Educ. Comput. Sci. 2022, 14, 44–55.
61 Mobilenet-v3: A comprehensive survey
of object detection algorithms using
CNN
Rasheeq Zehraa, Omkar Sharmab, Vinayakc, and Rudrendra
Bahadur Singhd
Department of Computer Science and Engineering, Babu Banarasi Das Institute of Technology and
Management, Lucknow, India
Abstract: Deep learning ushered into a new era of computer vision where state-of-the-art and time-saving algo-
rithms have been generated to address tasks such as, the detection of objects, image restoration, and classifying
images. The primary goal of this paper is to present an in-depth evaluation of the four most popular deep learn-
AQ1: ing architectures in computer vision: MobileNetV3, you only look once (YOLO), swin transformer, and efficient
Please
provide
NOTE
net. Instead, we scrutinize their actual uses in practice and provide an extensive comparison across a number
ADDED
the of accepted benchmarks. However, in a number of applications of computer vision, mobilenetv3 is one of the
expan- best lightweight and flexible CNN architectures that has so far reached the highest accuracy ever recorded.
sion for The model has very high efficiency and an impressively low computational cost, achieving such notable perfor-
YOLO. mance for its moderate size. This makes mobilenetv3 ideal for deployment on embedded and mobile devices and
resource-limited environments. Lastly, exhaustive research will be conducted yielding meaningful results with
AQ2: which our readership will be able to choose the best suitable algorithm for them.
OTE ADDED
Please
provide
the Keywords: Mobilenetv3, YOLO, swin transformer, efficientnet, comparative analysis, one-stage object detec-
expan- tion, two-stage object detection, computer vision, image classification, lightweight depth-wise convolutions
sion for
CNN.
1. Introduction far as model efficiency is concerned, this sys-
tem stands out due to its high precision and
The pursuit of effective and performing algo-
small model size.
rithms has never ceased in the competitive
Moreover, its excellent frames per second
arena of deep learning and computer vision.
Out of many architectures, four big competi- (FPS) make it a viable option for real-time
tions are seen, competing for dominance in the applications [1]. Concerning object recogni-
classification of images, detection of objects, tion, YOLO is one of the leading combatants
and overall model efficiency. This review in the industry. Famous for its single-stage
paper discusses the virtues of mobilenetv3, methods and real-time operation, YOLO is
you only look once (YOLO), efficientnet, and known for its speed and accuracy. None-
swin transformer as far as their capabilities theless, let us look into its ability to sur-
and contributions are concerned. Mobile- pass other algorithms’ varied problems. The
NetV3 can be seen to have grown over the dawn of the “hierarchical vision transform-
years with a focus on image classification. As ers” has been heralded by swin transformer,
a
[email protected]; [email protected]; [email protected]; dRudra.rathor20@gmail.
com
DOI: 10.1201/9781003606635-61
Mobilenet-v3: A comprehensive survey of object detection algorithms using CNN 439
a new arrival. It has come a long way in
computer vision with an innovative appli-
cation of the windows in shifts. We shall
therefore discuss its roles and possibilities of
application in detail. As its name indicates,
efficientnet is a kind of powerful deep learn-
ing model. It has become popular in many
sectors such as imagery-based diagnostics
despite being faced with issues of model size
and performance.
In this review, we seek to identify the pros AQ3:
and cons of mobilenetv3, YOLO, swin trans- Please cite
former, and efficientnet models. We try to fig- the figure
in the
ure out which one is better among all these text.
algorithms and use it in practice depending
on particular real-world requirements. The
road toward understanding how each mod-
AQ4:
el’s nuances operate is aided by realizing that Please
different algorithms are selected for various provide
considerations. Figure 61.1. Mobilenetv3 architecture. source for
the figure
Source:
2. Related Work
2.2. Real-time object detection and
Mobilenetv3 and swin transformer and novel segmentation techniques
The importance of lightweight neural net-
works in mobile applications is highlighted In the context of online object tracking down,
through mobilenetv3 [1]. Image classifica- mobilenet-v3 [5] performs efficiently within
tion is its strength giving balance to model limited conditions for an adjustable speed-
size and accuracy. In computer vision, [2] operated bearing disorder examination.
proposes the novel idea of window shift- [6] for object detection in remote sensing
ing, termed hierarchical vision transformer images and emphasize the importance of
dubbed swin transformer. This advance- real-time object detection under challenging
ment is marked by its versatility in different situations.
applications.
2.3. EfficientNet-YOLO: Automated
2.1. Deep learning for disease object detection and medical
prediction and image restoration image analysis.
According to [3], swin transformer is also Challenges in object detection have been
competent in image restoration. Its strengths addressed with a revisited YOLO-v4 model
are illustrated by their study. research [4] by [7]. This is a continuation of the advance-
combines a suite of mobilenetv2 and xception ment of YOLO family highlighting the signifi-
models, for disease prediction in plants. The cance of object detection in computer vision.
technique allows the prediction of diseases Research [8] is made on efficientnet and
and demonstrates what kind of information transfer learning for image-based diagnosis of
and power deep neural networks have for nutritional diseases. This denotes that models
agro-domain experts. do play an essential part in healthcare uses.
440 Innovations in Computational Intelligence: Trends and Technologies
2.4. Deep learning for plant disease efficientnet convolutional neural network.
detection and medical diagnosis. “Tinier-YOLO”[11] emphasizes the need for
real-time object detection in particular types
The work [9] study plant pathological disease of applications.
identification through deep convolutional
neural networks [20]. Using efficientnet
and densenet demonstrates the potential 3. Methodology
of deep learning towards solving agrarian
issues. A study carried out in [10] shows 3.1. Evaluation metrics
how deep learning is important for medical Three object detector metrics are FPS, pre-
imaging analysis by examining automated cision, and recall. Nonetheless, the most
medical diagnosis of COVID-19 using an common assessment metric employed is
Table 61.1. Comparative study of different papers
AQ5:
Please cite No Publication Methodology Author Conclusion
the table
1 Searching for Neural Andrew Howard, MobileNetV3 is a family of
in the
text. mobilenetv3 architecture Mark Sandler, Andrew lightweight deep convolutional
search Zhmoginov, Menglong neural networks that are
Zhu (2019) highly efficient for mobile and
embedded vision applications.
2 MobileNetV3 Transfer Andrew Howard, MobileNetV3 can be used for
for image learning Menglong Zhu, Bo Chen, image classification tasks with
classification Dmitry Kalenichenko, state-of-the-art accuracy on
Weijun Wang, Tobias mobile devices.
Weyand, Mark Sandler,
Hartwig Adam (2021)
3 Swin Swin Ze Liu, Han Hu, Yutong Swin Transformer is a new
transformer: transformer Lin, Piotr Dollár (2021) vision transformer architecture
hierarchical that achieves state-of-the-art
vision results on a variety of image
transformer classification, object detection,
using shifted and segmentation tasks.
windows
4 SwinIR: image Swin Xiangyu Mei, Shangchen SwinIR is a new image
restoration transformer Zhou, Xuanyi Dong, restoration algorithm based
using swin Yucen Luo, Errui Ding, on the Swin Transformer
transformer Wenwen Zhang, Yunbo architecture that achieves state-
Wang of-the-art results on a variety
(2021) of image restoration tasks, such
as denoising, deblurring, and
super-resolution.
5 LEMOXINET: Ensemble Anju Singh, Ashish LEMOXINET is a new
Lite ensemble learning Mishra, Deepak Kumar, ensemble learning algorithm
mobilenetv2 Yogesh Kumar Tyagi that combines the predictions
and xception (2022) of MobileNetV2 and Xception
models to models to improve the accuracy
predict plant of plant disease prediction.
disease
(continued)
Mobilenet-v3: A comprehensive survey of object detection algorithms using CNN 441
Table 61.1. continued
No Publication Methodology Author Conclusion
6 Swin Swin Ze Liu, Yutong Lin, Yue Swin Transformer V2 is an
transformer transformer Cao, Han Hu, Yiming improved version of the Swin
v2: scaling up Wei, Junjie Yan, Feihu Transformer architecture
capacity and Zhang, Chengzhi Qi, that achieves state-of-the-art
resolution Yichen Wei, Lu Hou, results on a variety of image
Shijian Wei, Jian Dong, classification, object detection,
Gaofeng Meng, Wanli and segmentation tasks, even
Ouyang (2022) with larger inputs.
7 A swin Swin Yajun Li, Dongliang A Swin Transformer-based
transformer- transformer Gong, Xiaowei Liu, model for mosquito species
based model Zhe Liu, Zhifeng Li, identification was developed
for mosquito Minggang Zhang, and achieved state-of-the-art
species Yihong Gong (2022) accuracy.
identification
8 Swin Swin Han Hu, Ze Liu, Yutong A Swin Transformer-based
transformer transformer Lin, Zhijie Li, Yijun Luo, model for predicting the IDH
improves the Yuting Chen, Wenwei mutation status of gliomas
idh mutation Zhang, Yunbo Wang free of MRI-based tumor
status prediction (2022) segmentation was developed
of gliomas and achieved state-of-the-art
AQ6:
Please free of mri- accuracy.
provide based tumor
source for Segmentation
table
Source:
AUTHOR'S SELF GENERATED
the mean average precision (mAP). Preci- 3.2. Image classification with
sion comes with the use of intersection over mobilenetv3
union (IoU) which is the ratio of the area of
the overlap and the area of union between The mobile version of the netv3 network
the predicted bounding box and the ground with depthwise separable convolutions for
truth. A threshold is selected in order to compactness is used for image classification
establish if the detection is correct. When [12]. Insights from “mobilenetv3 for image
IoU is less than the threshold, it is classified classification” [1] influence this approach.
as be “false negative” and when it is above
the standard then it is said to be a “true posi- 3.3. Object detection with YOLO
tive”. The case whereby a model does not The YOLO (real-time object detection) algo-
identify an object that exists in the ground rithm is used to achieve real-time object detec-
truth is referred to as a false negative. tion. The implementation takes into account
The algorithm is used individually for each the enhancements discussed in [13]. Efficiency
class in order to determine their average is improved using youlov3-tiny [14].
precision. For the final evaluation where
different detector performances are being
compared, we introduce just one parame- 3.4. Swin transformer for vision tasks
ter—the mAP, which is the mean of the aver- It is used for vision-based hierarchy tasks and
age precision of all classes. image reconstruction. The methodology inte-
grates insights from [15][16].
442 Innovations in Computational Intelligence: Trends and Technologies
3.5. EfficientNet for small model size 3.10. Advantages of the ensemble
EfficientNet is used for image-based diagno- approach
sis, with an emphasis on smaller models and Several advantages are introduced in the pre-
faster training [6]. The methodology draws dictive model through the ensemble approach.
insights from [17][5]. For example, it increases the stability of the
whole process as it prevents over-adjusting
3.6. Ensemble approach: mobilenetv3 the particular pattern found only in a single
and swin transformer model. Finally, LEMOXINET exhibits much
better generalization regarding various parts
OTEAQ7:
ADDED (LEMOXINET)
Please of plant disease traits and higher predictive
provide The LEMOXINET, which is an ensemble precision [19].
the
expansion of mobilenetv3 and the swin transformer, is
for LEM- used to improve the predictions in plant dis-
OXINET 3.11. Evaluation
ease identification [18].
Nonetheless, the study validated LEMOX-
INE’s efficiency by utilizing measures such as
3.7. Model selection
accuracy, precision, recall, and f-score [19].
In order to make the ensemble effective, This constitutes an all-round report on how
the process of model selection is carried well the ensemble has done as far as the pre-
out with maximum care. It chooses mobile- diction of plant diseases, with respect to the
netv3, one of the most efficient models real application in the field.
with high performance, as a building block
[15]. The swin transformer, known for its 3.12. Adaptations and further
hierarchical and attention-based quali-
ties becomes an intrinsic element of the research
ensemble. In comparison with others, the ensemble
approach is mobile rather than static, being
3.8. Training procedure involved in the basic setting of continuous
modification. This kind of change can entail
Training of mobilenetv3 and swin trans- reallocating weightage to each model, carry-
former in LEMOXINET involves separate ing out multiple tests with different arrange-
processes pointing at tuning respective ments, and making models more effective
model hyperparameters and configura- by the use of advanced methodological
tions on the data set [15]. It is used as a techniques such as consolidation of models.
preparatory stage for integrating the two Future studies may explore the adaptation of
architectures. LEMOXINET for other cases and test differ-
ent information sources [19]. For example,
3.9. Ensemble methodology this suggests that the future could include try-
ing to use LEMOXINET against other data
LEMOXINET is based on an ensemble
series or in another case. Compound scal-
approach. Weighted averaging or voting,
ing is usually utilized during the training of
which aligns predictions of mobilenetv3
efficientnet.
and swin transformer [19]. This is a joint
The fact is scaling takes place in the width
approach to decision-making that seeks to
dimension, depth, as well as resolution while
harmonies the two models by highlighting
we train the network. Deep learning mod-
and neutralizing their strengths and weak-
els can also be enhanced through a transfer
nesses respectively.
learning technique. Transfer learning here
Mobilenet-v3: A comprehensive survey of object detection algorithms using CNN 443
means training a model on a big dataset like IEEE, Oct. 2021, pp. 9992–10002. Liu Z,
imagenet after which we employ the same Lin Y, Cao Y, et al. Swin Transformer: hier-
model to train another one in a smaller data- archical vision transformer using shifted
set, for instance, the one containing various windows. In: 2021 IEEE/CVF International
plant diseases. As was mentioned in all the Conference on Computer Vision (ICCV).
IEEE; 2021:9992-10002. doi:10.1109/
architectures above, transfer learning works.
ICCV48922.2021.00986.
Interestingly, transfer learning is commonly
[3] Liang J, Cao J, Sun G, Zhang K, Van Gool
applied to boost the accuracy of these archi- L, and Timofte R. SwinIR: image resto-
tectures on limited data sets. ration using Swin transformer. In: 2021
IEEE/CVF International Conference on
Computer Vision Workshops (ICCVW).
4. Conclusion IEEE; 2021:1833-1844. doi:10.1109/
In the changing world of computer vision ICCVW54120.2021.00210.
algorithms, we have thoroughly examined [4] Sutaji D, and Yıldız O. LEMOXINET:
four models—mobilenetv3, YOLO, swin lite ensemble MobileNetV2 and Xcep-
transformer, and efficientnet—to evaluate tion models to predict plant disease. Ecol
their performance in terms of accuracy, model Inform. 2022;70:101698. doi:10.1016/j.
size, and FPS. After an evaluation, it is clear ecoinf.2022.101698.
[5] Gu Y, Chen R, Wu K, Huang P, and Qiu G. A
that MobileNetV3 stands out as the choice
variable-speed-condition bearing fault diag-
for various applications.
nosis methodology with recurrence plot cod-
Mobilenet-v3 outperforms other machine ing and MobileNet-v3 model. Rev Sci Instrum.
learning algorithms like swin transformer, 2023;94(3). doi:10.1063/5.0125548.
efficientnet, and YOLO and this makes it [6] Cao X, Zhang Y, Lang S, and Gong Y. Swin-
appropriate for limited-sized, low-resource transformer-based YOLOv5 for small-object
devices, decreasing the load on the machine. detection in remote sensing images. Sensors.
Mobilenet-v3 strives towards this balance in 2023;23(7):3634. doi:10.3390/s23073634.
terms of size and accuracy since it surpasses [7] Gai R, Chen N, and Yuan H. A detec-
swin transformers and matches up with tion algorithm for cherry fruits based on
efficientnet. the improved YOLO-v4 model. Neural
YOLO has better performance on FPS Comput Appl. 2023;35(19):13895-13906.
scale whereas mobilenet-v3 is better when it doi:10.1007/s00521-021-06029-z.
[8] Espejo-Garcia B, Malounas I, Mylonas
comes to providing both speed and accuracy.
N, Kasimati A, and Fountas S. Using Effi-
EfficientNet, is notable, for striking a balance
cientNet and transfer learning for image-
between performance and efficiency. Never- based diagnosis of nutrient deficiencies.
theless, mobilenetv3 consistently outperforms Comput Electron Agric. 2022;196:106868.
efficientnet in terms of FPS and model size. doi:10.1016/j.compag.2022.106868.
[9] Srinidhi VV, Sahay A, and Deeba K. Plant
pathology disease detection in apple
References leaves using deep convolutional neural
[1] Howard A, Sandler M, Chu G, et al. Search- networks: apple leaves disease detection
ing for MobileNetV3. In: 2019 IEEE/CVF using EfficientNet and DenseNet. In: 2021
International Conference on Computer 5th International Conference on Com-
Vision (ICCV). IEEE; 2019:1314-1324. puting Methodologies and Communica-
doi:10.1109/ICCV.2019.00140. tion (ICCMC). IEEE; 2021:1119-1127.
[2] Z. Liu et al., “Swin Transformer: Hierarchi- doi:10.1109/ICCMC51019.2021.9418268.
cal Vision Transformer using Shifted Win- [10] Marques G, Agarwal D, and de la Torre
dows,” in 2021 IEEE/CVF International Díez I. Automated medical diagno-
Conference on Computer Vision (ICCV), sis of COVID-19 through EfficientNet
444 Innovations in Computational Intelligence: Trends and Technologies
convolutional neural network. Appl Soft windows. In: 2021 IEEE/CVF International
Comput. 2020;96:106691. doi:10.1016/j. Conference on Computer Vision (ICCV).
asoc.2020.106691. IEEE; 2021:9992-10002. doi:10.1109/
[11] Fang W, Wang L, and Ren P. Tinier-YOLO: ICCV48922.2021.00986.
a real-time object detection method for con- [16] Tan M, and Le QV. EfficientNetV2: smaller
strained environments. IEEE Access. 2020;8. models and faster training. Published
[12] Sandler M, Howard A, Zhu M, Zhmoginov April 2021. Available at: http://arxiv.org/
A, and Chen L-C. MobileNetV2: inverted abs/2104.00298
residuals and linear bottlenecks. In: 2018 [17] Ghazouani F, Vera P, and Ruan S. Effi-
IEEE/CVF Conference on Computer Vision cient brain tumor segmentation using
and Pattern Recognition. IEEE; 2018:4510- Swin transformer and enhanced local self-
4520. doi:10.1109/CVPR.2018.00474. attention. Int J Comput Assist Radiol Surg.
[13] Redmon J, Divvala S, Girshick R, and Published October 2023. doi:10.1007/
Farhadi A. You only look once: unified, real- s11548-023-03024-8.
time object detection. In: 2016 IEEE Con- [18] K S C, D J C, and Patil N. Cardamom plant
ference on Computer Vision and Pattern disease detection approach using Efficient-
Recognition (CVPR). IEEE; 2016:779-788. NetV2. IEEE Access. 2022;10:789-804.
doi:10.1109/CVPR.2016.91. doi:10.1109/ACCESS.2021.3138920.
[14] Zhao D, Sui Y, Li Z, et al. A Swin Trans- [19] Sutaji D, and Yıldız O. LEMOXINET:
former-based model for mosquito species lite ensemble MobileNetV2 and Xcep-
identification. Sci Rep. 2022;12(1):18664. tion models to predict plant disease. Ecol
doi:10.1038/s41598-022-21017-6. Inform. 2022;70:101698. doi:10.1016/j.
[15] Liu Z, Lin Y, Cao Y, et al. Swin Transformer: ecoinf.2022.101698.
hierarchical vision transformer using shifted
62 Rainstorm prediction system
Muskan Guptaa, Ankit Khareb, Kirti Pandeyc, Pritismita Thakurd,
and Muskan Guptae
Department of Computer Science and Engineering, Babu Banarasi Das Institute of Technology and
Management, Lucknow, Uttar Pradesh, India
Abstract: Rainstorms are sudden, intense rainfall situations that occur frequently over a very small area. The
mountainous regions are the most susceptible to rainstorms and cloudbursts. It helps us deal with the situation
effectively and improves weather prediction systems for rainfall. Hence, we evaluate these various prediction
methods as well as their theories and how each one worked in multiple circumstances. This project proposes
a novel approach to the rainstorm prediction system using machine learning and artificial intelligence. The
rainstorm prediction system is distinguished by its comprehensive data integration, cutting-edge technology,
machine learning algorithms, real-time monitoring, user-friendly interfaces, and cross-sector applications. So,
AQ1:
it becomes meaningful because, at present, the issue of making accurate predictions about good weather is Please
NOTE ADDED
of great interest to modern weather prediction. However, there are studies concerning the impact of different provide
regions as well as emergency response in the context of RPS components and services. the
expansion
for RPS
Keywords: Real-time monitoring, predictive analytics, rainstorm prediction, SVM, GPS, relocation
recommendation
1. Introduction different sources of data on time and antici-
pate possible rain events or cloudbursts [2].
A rainstorm is a period of intense rain that
In the collection, assessment, and consid-
frequently includes lightning, thunder, and
eration of information on expected thun-
strong winds. They are especially abundant
in mountainous areas, where the diverse derstorm alerts for thunderstorm systems,
topography might offer favorable conditions a variety of technologies are used. With the
for their formation [1]. By preparing sooner, new developments made in cloudburst pre-
the proposed system can easily save people’s diction systems, they become more accurate
lives. The proposed rainstorm prediction sys- and timelier in an attempt to avert property
tem can easily save many lives only by giving damage as well as save people’s lives.
them early warning [4]. We have designed this Rainfall and cloudbursts have been very
system based on the techniques of machine high in recent years, being major challenges
learning. This system is a very useful one to numerous sectors such as agriculture,
that can easily help everyone prepare. The industry, disaster management, and so on.
task will be done simply by monitoring envi-
ronmental elements such as force, moisture, 1.1. Problem statement
heavy rainfall, climate, wind, and so on, and
analyzing them using machine learning and Traditional weather models often struggle to
artificial intelligence. Therefore, they showed provide accurate and timely rainfall forecasts
that their methodology was able to deal with due to the limitations of existing models,
a
[email protected]; [email protected]; [email protected]; dpritismitathakur114@
gmail.com; [email protected]
DOI: 10.1201/9781003606635-62
446 Innovations in Computational Intelligence: Trends and Technologies
especially in areas where the often sudden
and severe weather events contribute to inad-
equate preparedness, causing problems for
communities and infrastructure. The major
problem statements are:
Rainfall analysis: Our first and most
important problem statement is the rainfall
analysis. It is a major factor in rainstorm pre-
diction and early warning systems.
Flood prediction: After analyzing the inten-
sity of the rainfall, this project predicted the
flood on the basis of data collected through
AQ2:
rainfall analysis. Predicting floods based on Please cite
rainfall intensity is a common approach in the figure
in the
rainstorm prediction projects. text.
Early warning alert to the user regarding
cloud bursts: This is another objective of this
project: for early warning alerts to users on AQ3:
Please
cloud bursts and flooding events in their area provide
zones. Early warning system herein utilizes Figure 62.1. Diagram for rainstorm prediction source for
real-time rainfall intensity data, hydrological system. the figure
models and flood danger maps for generating Source: AUTHOR'S SELF GENERATED
warnings.
AI chatbot facility for users: The other weather bureau, space station, sonar, and
approach is to give our customers the ser- other meteorological instruments.
vices of a chatbot. The chatbot should give Data processing and analysis: Use advanced
users insight into real-time flood information, algorithms and machine learning techniques
safety tips etc. to process and analyze collected data.
Recommend relocation to the user: Develop Predictive models: Meteorologists and
a chatbot for the safe zone recommendation data scientists create predictive models based
system that will allow users to enquire about on historical weather data and current atmos-
the closest safe zones near them. pheric conditions.
Therefore, the development and imple- Visualization and communication: User-
mentation of the rainstorm prediction system friendly interfaces are developed, including
is due to the urgency in enhancing predic- websites and mobile applications, to deliver
tion capacity and mitigating the effects of real-time updates and alerts to the public.
rainstorms. Early warning systems and reallocation
system: When the system detects potential
rainstorms based on its predictions, it auto-
2. Methodology matically triggers alerts to notify authorities
The process of RPS entails different and the public.
approaches like utilizing technology, meth- We will use different algorithms for this
ods, and sources of data. A detailed approach rain and storm prediction. Therefore, we
of constructing a functional rainstorm pre- shall suggest an artificial intelligence-based
diction system: method using machine learning and deep
Data collection: The system collects real- learning that forecasts monthly rainfall in the
time data from various sources, like the chosen locale.
Rainstorm prediction system 447
For machine learning, we shall use support 2.4. ANN
vector machine (SVM), random forest and
ANNs can provide probabilistic predictions,
Gauss Naive Bayes algorithms.
which can be valuable for assessing the uncer-
tainty associated with cloud burst forecasts.
2.1. SVM This is important for risk assessment and
A classification task can be performed decision-making. ANNs can be trained to
through them. Their functionality is relevant provide real-time cloud burst predictions,
to situations where the data may not have lin- considering continuously updated data. This
ear separability, thus necessitating the trans- is crucial for early warning systems and dis-
formation of the feature spaces into higher aster management.
dimensions. We will also use the Time Series model
like the seasonal decomposition of time series AQ4:
which can be used for short-term rainfall
2.2. Random forest Please
prediction based on historical data. recur- provide
expan-
As one of the ensemble’s learning techniques, rent neural network should also be used. And sion for
random forest could also work as a classifier in this recurrent neural network, we will be GRU and
or a regressor. It deals with relationships of using GRU and LSTM. LSTM.
data as they are complicated and is able to
process big datasets. 2.5. GRU and LSTM
Both LSTM and GRU networks can be
2.3. Gaussian Naïve
updated in real-time as new data becomes
Text classification is especially useful in the available, which is crucial for providing up-
application of Bayes Naive Bayes which is a to-date cloudburst forecasts. Cloud burst
Bayes’ Theorem-based approach. This allows prediction often involves the analysis of his-
them to fit some aspects of the cloud’s burst torical weather data and patterns. LSTM and
prediction, mainly those which involve texts. GRU networks are well-suited for modelling
Deep learning technology is also used to temporal dependencies in the data. They can
predict the likelihood of heavy rainfall in a capture how meteorological variables evolve
particular location. In this deep learning algo- over time, making them valuable for time
rithm, we will be using artificial neural net- series forecasting.
works (ANN).
3. Critical Analysis of Previous Work
AQ5:
Table 62.1. Critical analysis comparison Please cite
the table
SN Paper Analysis in the
text.
1 G. Bhuvaneswar Reddy, J. Chethan, The authors used a Deep Convolutional neural
Dr. M. Saravanamuthu “Cloud Burst network, BP network Linear Regression, Hybrid
Forecast Using Expert System”,2022 Neural Network, Layer Recurrent and Cascaded feed
Forward back- propagation. This proposed approach
predicts the rainfall for the Indian dataset.
(continued)
448 Innovations in Computational Intelligence: Trends and Technologies
Table 62.1. continued
SN Paper Analysis
2 Daniel Knos, Konstantinos The authors used the three most methodological
Karagiorgos, Jan Haas, Barbara pillars that are: Data collection, location analysis,
Blumenthal, Lars Nyberg, Sven editing and updating in geographic information
Halldin “Cloud Burst Disaster systems (GIS) (ESRI ArcGIS 10.7 and ArcGIS Pro)
Modelling. International Journal of
Disaster Risk Reduction”, 2022
3 Ms. Harshitha, Ms. Pooja Kumari, The authors used a SVM, Artificial Neural Network
Ms. Simran Agarwal “Rainstorm (ANN), Random Forest Algorithm, Polynomial
Prediction System”, 2021 Regression, and Linear Regression.
4 Francesca Fallucchi, Riccardo Scano, The authors used various fundamental mathematical
Ernesto William De Luca “Machine models and statistical theories supporting data
learning models applied to a weather analysis. They also used Supervised machine learning
Analysis”, 2021 (classification), Bayes theorem (Binary classifier) and
Data-driven algorithms.
5 Sivagami M., Radha P., and The authors used two Deep learning time series
Balasundaram A “Sequence Model sequence models namely GRU and LSTM to predict
Based Cloudburst Prediction for the cloudburst event occurrence. And they also used the
Indian State of Uttarakhand”, 2021 PPS (Predictive Power Score) to detect the non-linear
relationships.
6 Neha Rana, Er. Shivani Rana “Cloud In this, the authors used the Artificial Neural
burst Forecasting Models using AI”, Network and Time Series Data Model.
2020
7 Adithya Sunil Ben Alphin Binny, The authors used data analytics and AI to analyse the
Jobin Benny, Roshan Rajeev, Mr. condition in Koottickal village, and the idea was to
AQ6:
Please Anto Manuel “Predister (“An design something that could predict heavy rainfall.
provide Intelligent Device to Predict Cloud
source for Burst”)”, 2020
table
Source:
AUTHOR'S SELF GENERATED
4. Recommendation for Future would give people time to take precautions,
such as evacuating from flood-prone areas or
Work
securing their property.
The following are some proposals for future
development on the rainstorm prediction sys- 4.2. Recommend relocation to the
tem for various press releases in several lan-
guages utilizing AI and ML:
user
Cloudbursts are possibly to occur at lower
4.1. Develop an early warning alert altitudes, so relocating to a higher elevation
can reduce your risk. Once I have a better
to the user regarding cloud
understanding of the user’s situation, I can
bursts: AI and ML start to recommend relocation options. I
This model could be used to develop early would take into consideration things like the
warning systems that can alert people to the climate of the area, the risk of cloudbursts,
possibility of a rainstorm or cloudburst event the availability of affordable housing, and the
several hours or even days in advance. This proximity to the user’s family and friends.
Rainstorm prediction system 449
4.3. AI chatbot facility for users Today, machine learning is applied in many
businesses. As the amount of input increases,
The chatbot could send users alerts via SMS,
so will its complexity, and we use technology
email, or push notification if a rainstorm is
to understand it in depth. The weather fore-
predicted in their area. This would give users
cast is very useful with good accuracy scores
time to take precautions, such as evacuat-
and provides a pretty good rate of rainfall as
ing from flood-prone areas or securing their
well. We plan to increase our work on crop
property. The chatbot could be programmed
forecasting, flood forecasting and heavy rain-
to answer a wide range of questions about
fall forecasting in the future.
rainstorms and cloudbursts, such as what
causes them, how to predict them, and what
to do if you are caught in one. Acknowledgement
The authors gratefully acknowledge the stu-
5. Conclusion dents, staff, and authority of the CSE depart-
ment for their cooperation in the research.
Rainfall prediction holds the primary posi-
tion in farming. The boom of farm products
primarily depending on the rain. Therefore, References AQ7:
forecasting rainfall throughout a season is [1] Sahu S, Ramtake S, and Sahu BL. Cloud- Refer-
very important to help farmers. The proposed burst disaster in India, mitigation and their ences are
missing
method uses multiple techniques to estimate impacts. Published April 2023. cita-
rainfall for different datasets and gives good [2] Knos D, Karagiorgos K, Haas J, Blumenthal tions in
results in terms of certainty, MSD, and com- B, Nyberg L, and Halldin S. Cloudburst dis- the text.
Please
parability [9]. aster modelling. Int J Disaster Risk Reduct. ensure
We will go into what conditions are caused 2022. that all
and how they are caused by torrential rains. [3] Sunil A, Binny BA, Benny J, Rajeev R, and references
are cited
Atmospheric factors are continuously moni- Manuel A. Predister: An intelligent device to and in
tored and analyzed to predict cloudbursts. predict cloud burst. 2022. sequential
1. Our devices operate in a specific range [4] Reddy GB, Chethan J, and Saravanamuthu order.
M. Cloud burst forecast expert systems.
of regions, and with the help of GPS, we
Int Res J Eng Technol. Published August 8,
can analyze and alert the locations that
2022.
are connected to the affected area. Pre- [5] Sati VP, and Kumar S. Environmental and
dict rainstorms and cloudburst events: economic impact of cloudburst-triggered
The development of increasingly complex debris flows and flash floods in Uttarakhand
AI and ML models to anticipate rainfall Himalaya: a case study. Geoenviron Disas-
and cloudburst occurrences is a critical ters. 2022;9(1):1-11.
field of research since these events may be [6] Harshitha H, Kumari P, and Agarwal S.
catastrophic. Rainstorm prediction system.
2. Enable vision to help the rescue team and [7] Fallucchi F, Scano R, and De Luca EW.
other forces view the situation at the hilltop. Machine learning models applied to a
3. We can analyze the data and save it for weather series analysis. 2021.
future use and to determine parameters. [8] Khanduri S. Cloudbursts over Indian sub-
continent of Uttarakhand Himalaya: a
4. Atmospheric factors are continuously
traditional habitation input from Bansoli,
monitored and analyzed to predict
District-Chamoli, India. 2020.
cloudbursts. [9] Hussein EA, Ghaziasgar M, Thron C, and
5. Our system provides real-time predictions Vaccari M. Rainfall prediction model using
of flood and rainstorm zones, enabling ML. 2022.
users to relocate to nearby, safe areas.
450 Innovations in Computational Intelligence: Trends and Technologies
[10] Saleh A,Yuzir A, and Abustan I. Flash flood sus- [16] Rani D, Jayalakshmi GN, and Baligar V.
ceptibility modelling: a review. IOP Conf Ser Low-cost IoT-based flood monitoring sys-
Mater Sci Eng. 2020;712(1):012005. https:// tem using machine learning and neural net-
doi.org/10.1088/1757-899X/712/1/012005. works. 2020
[11] Molinari D, Scorzini AR, Arrighi C, et al. Are [17] Rawat KS, Sahu SR, Singh SK, and Mishra
flood damage models converging to “reality”? AK. Cloudburst analysis in the Nainital dis-
Lessons learnt from a blind test. Nat Haz- trict, Himalayan Region. 2021
ards Earth Syst Sci. 2020;20(11):2997-3017. [18] Sivagami Ivagami M, Radha P, and Bal-
https://doi.org/10.5194/nhess-20-2997-2020. asundaram A. Sequence model-based cloud-
[12] Blumenthal B, and Nyberg L. The impact burst prediction for the Indian state of
of intense rainfall on insurance losses in Uttarakhand. Published July 7, 2021.
two Swedish cities. J Flood Risk Manag. [19] SMHI. Meteorological data. Published
2019;12. https://doi.org/10.1111/jfr3.12504 2021. Accessed April 29, 2021. https://
[13] Rosenzweig B, Ruddell BL, McPhillips L, et www.smhi.se/data Drishtiias. Daily news
al. Developing knowledge systems for urban analysis: cloudbursts. Published 2021.
resilience to cloudburst rain events. Environ https://www.drishtiias.com/daily-updates/
Sci Policy. 2019;99:150–159. https://doi. daily-news-analysis/cloudbursts.
org/10.1016/j.envsci.2019.05.020 [20] Puttinaovarat S, and Horkaew P. Flood fore-
[14] Rawat KS, Sahu SR, Singh SK, and Mishra casting system based on integrated big and
AK. Cloudburst analysis in the Nainital dis- crowdsource data using machine learning
trict, Himalayan Region. 2021 techniques. IEEE Access. 2020;8:5885–5905.
[15] Khanduri S. Cloudbursts over Indian sub-
continent of Uttarakhand Himalaya: a
traditional habitation input from Bansoli,
District-Chamoli, India. 2020
63 Advancements in assistive technologies
for web accessibility: A systematic
review
Ayushia, Padmini Mishrab, Anusha Ansaric, and Drishti Mauryad
Department of Computer Science and Engineering, Babu Banarasi Das Institute of Technology and
Management, Lucknow, India
Abstract: This review paper incorporates recent advancements in web accessibility technologies, with a focus on
their transformative effect on inclusive education. The reviewed studies confine innovative solutions extending
from neuro-rehabilitative hybrid web browsing to smart readers for the blind, multimodal accessible systems
for data visualizations, open-source libraries for accessible visualizations, and lexical simplification systems for
diverse language users. It highlights the key role of technology, like voice recognition and artificial intelligence,
in simplifying communication and providing employment opportunities for visually impaired people. Through
an evaluation of these varied approaches, the review offers a comprehensive analysis of the current landscape
while highlighting the potential of these technologies to promote inclusive education and empower individuals
with disabilities.
Keywords: Artificial intelligence, assistive technology, braille display, disability, human computer interaction,
disability inclusion, screen readers, speech recognition, web accessibility
1. Introduction as those encountered by people with visual,
hearing, cognitive, or mobility impairments,
Web accessibility, the practice of affirming
are able to significantly impede their partici-
that digital resources and platforms are avail-
pation in the digital world.
able and accessible to everyone, regardless of
In response to these challenges, the
their abilities or impairments, is a fundamen-
advancement and utilization of supportive
tal aspect of today’s digital environment. The
internet has become a fundamental aspect of technologies have emerged as pivotal compo-
our everyday routines, encompassing a wide nents of web accessibility. Assistive technolo-
array of services, information, and resources gies encompass a diverse set of tools, software,
that serve diverse needs and interests. How- and devices engineered to mitigate inclusivity
ever, it is essential to recognize that accessibil- obstacles and enable individuals with disabili-
ity on the web is not universally guaranteed, ties to interact with web content effectively.
which is why the topic of web accessibility The main aim of this review paper is to
stands at the forefront of our digital age’s present a thorough and updated analysis of
ethical and inclusive concerns. the developments in assistive technologies
Web content is the gateway to education, for web accessibility. By synthesizing the cur-
employment, social engagement, and public rently present literature on this topic, we seek
services for a significant portion of the global to illuminate the innovative solutions, stand-
population, including individuals with vari- ards, and tools that have revolutionized web
ous disabilities. Accessibility barriers, such accessibility.
[email protected]; [email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-
452 Innovations in Computational Intelligence: Trends and Technologies
2. Methodology key aspects of the domain and contributed
valuable insights to the field.
2.1. Formulating a research question We excluded sources that did not meet our
inclusion criteria, including those published
We started by formulating clear and well- before 2020, written in languages other than
defined questions to guide our review, such as: English, lacking peer review, or not directly
• What are the recent advancements in assis- related to the core aspects of our topic.
tive technologies for web accessibility?
• How do these innovations impact the
user experience for individuals with 3. Thematic Overview
disabilities? One of the most significant milestones in
• What is the role of new technologies, such the world of online inclusivity standards is
as artificial intelligence, machine learning, the Web Content Accessibility Guidelines
and IoT, in improving web accessibility?” (WCAG) created by the World Wide Web
Consortium. WCAG is an internationally rec-
2.2. Search criteria, databases, and ognized set of guidelines that outlines how
sources web content can be made more inclusive for
individuals with disabilities. These recom-
Our search criteria included keywords and mendations encompass multiple facets, such
phrases related to the topic of review. These as providing text substitutes for non-text
criteria were carefully chosen to cover vari- information, ensuring keyboard navigation,
ous aspects of the subject, ensuring a compre- offering easy-to-understand content, and
hensive search. We used Boolean operators designing adaptable web content. WCAG
(e.g., AND, OR) to combine search terms operates on a scale of success criteria, offer-
effectively and maximize the retrieval of rel- ing developers and designers a structured
evant studies. framework to ensure that their websites meet
We performed searches in various reliable a range of accessibility requirements.
online resources, including academic search As technology continues to evolve, web
engines and digital libraries. Each database accessibility standards will remain a corner-
was chosen for its relevance to the field and stone of the digital landscape, ensuring that
its coverage of academic literature. This the benefits of the internet are available to all,
multi-database approach helped us capture a regardless of their abilities.
broad spectrum of sources.
The primary focus was on academic 3.1. Braille displays
sources, such as peer-reviewed journal arti-
cles, conference papers, and scholarly books. In a study, the respective authors introduce
We ensured that the sources were recent (pub- a novel reconfigurable Braille display instru-
lished after 2020) to reflect the latest develop- ment employing a magneto-resistive blend
ments in the field. featuring triple shape memory. The study
explores advancements in Braille display tech-
nology with a focus on improving the user
2.3. Search inclusion and exclusion
engagement for users with visual disabilities.
criteria for selecting studies The device employs magneto-resistive materi-
We included studies that were published after als and shape memory technology to create a
2020 to ensure the most up-to-date informa- smart Braille display [1].
tion. Additionally, studies had to be written In another paper, a novel refreshable Braille
in English to ensure consistency in the review. display is presented, emphasizing a multi-lay-
We included sources that directly addressed ered electromagnetic driving system for Braille
Advancements in assistive technologies for web accessibility: A systematic review 453
dots. This innovation enhances the tactile feed- the ongoing efforts to enhance accessibility
back of Braille displays, simplifying access for through innovative technologies [6].
individuals with visual impairments to access These recent advancements in screen
digital material. The paper outlines the tech- reader technology have significantly contrib-
nology’s working principles and its applica- uted to improving the web experience for
tions in assistive technology [2]. individuals with visual impairments.
Introducing a Braille reading system
grounded on an electro-tactile interface fea- 3.3. Voice recognition and speech to
turing a adaptable electrode array, the authors text technologies
of the respective paper present a method that
provides tactile feedback through electro-tac- In their paper, the respective authors demon-
tile stimulation. This allows individuals with strate how machine learning algorithms can
visual impairments to read Braille characters be trained to understand and transcribe the
electronically. The article outlines the con- unique speech patterns of users with speech
struction and potential applications of this disorders. This application leverages voice
system, contributing to the field of assistive recognition technology to convert spoken
technologies [3]. words into text, offering a lifeline to those
who may struggle with traditional communi-
cation methods [7].
3.2. Screen readers In another study, the authors of the respec-
tive research introduce an innovative appli-
In their recent research, the respective cation focused on enhancing the digital
authors unveiled Olli, an extensible visu- experience of visually impaired users. This
alization library that focuses on enhancing system enables blind individuals to access
screen reader accessibility. Olli is designed to and send emails through voice commands
provide an accessible and inclusive experi- and speech recognition powered by artificial
ence for data visualization, a critical aspect intelligence [8].
of web content. The authors presented Olli as
a promising tool to make complex data visu-
alizations more accessible to users with visual 3.4. Natural language processing
impairments [4]. A research paper presents a Lexical Simpli-
Introducing VizAbility, a novel approach fication System designed to enhance web
to accessible data visualization, the authors of accessibility. This system employs NLP tech-
the respective paper demonstrate a method to niques to simplify the content of web pages,
combine keyboard navigation and conversa- making them more comprehensible for users
tional interaction, making it easier for screen with cognitive or reading disabilities. The
reader users to explore and comprehend data study recognizes that the language complex-
visualizations. This innovative approach con- ity of web content can be a significant bar-
tributes to a more inclusive web environment, rier to access for individuals with disabilities,
allowing individuals with visual impairments including those with limited reading skills [9].
to engage with data-driven content seam- In related context, another study highlights
lessly [5]. the transformative role of NLP in improv-
Another extensive study introduces a Rasp- ing the lives of people with disabilities. The
berry Pi-based Smart Reader designed to aid respective authors of the research underlines
individuals with visual impairments. This how NLP-driven tools and applications can
device facilitates the reading of printed text and help individuals with various disabilities,
converts it into audible speech. Although it pri- including those with visual, hearing, or cogni-
marily addresses offline content, it showcases tive impairments, interact with online content
454 Innovations in Computational Intelligence: Trends and Technologies
more effectively. The study also emphasizes an innovative approach that streamlines
the importance of multimodal interactions, web content to enhance user comprehension
combining NLP with other technologies like and navigability [14].
speech recognition and gesture control to pro- Another pioneering research study pre-
vide a comprehensive assistive experience [10]. sents a hybrid brain-computer interface (BCI)
web browser. This novel browser combines
3.5. Image recognition steady-state visual evoked potentials and
gaze-tracking technology to provide a more
In a detailed research, the respective authors efficient and user-friendly web browsing
of the study leverage Convolutional Neural experience. This innovation aids individuals
Networks (CNNs) to recognize hand gestures with motor or cognitive disabilities [15].
for individuals with disabilities. It enables ges- In another recent research, the respective
ture recognition that can facilitate web brows- authors present a unique approach by utilizing
ing, content navigation, and interaction with the triboelectric properties of the human body
online applications, making web resources for human-computer interactions. This intrigu-
more accessible to a broader audience [11]. ing method opens up possibilities for touchless
In a study related to the context, the and gesture-based web interactions [16].
respective authors introduce an innovative In a study within relatable context intro-
Human-Computer Interaction (HCI) appli- duces a feedback-based system for augmenta-
cation which employs facial movements to tive and alternative communication grounded
control the mouse cursor, thereby providing in human-computer interaction (AAC). AAC
an alternative means of computer interaction. is crucial for individuals with speech or lan-
This technology enhances web accessibility guage impairments, offers a promising solu-
by allowing users to navigate web content tion to improve their web browsing and
through facial gestures, ultimately improving communication experiences [17].
the overall web experience [12].
While introducing “DeepNAVI,” the
authors of respective paper introduce a 4. Critical Analysis
smartphone navigation assistant powered by In this critical analysis, the strengths and
deep learning, designed for individuals with weaknesses of assistive technologies for web
visual impairments. This smartphone appli- accessibility are thoroughly examined, with
cation leverages image recognition and deep a central focus on identifying and addressing
learning to provide real-time auditory cues existing gaps and challenges.
to users, guiding them through their sur- The innovation of Braille displays is
roundings. When navigating web content, acknowledged, but their prohibitive cost poses
individuals with visual impairments heavily a significant obstacle to widespread acces-
rely on screen readers and assistive technolo- sibility. Screen readers, a fundamental tool,
gies. DeepNAVI extends this functionality by are contingent upon proper web content cod-
providing context-aware guidance based on ing for optimal functionality. The potential of
image recognition [13]. voice recognition and speech-to-text technolo-
gies to enhance communication is recognized,
3.6. Innovations in human computer yet their efficacy is hampered by a need for
interaction and web browsers improved accuracy. Natural Language Process-
ing (NLP) simplifies content but grapples with
One notable publication delves into the challenges related to subtle language nuances.
creation of a web simplification prototype The empowering capabilities of image recogni-
to cater to cognitive-disabled users. The tion for users are acknowledged, but accuracy
authors of the respective research propose
Advancements in assistive technologies for web accessibility: A systematic review 455
concerns arise, particularly in varying light- 6. Conclusion
ing conditions. Human-Computer Interaction
(HCI) and web browser innovations strive for In this comprehensive review paper, we have
inclusivity, but their effectiveness is contingent explored the multifaceted realm of assistive
upon standardization and thoughtful consid- technologies for web accessibility. While signifi-
eration of user needs. cant progress has been made, there remains a
call to action. From evolving web accessibility
standards to enhancing Braille display technol-
5. Future Research Work ogy and pushing boundaries in image recogni-
To advance web accessibility and assistive tion, we have outlined research directions that
technologies, we suggest several research can shape a more inclusive digital world.
directions that can help evolve the domain of As we step into the future, the principles
web accessibility to new heights. of web accessibility, user-centric design, and
• Develop standards that include emerging ethical considerations must remain at the
tech like AR, VR, voice interfaces, and forefront of our efforts. The journey towards
IoT. Collaboration between W3C, tech- a digitally inclusive society continues, and
nologists, accessibility experts, and indi- this review paper acts as a guide, directing us
viduals with disabilities is crucial. towards the path of progress and collabora-
• Make Braille displays more afford- tion in the domain of assistive technologies
able, portable, and adaptable. Scope of for web accessibility.
improvement in refreshable Braille tech
with faster response times and higher Acknowledgment
resolution through collaboration between
academia, tech companies, and users with The authors gratefully acknowledge the stu-
visual impairments. dents, staff, and authority of Computer sci-
• Create more sophisticated screen reader ence and engineering department for their
interactions with multimodal interfaces, cooperation in the completion of this paper.
combining speech, gestures, and haptic The appreciation is also extended towards
feedback. Collaboration between HCI, AI the researchers and authors whose insightful
researchers, and individuals with visual studies and publications formed the founda-
impairments is key. tion of this review.
• Enhance accuracy and adaptability of voice
recognition. Simplify complex web content References
using NLP and cater to unique user needs.
[1] Hu, Tao, Shouhu Xuan, Yinduan Gao,
Investigate expanded applications like real-
Quan Shu, Zhenbang Xu, Shuaishuai Sun,
time object recognition, scene descriptions, Jun Li, and Xinglong Gong. 2021. “Smart
and image-based navigation. Refreshable Braille Display Device Based
• Make web content adaptive to a wider on Magneto-Resistive Composite with Tri-
range of disabilities. Allow users to con- ple Shape Memory.” Advanced Materials
trol content with facial gestures, voice Technologies 7 (1): 2100777. https://doi.
commands, or brain-computer interfaces org/10.1002/admt.202100777.
through collaboration between HCI, AI, [2] Chen, Hao, Wentao Tao, Chang Liu, Qi
and individuals with motor disabilities. Shen, Wu Yuecheng, Liuxia Ruan, and
• Involve individuals with disabilities in Wenzhen Yang. 2023. “A Novel Refresh-
every design stage. Collaboration between able Braille Display Based on the Layered
UX designers, accessibility experts, and Electromagnetic Driving Mechanism of
Braille Dots.” IEEE Transactions on Haptics
individuals with disabilities can create
more inclusive digital products.
456 Innovations in Computational Intelligence: Trends and Technologies
16 (1): 96–105. https://doi.org/10.1109/ [11] E. P. Shadiya Febin, and Arun T Nair. 2022.
toh.2023.3241952. “Hand Gesture Recognition for Disabled
[3] Zhou, Ziliang, Yueying Yang, and Hong- Person with Speech Using CNN.” Lecture
hai Liu. 2022. “A Braille Reading System Notes on Data Engineering and Communica-
Based on Electrotactile Display with Flex- tions Technologies, January, 239–49. https://
ible Electrode Array.” IEEE/CAA Journal of doi.org/10.1007/978-981-16-7610-9_17.
Automatica Sinica 9 (4): 735–37. https://doi. [12] R Shashidhar, K Snehith, P K Abhishek,
org/10.1109/jas.2022.105476. Abhishek B Vishwagna, and N Pavitha.
[4] Blanco, Matt, Jonathan Zong, and Arvind 2022. “Mouse Cusor Control Using Facial
Satyanarayan. 2022. “Olli: An Extensible Movements - an HCI Application.” 2022
Visualization Library for Screen Reader International Conference on Sustain-
Accessibility.” Vis.csail.mit.edu. October 19, able Computing and Data Communica-
2022. https://vis.csail.mit.edu/pubs/olli. tion Systems (ICSCDS), April. https://doi.
[5] Gorniak, Joshua, Yoon Kim, Stephen Gwon, org/10.1109/icscds53736.2022.9760914.
Donglai Wei, and Nam Wook Kim. 2023. [13] Kuriakose, Bineeth, Raju Shrestha, and
“VizAbility: Multimodal Accessible Data Frode Eika Sandnes. 2023. “DeepNAVI:
Visualization with Keyboard Navigation A Deep Learning Based Smartphone Navi-
and Conversational Interaction.” ArXiv gation Assistant for People with Visual
(Cornell University), October. https://doi. Impairments.” Expert Systems with Appli-
org/10.48550/arxiv.2310.09611. cations 212 (February): 118720. https://doi.
[6] Ravi, A, Sk Khasimbee, T Asha, T Naga org/10.1016/j.eswa.2022.118720.
Sai Joshna, and P Gnana Jyothirmai. 2020. [14] Kamran, Maira, Marium Malik, Muham-
“Raspberry Pi Based Smart Reader for Blind mad Waseem Iqbal, Muhammad Anwar,
People.”IEEE Xplore. July 1, 2020. https://doi. Muhammad Aqeel, and Sana Ahmad. 2022.
org/10.1109/ICESC48915.2020.9155941. “Web Simplification Prototype for Cogni-
[7] Mulfari, Davide, Gabriele Meoni, Marco tive Disabled Users.” Edited by zheng yan.
Marini, and Luca Fanucci. 2021. “Machine Human Behavior and Emerging Tech-
Learning Assistive Application for Users nologies 2022 (March): 1–14. https://doi.
with Speech Disorders.” Applied Soft Com- org/10.1155/2022/5817410.
puting 103 (May): 107147. https://doi. [15] Lin, Xinyuan, W.Q Malik, and Shaomin
org/10.1016/j.asoc.2021.107147. Zhang. 2019. “A Novel Hybrid BCI Web
[8] Biruntha, S., Gaja Priya, M., Kiruthika, R., Browser Based on SSVEP and Eye-Track-
Indupriya, N., and Ashwini, M. “Voice- ing,” October. https://doi.org/10.1109/
Based Email for Blind People Using Speech biocas.2019.8919087.
Recognition through Artificial Intelligence.” [16] Zhang, Renyun, Magnus Hummelgård, Jonas
International Journal of All Research Educa- Örtegren, Martin Olsen, Henrik Anders-
tion and Scientific Methods (IJARESM), vol. son, Ya Yang, Håkan Olin, and Zhong Lin
9, 2021. Available: https://jit.ac.in/journal/ Wang. 2022. “Utilising the Triboelectricity
cse/CSE_20-21/9.pdf. of the Human Body for Human-Computer
[9] Alarcon, Rodrigo, Lourdes Moreno, and Interactions.” Nano Energy 100 (Septem-
Paloma Martinez. 2021. “Lexical Simplifi- ber): 107503. https://doi.org/10.1016/j.
cation System to Improve Web Accessibil- nanoen.2022.107503.
ity.” IEEE Access 9: 58755–67. https://doi. [17] Yubin Liu, C. B. Sivaparthipan, and Achyut
org/10.1109/access.2021.3072697. Shankar. 2021. “Human–Computer Inter-
[10] Nadeem, Warda, Vinod Kumar Shukla, action Based Visual Feedback System for
Preetha V K, Gagandeep Kaur, and Arad- Augmentative and Alternative Communica-
hana Balodi Bhardwaj. 2022. “Role of tion.” International Journal of Speech Tech-
Natural Language Processing in Improv- nology, October. https://doi.org/10.1007/
ing Lives of Disabled.” IEEE Xplore. s10772-021-09901-4.
October 1, 2022. https://doi.org/10.1109/
ICRITO56286.2022.9964626.
64 Predicting bitcoin prices: A machine
learning approach for accurate
forecasting
Rishabh Jaina, Shekhar Srivastavab, and Prakhar Shuklac
Computer Science and Engineering, Babu Banarasi Das Institute of Technology and Management,
Lucknow, India
Abstract: This project investigates the active realm of Bitcoin price forecasting through the glass of machine
intelligence models, including Logistic Regression, Support Vector Machines (SVM), and XGBoost Classifier.
Leveraging a different dataset including historical and actual-occasion Bitcoin price dossier, the study employs
an orderly method for dossier collection, feature collection, model preparation, and judgment. The aim is to
embellish the veracity of short-term and unending forecasts, making the challenges posed apiece explosive cryp-
tocurrency retail. The project extends further hypothetical exploration, climactic in the incident of a convenient
web connect. This connects employs HTML, CSS, and Flask API to provide authentic-opportunity forecasts,
extending the gap betwixt leading predictive models and proficient uses.
Keywords: Bitcoin, Price Prediction, Machine Learning, Logistic Regression, Support Vector Machines, XG-
Boost Classifier, Cryptocurrency, Web Interface, Forecasting, Real-time, HTML, CSS, Flask API
1. Introduction scope of vacillation, translation it an enigma
for two together experienced sellers and new-
Bitcoin is a distributed electronic cash that comers. The active type of Bitcoin’s price is
uses cryptography for protection and is not intertwined accompanying a large group of
conditional some management or monetary determinants, containing advertise belief,
organization. It was created in 2008 by an technological progresses, supervisory inci-
individual or group of things utilizing the alias dents, and macroeconomic occurrences.
Satoshi Nakamoto (2008) accompanying a Understanding and calling the price of Bit-
paper named “Bitcoin: A Peer-to-Peer (P2P) coin is not merely an occupation for finan-
Electronic Cash System” [1]. In the rapidly ciers pursuing monetary gain, but again a
developing countryside of monetary markets, subject of superior importance for managers
cryptocurrencies have arisen as a causing trou- forming cryptocurrency procedures, investi-
ble force, challenging established contribution gators solving the intricacies of blockchain
example and attracting the interest of financi- electronics, and technologists construction
AQ1:
Please
ers, professors, and policymakers alike. Simi- healthy business algorithms. Forecasting has
note the lar to other property, to a degree stocks and enhanced an increasingly complex process,
cross-
reference
merchandise, Bitcoin price forecasts are a suc- particularly contemporary, place markets are
NOTE ADDED
2-3, 5, cession of unending predictions cause Bitcoin sufficiently affiliated, and facts circulates eas-
7, 12. prices too change over opportunity [4]. ier and faster [6].
And also
not in As Bitcoin’s upstandingness persists to One of ultimate main facets of utiliz-
sequential evolve, its price has shown an extraordinary ing cryptocurrencies, two together as an
order.
[email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-64
458 Innovations in Computational Intelligence: Trends and Technologies
intermediary and as an advantage, is the valuable for those powers the one purchase
wonted amount forecast. Predicting the price Bitcoin, to a degree u ndertaking speed, dis-
of Bitcoin, as a favorite mathematical cash, semination, DE centrality, and the big com-
has been a most of learning. Besides, plenty puter world of family curious in speaking and
research and procedures have happened done providing relevant news about mathematical
to call the price of mathematical currencies. currencies, principally Bitcoin [8].
For example, affected affecting animate nerve As we begin undertaking this journey
organs networks are usual to envision finan- through the extant history on Bitcoin price
cial markets by utilizing mechanics signs. indicator, it enhances clear that the request of
Plus, retail belief indices were secondhand as machine learning models, containing Logis-
stock exchange predictors. Furthermore, the tic Regression, SVM, and XGBoost Classifier,
Estimation Maximum Likelihood (EML) was shows an important and creative approach to
used to show the Bitcoin display is entirely forwarding the complicatedness of cryptocur-
adept. Nevertheless, making accurate indica- rency price forecasting, individual that persists
tors in a complex and swift examining foun- to challenge the confines of fiscal information.
dation is still assuredly a disputing issue [3]. The article revenue in this manner. In Sec-
The objectives concerning this composi- tion 2, we will outline the methods second-
tion review are three times as many. Firstly, hand for selecting and judging the research on
we aim to support an awareness into the evo- Bitcoin price prediction. In Section 3, we will
lution of Bitcoin price indicator models and support a having a theme survey by classifica-
the various methods working in guessing. In tion the history into key themes and empha-
particular, this review will devote effort to size main believes and models. In Section 4,
something the application of machine intel- our fault-finding study will assess substances
ligence methods to a degree Logistic Regres- and restraints of existent research, concentrat-
sion, Support Vector Machines (SVM), and ing on study feature and dependability. Section
XGBoost Classifier in forecasting Bitcoin 5 will synthesize judgments from miscellane-
prices. Secondly, we endeavor to precariously ous studies to draw encompassing judgments
analyze the substances and disadvantages and review their implications. Section 6 will
of existent research, peeling arrive the chal- offer approvals for future research, stressing
lenges associated with forecasting the price uncharted districts and key questions needing
of an advantage as explosive and singular as further study. Lastly, Section 7, our conclusion,
Bitcoin utilizing these models. Lastly, we will will recap main judgments, repeat the review’s
combine the findings from differing studies, meaning, and, if appropriate, offer policy,
illustration encompassing ends and reviewing practice, or research pieces of advice.
the implications for the field of cryptocurrency
research and expenditure practices. Forecast-
ing Bitcoin price is extremely important for 2. Methodology
two together advantage managers and liber- This division outlines the inclusive methods
ated financiers. Although Bitcoin is a currency, for our project on Bitcoin price prognosis.
it cannot be intentional as another usual bills The methods cover dossier group, feature
place financial theories about exposed inter- pick, model preparation, forecasts, nick, con-
est equality, future cash-flows model, and firmation, and the last exercise and accom-
ability to purchase balance matter, since vari- modating of your project.
ous standard determinants of the connection
middle from two points supply and demand
cannot be used in the mathematical currency 2.1. Data collection
advertise like Bitcoin. On the individual BTC Data Source: To introduce the project,
help, Bitcoin has various traits that make it draw classical Bitcoin price dossier from a
Predicting bitcoin prices: A machine learning approach for accurate forecasting 459
confidential information, to a degree crypto- Regression Metrics: Assess reversion mod-
currency exchanges or dossier providers. els utilizing versification like Mean Absolute
Data Gathering: Acquire dossier connect- Error (MAE) and Root Mean Square Error
ing a solid period of time, by preference a (RMSE).
6-period dataset accompanying absolute-
period costing. 2.6. Web framework
Data Cleaning: Perform dossier cleans-
ing to away gone principles, outliers, and HTML/CSS Framework: Develop a handy
discrepancies. netting connect for your project utilizing
Data Scaling: Normalize or scale the dos- HTML and CSS.
sier to guarantee that all facial characteristics
have the alike pressure. 2.7. Flask API
Handshaking: Create an API utilizing Flask
2.2. Feature selection to authorize ideas middle from two points
Feature Selection: Choose appropriate computer network connect and the model
face that have ultimate affect Bitcoin price prophecies.
forecasting.
6-Month Dataset (Real-Time): Create a 2.8. Hosting on Web
dataset forming ultimate current 6 months of Hosting: Deploy the project on a net enter-
authentic-period Bitcoin price dossier. taining terrace.
Train-Test Split: Divide the dataset into This procedure specifies a structured
preparation and experiment subsets, illustrat- approach to our Bitcoin price forecast pro-
ing the split percentage (for instance, 80% for ject, including data group, model prepara-
preparation and 20% for experiment). tion, judgment, and deployment.
2.3. Training models
Training of Classification Models: Train
machine intelligence categorization mod-
els (for instance, Logistic Regression, SVM,
XGBoost Classifier) to conclude price flows
(for instance, ‘up’ or ‘unhappy’).
Training of Regression Models: Develop
reversion models to think the real Bitcoin price.
2.4. Forecasts
Classification Forecasts: Use categorization
AQ2:
models to form prophecies about the man- Please
agement of Bitcoin prices. provide
Regression Forecasts: Employ reversion NOTE
figure
ADDED
caption
models to anticipate particular price principles. and
in-text
citation.
2.5. Scoring and validation AQ3:
Classification Metrics: Evaluate categoriza- Please
provide
tion models utilizing versification like verac- Figure 64.1. source for
ity, accuracy, recall, and F1-score. Source: AUTHOR'S SELF GENERATED
the figure
460 Innovations in Computational Intelligence: Trends and Technologies
3. Thematic Overview what way or manner business-related signs
and tactics developments shape the crypto-
This survey is devised to offer bookworms currency advertise. Cryptocurrency display
a comprehensive understanding of the dif- interplays are an idea that survey Bitcoin’s
ferent approaches and methods used in this relationships accompanying additional cryp-
place field of study. The information is cat- tocurrencies. This involves cross-advantage
egorized into various key having a theme equivalences and potential spillover belong-
clusters, each focusing on various facets of ings inside the fuller cryptocurrency adver-
Bitcoin price prediction. One outstanding tise, recognizing the interconnectedness of
having a theme cluster focuses on predicting miscellaneous mathematical property. Vola-
models and methods. This contains a range of tility and risk evaluation are important for
machine learning models, containing Logistic investors, and research in this place cluster
Regression, Support Vector Machines (SVM), focuses on approaches to quantifying Bitcoin
and XGBoost Classifier. Researchers situ- price airiness and risk. Understanding risk
ated on sides have investigated the develop- and allure calculation is fundamental for cog-
ment of these predicting techniques, judging nizant investment resolutions.
their substances, defect, and their relevance to Finally, a multidisciplinary approach
guessing Bitcoin prices. is checked in studies that mix diversified
Market sentiment and friendly television perspectives and dossier beginnings. This
reasoning show another essential thematic involves joining emotion analysis accompa-
cluster. This region investigates the influence nying mechanics signs or macroeconomic
of display belief and the use of public televi- determinants, professed the potential for
sion data in anticipating Bitcoin prices. Stud- interdisciplinary research to embellish the
ies inside this idea determine the influence of veracity of Bitcoin price forecasting. By clas-
sentiment reasoning finishes and their unifi- sification the drama into these thematic clus-
cation into predicting models, acknowledging ters, we aim to support lectors accompanying
the importance of observable determinants an organized foundation for comprehending
in price activities. Fundamental reasoning is the various streets of research in Bitcoin price
a detracting category that considers deter- forecast. This approach will authorize schol-
minants in the way that blockchain dossier, ars to navigate the complicatedness of the
network versification, and on-chain signs. field and gain deeper acumens into the versa-
Researchers assess the pertinence and predict- tile policies and judgments offered by existent
ing capacity of these fundamental variables, research.
stressing their affect Bitcoin’s price dynamics.
Market signs and mechanics reasoning, fre-
quently secondhand in established financial 4. Critical Analysis
markets, again have their place in Bitcoin price
prophecy [2]. This idea includes components We precariously evaluate the strengths and
like moving averages, business books, and defect of the existent history on Bitcoin price
chart patterns, and checks by what method prophecy, with a devote effort to something
these established techniques maybe suitable the request of machine intelligence models,
and used to the cryptocurrency display. containing Logistic Regression, Support Vec-
The influence of business-related and tor Machines (SVM), and XGBoost Classifier.
macro-economic determinants on Bitcoin This critical reasoning inquires to supply an
price authorizes another having a theme inclusive evaluation of the current state of Bit-
cluster. This cluster delves into the impact coin price prognosis research. In conditions of
of macroeconomic occurrences and regula- predicting capacity, many studies have high-
tory changes on Bitcoin’s price, surveying in lighted the influence of machine intelligence
Predicting bitcoin prices: A machine learning approach for accurate forecasting 461
models, exceptionally when handling large price prediction is a complex and disputing
datasets and complicated display patterns. endeavor, meaningful progress has existed
These models have illustrated their ability to fashioned in understanding the dynamics of
capture market movement and offer valuable the cryptocurrency display. Machine learning
understandings into Bitcoin price currents. models have showed expected valuable forms
However, the literature is not outside allure in capturing and defining complex patterns
defect and challenges. Data characteristic and currents in Bitcoin prices. These models,
and availability wait determined issues. Inac- when suitably trained and ratified, can offer
curate or wanting dossier can undermine the correct understandings into temporary price
acting of predicting models and limit their movements [10,11].
actual-realm applicability. Addressing dossier However, the associations of these ver-
kind is authoritative for boosting the reliabil- dicts too underline several main warnings.
ity of Bitcoin price prognoses. The extreme excitability and susceptibility
Long-term sameness in Bitcoin price rem- of Bitcoin prices to outside determinants, to
nants restricted, a characteristic inherent to a degree display sentiment, revelation occur-
the cryptocurrency display. While advances rences, and supervisory changes, create long-
have existed created concisely-term predic- term prophecy a difficult challenge. The
tions, the challenge of correctly guessing Bit- cryptocurrency display’s hereditary unpre-
coin prices over lengthened periods persists dictability, accompanying the moral con-
[9]. Overfitting and inference pose another cerns encircling dossier collection and study,
challenge. Some studies have elevated con- demands carefulness and responsible research
cerns about overfitting, place models surpass practices [13].
on training dossier but struggle when accused The useful suggestions of correct Bitcoin
hidden dossier. Achieving robust inference, price guess extend to an off-course range of
exceptionally in the explosive cryptocurrency colleagues. Investors can form more conver-
retail, is an ongoing challenge that demands sant decisions about their capital and busi-
consideration. The basic excitability of the ness designs. Regulators and policymakers
cryptocurrency display, coupled with allure can benefit from visions into the factors doing
susceptibleness to outside determinants, pre- cryptocurrency prices, admitting bureaucracy
sents a dynamic atmosphere that is to say to art more effective tactics. Technologists
troublesome to forecast with complete verac- can cultivate business algorithms and uses
ity. Sudden display shifts sparked by news that respond to evident-period retail environ-
occurrences, supervisory changes, or retail ments with better veracity.
belief add a factor of doubt to Bitcoin price
forecasting, confusing the development of
trustworthy guessing models. 6. Recommendations for Future
Research
Firstly, skilled is a need for persisted empha-
5. Synthesis and Implications sis on dossier character and dependability.
In this portion, we combine the findings from Researchers should investigate designs to
the various array of studies on Bitcoin price address issues had connection with inaccu-
indicator. By collect the composite wisdom rate and wanting dossier, as reconstructing
of investigators in this place field, we aim to the status of input dossier is essential for
draw encompassing decisions and discuss the improving the accomplishment of predictive
fuller suggestions of these verdicts. models. Furthermore, the production of pat-
The combination of this thorough mate- terned datasets accompanying consistent and
rial of brochure tells that while Bitcoin finest news would expedite more robust and
462 Innovations in Computational Intelligence: Trends and Technologies
reproducible research or in general area. Sec- allure price action. Our exploration of the
ondly, investigators endure focus on evolving existent methodical study of part of material
more cosmopolitan and adjusting models that world has lighted several key understandings
can address the challenges of overfitting and and challenges inside this vital field.
generalization. Building models that exhibit The literature underlines the variety of
powerful inference capacities, particularly in methods and data beginnings used to Bit-
the framework of the very changeable crypto- coin price forecast. Researchers have har-
currency display, is a priority. This concedes nessed the capacity of machine intelligence,
possibility include surveying techniques to established economic analysis, retail emo-
a degree ensemble systems, deep education tion, friendly media dossier, blockchain
architectures, or recurrent affecting animate versification, and macroeconomic signs to
nerve organs networks to reinforce predicting develop predicting models. This difference of
accuracy and strength. approaches specifies a comprehensive view
Standardization of model judgment and of the elaborate determinants doing Bitcoin
confirmation criteria is a fault-finding pro- prices. However, our critical study has still
gress. Future research endures aim to estab- emphasized several continuous challenges.
lish a set of generally approved versification Data character and chance, overfitting,
and evaluation processes to allow significant model evaluation and confirmation tests,
corresponding between various models and advertise volatility, and moral concerns in
studies. This uniformity would advance trans- dossier analysis wait important issues. The
parency and dependability engaged of Bitcoin innately dynamic character of the cryptocur-
price prophecy. The role of outside determi- rency advertise, from rapid changes affected
nants in Bitcoin price motions, such as super- by outside determinants, poses a consider-
visory changes, macroeconomic occurrences, able challenge to the growth of trustworthy
and advertise sentiment, bear stretch to be a guessing models.
focus of research. Understanding by virtue of
what these factors communicate accompany-
ing price action and developing models that References
can accommodate to changeful environments [1] Chen J. Analysis of Bitcoin Price Prediction
is principal. This includes fact-finding designs Using Machine Learning. Journal of Risk
real-time dossier unification and reasoning, and Financial Management. 2023; 16(1):51.
allowing for up-to-date prognoses that join [2] Nikola Gradojevic, Dragan Kukolj, Rob-
with the vital type of the cryptocurrency ert Adcock, Vladimir Djakovic, Forecasting
Bitcoin with technical analysis: A not-so-
retail.
random forest?, International Journal of
Forecasting, Volume 39, Issue 1, 2023, Pages
7. Conclusion 1-17, ISSN 0169-2070.
[3] Shahab Rajabi, Pardis Roozkhosh, Nasser
In this paper, we have delved into the versa- Motahari Farimani, MLP-based Learnable
tile sphere of Bitcoin price prediction, accom- Window Size for Bitcoin price prediction,
panying the devote effort to something the Applied Soft Computing, Volume 129, 2022,
application of machine intelligence models, 109584, ISSN 1568-4946.
containing Logistic Regression, Support Vec- [4] Ye Z, Wu Y, Chen H, Pan Y, Jiang Q. A
tor Machines (SVM), and XGBoost Classifier. Stacking Ensemble Deep Learning Model
The cryptocurrency advertises, and Bitcoin in for Bitcoin Price Prediction Using Twitter
particular, has signed superlative tumor and Comments on Bitcoin. Mathematics. 2022;
volatility in current age, cueing a surge in 10(8):1307.
[5] Carbó, José Manuel and Gorjon, Sergio,
research aimed at understanding and guessing
Application of Machine Learning Models
Predicting bitcoin prices: A machine learning approach for accurate forecasting 463
and Interpretability Techniques to Identify via machine learning, The Journal of Finance
the Determinants of the Price of Bitcoin and Data Science, Volume 7, 2021, Pages
(April 19, 2022). Banco de Espana Working 45-66, ISSN 2405-9188.
Paper No. 2215. [10] Cocco L, Tonelli R, Marchesi M. 2021. Pre-
[6] Guarino, A., Grilli, L., Santoro, D. et al. To dictions of bitcoin prices through machine
learn or not to learn? Evaluating autono- learning based frameworks. PeerJ Computer
mous, adaptive, automated traders in crypto- Science 7:e413.
currencies financial bubbles. Neural Comput [11] Akyildirim, E., Cepni, O., Corbet, S. et al.
and Applic 34, 20715–20756 (2022). Forecasting mid-price movement of Bitcoin
[7] Shahab Rajabi, Pardis Roozkhosh, Nasser futures using machine learning. Ann Oper
Motahari Farimani, MLP-based Learnable Res (2021).
Window Size for Bitcoin price prediction, [12] Dirk G. Baur, Lai Hoang, The Bitcoin gold
Applied Soft Computing, Volume 129, 2022, correlation puzzle, Journal of Behavio-
109584, ISSN 1568-4946. ral and Experimental Finance, Volume 32,
[8] Lamothe-Fernández P, Alaminos D, Lam- 2021, 100561, ISSN 2214-6350.
othe-López P, Fernández-Gámez MA. Deep [13] Baur, D.G., Dimpfl, T. The volatility of Bit-
Learning Methods for Modeling Bitcoin coin and its role as a medium of exchange
Price. Mathematics. 2020; 8(8):1245. and a store of value. Empir Econ 61, 2663–
[9] Patrick Jaquart, David Dann, Christof Wein- 2683 (2021).
hardt, Short-term bitcoin market prediction
65 A comprehensive review on multi
disease prediction web app:
QuadraDiag
Khushi Saxenaa, Keerti Saxenab, Vishesh Guptac, Swatid, and
Kamlesh Kumare
Computer Science and Engineering BBNITM, Lucknow, India
Abstract: In today’s environment, artificial intelligence and machine learning are major players. They are used
in the medical fields as well as in self driving cars. A lost of patient data is produced by the medical industry, and
there are several ways to handle this data. Thus, we have developed a forecasting system proficient in identify-
ing numerous conditions forthwith with automated learning. Most of the current systems have poor accuracy
and can only forecast one disease at a time. Inaccurate results can gravely endanger a patient’s health. Heart,
liver, Parkinson’s, and diabetes are the four disorders we have currently considered; many more diseases may be
included in the future. The user needs to provide a number of parameters.
Keywords: Diabetes, Heart, Liver, and Parkinson’s disease, Machine learning, Xgboost, SVM, Logistic Regres-
sion, Pickle file, Flask API
1. Introduction utilized to anticipate various ailments. We are
going to investigate Liver, Parkinson’s, heart
In digital age, data is an extremely valuable
disease, and diabetes. Later on, more ailments
resource, with massive quantities being gen-
could be added. We’re going to leverage ML
erated in numerous industries. A data is a
techniques and Flask into numerous medical
valuable resource, with enormous volumes
prognosis. The model’s behavior is preserved
of data being generated across all industries.
All data pertaining to patients is referred to through the use of Python pickling. The signif-
as data in the healthcare industry. This arti- icance of this system analysis lies in the inclu-
cle proposes an architecture for illness antici- sion of all illness-causing parameters during
pation in the health sector. A large number the analysis procedure, making illness iden-
of current models focus on a single disease tification more accurate and successful. The
per analysis. Such analyses include those for finalized model’s demeanor is set to be stored
skin disorders, diabetes, and cancer. A single in a .pkl file.We frequently witness folks lose
system that is capable of looking at several their lives as a result of delayed medical care.
diseases at once does not currently exist. The healthcare sector is time-constrained
Therefore, our focus is on giving customers and unable to prioritize which patients to
precise and instantaneous disease forecasts, treat first. However, the healthcare sector
including the symptoms they may experience also produces a vast amount of data about
in addition to the projected sickness. Thus, we the health of its patients. From this data,
are putting forth a Flask-based method that is high-level insights can be obtained. Thus, we
[email protected]; [email protected]; [email protected];
a
[email protected]; [email protected]
d
DOI: 10.1201/9781003606635-65
A comprehensive review on multi disease prediction web app: QuadraDiag 465
have chosen to develop the “Multiple Dis- because many localized diseases have unique
ease Prediction System” project combining characteristics in various locations.
this data and cutting-edge machine learning
techniques. In order to treat patients who are
most at risk for a certain condition, our study 2. Methodology
combines three machine learning models to Prediction model development in diverse
detect individuals with diabetes, renal disease, machine learning algorithms shares a com-
and heart disease early on. We initially identi- mon methodology, adhering to a standard-
fied the problem statement and the kinds of ized framework. This uniformity streamlines
data that will be needed for our project. For the process and fosters consistency in the
each of our three machine learning models, creation of predictive models across different
we gathered three distinct datasets from Kag- algorithmic approaches.
gle. Following data collection, we correctly
examined the information and created visu- 2.1. Data set
als to aid in comprehension. Next, we used
categorical feature encoding and null value For diabetes prediction some of the data-
imputing to clean up the data. The dataset sets taken from Kaggle as well as from the
was later partitioned into training and testing UCI machine learning library. The majority
sets, where 80% of the data was designated of the datasets used to forecast liver disease
for training our predictive model, while the come from electronic health records, ILPD
remaining 20% was reserved for testing its benchmark dataset and some datsets for
efficiency. Following that, we tested a num- heart disease prediction are drawn from the
ber of classification techniques on each of the UCI patient database for heart illness and the
three datasets, incorporating supervised clas- patient dataset for heart disease from Hun-
sification statistical learning techniques. garian, Switzerland and Cleveland. And for
Health and medicine play a significant role parkinson’s disease prediction the datasets
in both human life and economic progress. are taken from PPMI.
Applications for technology-assisted health
care have grown dramatically over the past 2.2. Data Preprocessing
two decades. This strategy will also result Data preprocising is the initial step in data
in reduced treatment costs and anxiety dur- analysis, involving Raw data must be cleaned,
ing the last stages, which will enable prompt transformed, and arranged to make it ready
delivery of quality care and a decline in the for analysis. It involves operations like scal-
death rate. Furthermore, it is challenging to ing, encoding variables, addressing missing
forecast when a disease epidemic will occur values, and eliminating duplicates.
AQ1:
OTEPlease
ADDED
provide 2.3. Data processing
figure
caption Data processing entails organising and pre-
and paring data for model evaluation and train-
in-text
citation. ing. This covers operations such as feature
selection, data cleaning, normalisation, and
AQ2: partitioning data into training and testing
Please
provide sets. Machine learning algorithms can only
source for discover patterns and produce precise predic-
the figure Figure 65.1. tions or classifications if the data is processed
Source: AUTHOR'S SELF GENERATED properly.
466 Innovations in Computational Intelligence: Trends and Technologies
2.4. Data mining algorithms incorrect catogrization. Widely accepted sup-
port vector machine execution include MAT-
In order to find patterns, relationships, and
LAB, LIBSVM, and scikit-learn.
insights inside enormous datasets, data min-
ing algorithms are computer techniques. To
wring useful information out of unprocessed 2.4.4. Naïve Bayes algorithm
data, these algorithms are used in a variety of It is a simple approach to classifier construc-
domains such as data analysis and machine tion. This categorizer relies on the Bayes
learning. theorem and operates probabilistically. Since
each feature’s value is independent of every
2.4.1. Decision tree other feature’s value, all Naive Bayes classi-
fiers function with the class variable as their
This is a model or tree structured graph of
input. The Bayes Theorem as follows P(Y/X)
determination and their potential results,
= P(X&Y)/P(X).
embracing utility and random occurrence of
results, that is used as a decision assistance
tool. A decision tree can function as swiftly 2.4.5. ANN algorithm
transmute into a set of guidelines established An artificial neural network is a algorithmic
through mapping each leaf node commencing model that emulates the architecture and
from the initial node to each sole individual. purpose of anatomical neural network. The
At last, by adhering to these guidelines, suit- insight that traverses over an artificial neural
able conclusions can be drawn. network impresses its configuration because,
depending on its input and output, a neural
2.4.2. KNN algorithm network adapts or learns at every stage. When
it comes to modelling complex interactions
KNN is a well-liked as well as simple ML
between inputs and outputs or identifying
technique for regression and categorizing. In
patterns, A class of non-sequential empirical
KNN, an item is categorized or given a value
data mimicking approaches is ANN.
determined by the predominant classes, which
is the arithmetic mean calculated from the
k-nearest data points within a variable space. 2.4.6. Random forest
It is certain non-parametric, instance-based Random Forests are a kind of closest-neigh-
method that makes no basic notions regard- bor predictor that can be used as an ensemble
ing dispensation of the data. The importance learning tool for regression and classifica-
of ‘k’ and the distance metric selection deter- tion techniques. It creates multiple DT during
mine how effective KNN is. With larger data- preparation and findings of the class that is
sets, it may be computationally costly, but it the intermediate of the classes that are pro-
performs well for situations involving both duced by every single tree independently.
regression and classification.
2.4.7. XGboost algorithm
2.4.3. SVM algorithm
A popular and powerful machine learning
Using a hyperplane, this supervised learning method that performs exceptionally well in
method splits the data into two groups. The supervised learning tasks—particularly those
SVM seeks to strategically maximize the mar- using structured data—is called eXtreme
gins width, or the separation the optimal dis- Gradient Boosting, or XGBoost. XGBoost
tance separating the hyper-plan and the two uses gradient boosting, which iteratively
closest data points belonging to each other, minimizes the loss function, to improve the
in the direction of lessen the likelihood of model’s accuracy.
A comprehensive review on multi disease prediction web app: QuadraDiag 467
3. Literature Review outcomes tailored interventions. SVM model
mainly focus on managing and refining data
J. Jasmin et al. (2023) [1] Parkinson’s Dis- through Pandas libraries, executing model
ease (PD) debilitating neurodegenerative dis- selection, comparison of models, model train-
ease and cannot be diagnosed through blood ing and testing, upgrading its performance
tests. It affects body movements and is mostly and exporting the train model for future use.
found in people above 60. In this paper, they The integration of the trained model into an
collected a dataset that consists of data from application enables disease prediction in real-
voice recordings of PD patients and normal world scenarios, empowering healthcare pro-
individuals. Here they use 4 data-driven learn- fessionals, researchers.
ing algorithms naïve bayes, random forest, Mr. Valle Harsha Vardhan et al. (2023)
xgboost, and svm, and compared their accu- [4] The application of machine learning for
racy, sensitivity, and specificity. Which are of early heart disease prediction is of significant
naïve bayes 69.23%, 68.75%, 71.43%, ran- societal importance due to the high global
dom forest 94.87%, 100.00%, 71.43%, svm mortality rates associated with cardiac ill-
87.18%, 96.88%, 42.86%, xgboost 94.87%, nesses. Detecting such conditions in their
100.00%, 71.43% respectively. early stages empowers high-risk individuals
Poojaa C et al. (2023) [2] highlighted that to make lifestyle adjustments, a substantial
the main aim of this paper is to design a weara- advancement in healthcare. SVM, D-Tree,
ble device system that can help the PD patient R-Forest, N-Bayes, Adaptive Boosting, and
in diagnosis of the disease using autonomous X-Gradient Boosting are among the seven
learning algorithms as PD is a fast-growing machine learning techniques used in this
condition and affect 1 out of every 1000 peo- work. System efficiency is improved through
ple, being more inclined towards the popula- feature selection from the dataset’s 76 prop-
tion above 60 years. In this paper, they take erties, with a focus on achieving greater
a dataset from PPMI for conducting data accuracy. The results indicate that, out of the
analysis to extract high-accuracy features seven approaches, the X-Gradient Boosting
for designing the accurate system. Their out- classifier obtains the greatest performance
come unveil the accuracy rates concerning (81%) according to evaluation criteria such
SVM, RF, knn, linear regression, along with as the C-matrix, accuracy, precision, re-call,
xgboost were 66.0714%, 75%, 73.0303%, and f1-score.
64.2857%, and 82.1429%. Two tasks have Dr. T. Venu Gopal et al.(2023) [5] In order
been chosen by them to predict the PD. Task1 to achieve early detection and potentially life-
1min normal walk in a straight line and task2 saving interventions, processing raw health-
1min normal walk while serially subtracting care data for heart conditions using machine
a single digit number from a three-digit num- learning techniques is essential. In the medi-
ber throughout. cal field, predicting heart diseases is a difficult
Parshant et al. (2023) [3] explored the but vital task. Early detection makes timely
application of machine learning techniques preventive measures possible, which can sig-
for the prediction of multiple diseases, with nificantly lower mortality rates. In order to
a specific focus on heart disease, diabetes, effectively predict heart disease, this study
and Parkinson’s disease. For achieving higher uses a variety of data mining techniques. It
accuracy of 98.3 % we use Support Vec- places special emphasis on utilizing a small
tor Machine (SVM) model for developing a set of attributes to produce predictions that
framework of multi disease prediction model. are both accurate and efficient. The R- Forest,
The result mainfests the capacity supervised D-Tree, G-Naive Bayes, and L-Regression
machine learning is essential in the health- algorithms were four that were employed.
care sector to predict and enhance patient
468 Innovations in Computational Intelligence: Trends and Technologies
Gaussian Naive Bayes showed t best accuracy impairs a patient’s quality of life, has no
out of all of these. known cure, and causes significant disability.
Ruhul Amin et al. (2023) [6] investigated Dopamine is a chemical messenger created by
enhanced feature extraction systems in this neurons in this part of the brain. One neu-
study for liver patient categorization with ral-transmitter that assests monitor and syn-
the use of statistical machine learning tech- chronize organism movements is dopamine,
niques, utilising dimensionality reduction which acts as a bridge between the brain
strategies including PCA, FA, and LDA. An and nervous system elements. Speech, writ-
enhanced descriptor space which maximizes ing, walking, and other basic tasks become
class separation, ledgers the peak variation harder to perform as dopamine levels in the
amid the evidences, and considers the corre- brain fall. The early stages of Parkinson’s dis-
lation betwixt studied variables was retrieved ease can be recognised and detected by this
with the system’s assistance. Through a sim- research using predictive modeling protocol.
ulation exercise, we were able to duplicate Two examples of automated learning classi-
the conclusion with the suggested method- fication procedure that are used to diagnose
ology plus achieving standard accuracy of Parkinson’s disease in individuals are Support
91.4 percent in the congregation taxonomy Vector Machine and XGBoost. Consequently,
process. Making use of recognition rate and the XGBoost algorithm performs better than
AUC by R-Forest computational intelligence the S-Vector Machine in terms of precision.
based on validation protocols have increased Arumugam K et al. (2021) [9] aims to
between 1% and 18.5%, or about 89% accu- create a Diabetes related heart disease using
racy, by harnessing suggested strategy as for machine learning techniques as diabetes-
the difficult India Liver Patient Dataset Kag- related heart disease occurs in diabetic per-
gle benchmark dataset in comparison to the sons and a chronic diabetes disease arises
reference-based approaches. when the pancreas fails to create enough insu-
Deepika Bhupathi et al. (2022) [7] Because lin or when the body fails to utilize insulin
liver disease is difficult to diagnose due to the so it is very complex to procedure to predict
subtle nature of its symptoms, research is nec- such disease. They explored and discovered
essary to identify algorithms that can more that after comparison with most experienced
accurately forecast this terrible illness. The physicians can diagnose with 79.97% accu-
steps in the suggested LDP approach offer racy conversely, 91.1% of the machine learn-
improved phase alignment. After the data- ing systems could recognize. After comparing
set has been chosen, the missing values are classification algorithms Naïve Bayes, SVM
replaced and the dataset is balanced as part and Decision Tree they found out that the
of the preprocessing procedure. Subsequently, precision by the D-Tree came about giving
five distinct supervised learning techniques— the highest accuracy among them.
SVM, Naïve Bayes, KNN, LDA, and CART— Cameron R et al. (2020) [10] Heart plays
are used with R, and the accuracy is measured an important role in our body functioning
using confusion matrix metrics. The outcome and we can’t ignore that. A study tells us that
demonstrates that K-NN has a superior accu- 1,000,000 Americans in the United States each
racy of 91.7% in predicting liver illness. In year are diagnosed with heart failure and that
this study, autoencoders are used as a test is a serious problem. That’s why Cameron R.
case to better understand how well unsuper- Olsen provides a platform that will be cost-
vised algorithms can classify data. effective, easy to use, and provide accurate
Azizkhan F. Pathan et al. (2022) [8] Par- disease prediction that will help patients to
kinson’s disease (PD) is the second most com- easily diagnose heart-related disease. So our
mon neurological condition that significantly aim is to build an statical learning model that
A comprehensive review on multi disease prediction web app: QuadraDiag 469
will predict heart-related disease by taking related mortality. Next, SVM is employed
the patient’s symptoms as input. We’ll employ during investigation to predict hepatic illness.
the SVM, RNN, Random Forest, and Logistic Archana Singh et al. (2020) [13] The main
Regression techniques. aim of the paper is that heart plays an impor-
Akkem Yaganteeswarudu et al. (2020) [11] tant role in living organisms. Accurate diagno-
found that a lot of the health care systems in sis and prediction of heart-related diseases are
place today only concentrate on one type of essential because they can potentially lead to
illness, and that in order for an organisation fatal cardiac events so for predicting this kind
to evaluate the health of its patients, it must of crucial health condition we use machine
develop a variety of models and follow a num- learning and artificial intelligence. In this
ber of steps. The intent concerning to evolve paper, these algorithm help in calculating the
a infrastructure aforementioned allowed to precision of data-driven learning model sup-
simultaneously outlook many diseases on the porting the prediction of heart ailment using
same platform. The diseases that are being KNN, decision tree, svm, linear regressive by
considered in this article include diabetes, using University of California repositor data-
cancer, heart disease, and diabetic retinopa- frame for development of model and analysis
thy. The method makes use of the Flask API comparison of algorithm and their accuracy
and a machine learning technique. Flask is a is also done that are as follows: Decision tree
Python micro web framework that makes it 79%, SVM 83%, k-nearest neighbour 87%,
simple to develop online apps and APIs. The Linear regression 78%.
user can load a pickle file to obtain the model Mitushi Soni et al. (2020) [14] Diabetes
behaviour corresponding to the chosen ail- occurs when the body does not create enough
ment while utilising the Flask API. The dataset insulin. In accordance WHO, 422 million
used is the PIMA India dataset for diabetes, people worldwide mostly in low- and middle-
University of Califonia, Irvine machine learn- income countries struggle with diabetes. And
ing repository for Diabetes Retinopathy, sup- by the year 2030, this might rise to 490 bil-
porting cardiopathy the dataset of patients lion. Nonetheless, High blood sugar is a con-
from Hungarian, Switzerland and Cleveland, ventional disease in numerous countries like
and Cancer, a Brest Cancer Wisconsin data- Canada, India. With over 100 million peo-
set is used. Sir has used Random Forest for ple living in India now, there are actually 40
Heart, Diabetes and SVM for Cancer whereas million diabetics living there. The suggested
TensorFlow CNN is used for Diabetes Retin- strategy makes use of a number of taxonomy
opathy as images are used for it. and collaborative learning techniques.
A. Sivasangari et al. (2020) [12] Liver ill- Chayakrit Krittanawong et al. (2020) [15]
nesses are a major cause of death in India The prediction of cardiovascular diseases,
and a worldwide concern. Early diagno- such as heart failure, stroke, coronary artery
sis of liver sickness can be challenging, but disease, and cardiac arrhythmias, was evalu-
machine learning techniques in an auto- ated using machine learning (ML) algorithms
mated approach have shown promise. This in a thorough investigation. Prospective find-
study assessed the performance of the SVM, ings emerged from the analysis of data from
D-Tree, and R-Forest blueprint in terms of 344 studies involving 3,377,318 participants.
authenticity, precision, and sensitivity. The High accuracy was achieved in the prediction
accuracy rates that were found were 95%, of coronary artery disease, with AUC values
87%, and 92%, in that order. These results associated with 0.93 and 0.88 for amplifying
show how machine learning may be used to and tailor-made methods, consecutively. The
improve liver disease early identification and SVM, boosting, and CNN algorithms showed
care, which will improve prognosis and lower AUC values of 0.92, 0.91, and 0.90 for stroke
470 Innovations in Computational Intelligence: Trends and Technologies
prediction. It was discovered that the predic- caused by Diabetes per year. The crucial part
tion of cardiac arrhythmia and heart fail- is to diagnosis the diabetes in its early stage
ure was a suitable use for SVM. The study for its treatment. Deep neural networks have
illustrated the potential of machine learn- been used in their study to predict diabetes.
ing algorithms in cardiovascular care while Their study tells us that using logistic regres-
emphasising the need for additional research sion they obtained highest accuracy of 78%,
and integration into clinical practice. enhanced genetic algorithm acquired correct-
Rudra A. Godse et al. (2019) [16] proposed ness of 80.4%, secured precision of 96.71%
a system that will have a simple and elegant utilizing customized K-Nearest Neighbour
UI and also be time efficient which will help and S-Vector Machine and precision of
in creating a connecting bridge between Doc- 97.47% using S-Vector Machine alongside
tors and Patients. They have analyzed dif- effective development. Their approach out-
ferent algorithms and have chosen the best performs the state of the art with an accuracy
algorithm for the prediction of various dis- of 98.35% when fivefold cross-validation is
eases. The models used in this article mostly performed using a deep neural network.
apply Supervised Learning. Supervised Learn- Deepti Sisodia et al. (2018) [19] The study
ing is an approach in which the data is sepa- aims to utilize machine learning algorithms
rated into two sections. One is being trained, to predict diabetes at an early stage, address-
and the other is testing, the model is trained ing the critical issue of late identification
using them and then asked for new values for and treatment of this chronic disease. From
prediction. This article uses an 80/20 split, a dataset sourced from the UCI machine
which divides records among 80% directed learning repository, we compared three clas-
to preparation and 20% for screening. Such sification algorithms: SVM, Naive Bayes, and
resulting model is then used to make predic- Decision Tree. Evaluation metrics included
tions using a questionnaire. Precision, Accuracy, F-Measure, and Recall.
Priyanka Sonar et al. (2019) [17] dia- Naive Bayes yielded the highest accuracy,
betes is one of the most deadly diseases in achieving 76.30%, surpassing the other algo-
the world since it can lead to a wide range rithms. Early identification of diabetes is of
of conditions, such as blindness. This paper paramount importance due to the severe com-
contains employs machine learning methods plications associated with untreated diabetes,
to identify diabetes since it is a forecastable and the study’s results suggest that machine
condition that is simple to predict whether learning, particularly Naive Bayes, can play a
or not the patient is unwell. The purpose of vital role in achieving this goal, with potential
the evaluation supposed to develop a frame- applications for other diseases in the future.
work that would let user correctly diagnose Chieh-Chen Wu et al. (2018) [20] arti-
themselves with diabetes. In this case, they cle says that fatty liver disease is a common
primarily used four algorithms: SVM, D-Tree, clinical problem and increases mortality rate.
and Naïve Bayes. Centered on an accuracy Therefore accurate identification of such
comparison of 85%, 77%, and 77.3%, they issues will be considered a great help in soci-
chose to use SVM for the prediction of dia- ety. The article uses machine learning tech-
betes after analysis. They also used the ANN niques as over the past decades Biopsy has
algorithm to observe the network’s responses been used for diagnosis and this is a way
after the training phase. costly method. The liver protection project at
H. Abbas et al. (2019) [18] This research New Taipei City Municipality Hospital Ban-
paper tells us that the Diabetes is a chronic dis- qiao Branch provided the textual and numer-
ease that has to be prevented before distresses ical dataset that was used. Patients under 30
people. A large number of death are being who had an ultrasound-confirmed case of
A comprehensive review on multi disease prediction web app: QuadraDiag 471
fatty liver were excluded. To select the best patient outcomes. The versatility of the sys-
model, they looked at five machine learning tem in handling various medical conditions is
models: Random Forest, LR, SVM, N-Bayes, evidenced by the application of multiple algo-
and ANN. Where R-Forest gave highest accu- rithmic modelling modalities, such as S-Vec-
racy about 91% and was chosen as the best tor Machine, D-Tree, Naïve Bayes, R-Forest,
model, whereas SVM, Naïve Bays, Logistic and XGBoost.
Regression and ANN gave 76.30%, 82.92%, This paper also references other relevant
64.90% and 74.10% respectively. studies that emphasize the need of early ill-
ness identification and the ability of machine
learning to predict a diversity of malady,
4. Recommendation for Future entailing glycemic disorder, cardiologic disor-
Work der, liver affliction, and Parkinson’s disease.
The limitation includes the challenges related These studies show that feature engineering
to data availability as most of the previous processes have the ability to realise exact-
research and projects have used the same ness proportionality that can improve health-
datasets for training and testing so it reduces care quality by enabling accurate and timely
their accuracy in some or the other way. diagnosis.
They have offered a solitary platform with a
complicated user interface that makes it feel References
hard for users to access the platform for a
single illness prediction. Overcoming these [1] J. Jasmin, R. Jenisha, S.G. Rahul, Shriram,
S.S. KavyaSruthi, A.K. Santhiya, K. Kaavy-
limitations in future work entails improving
atamizhan MR. T. Goutham, Parkinson’s
data collection from various mediums. To Disease Prediction Using Deep Learning,
enhance the accuracy of the system, the best Journal of Xi’an Shiyou University, Natural
supervised learning algorithm for each kind Science Edition, 2023.
of disease will be selected from a range of [2] Poojaa C, John Sahaya Rani Alex, Early
algorithms. There will be a streamlined and Detection of Parkinson’s Disease using
passive experience for patients and health- Motor symptoms and Machine Learning,ar
care professionals thanks to the provision Xiv:2304.0925,Cornell University, 2023.
of a more user-friendly and unified platform [3] Parshant, Anu Rathee, Multiple Disease
for a range of disease diagnoses. In the end, Prediction Using Machine Learning, Iconic
this will improve patient care and outcomes Research and Engineering Journals, Ire
by facilitating accurate disease detection and 1704650, 2023.
early intervention. [4] Valle Harsha Vardhan, Uppala Rajesh
Kumar, Vanumu Vardhini, Sabbi Leela Var-
alakshmi, A. Suraj Kumar, Heart Disease
5. Conclusion Prediction Using Machine Learning, Journal
Of Engineering Sciences, Vol 14 Issue 04,
In its conclusion, the paper discusses the 2023.
development of a “Multi Disease Prediction [5] T. Venu Gopal, Heart Disease Prediction
Web App” that forecasts multi-systematic dis- Using Machine Learning Techniques, Issn
eases simultaneously, incorporating glucose- 2457-0362, Volume 13, Issue 07, Jul 2023.
intolerance, liver disease, movement disorder, [6] Ruhul Amin, Rubia Yasmin, Sabba Ruhi,
and cardiac illness, using machine learning Habibur Rahman, Shamim Reza, Predic-
algorithms. The system’s ability to provide tion of chronic liver disease patients using
precise and timely disease projections is what integrated projection based statistical fea-
makes it so important; these forecasts can ture extraction with machine learning algo-
help with early intervention and improve rithms, Informatics in Medicine Unlocked,
Elsevier; 2023.
472 Innovations in Computational Intelligence: Trends and Technologies
[7] Deepika Bhupathi Christine Nya-Ling Tan, [14] Mitushi Soni, Sunita Varma, Diabetes Pre-
Sreenivas Sremath Tirumula, Sayan Kumar diction using Machine Learning Tech-
Ray, Liver Disease Detection Using Machine niques, International Journal of Engineering
Learning Techniques, Research Gate, 2022. Research and Technology (IJERT), Vol. 9
[8] Azizkhan F Pathan, Muskan M, Pooja K, Issue 09, September-2020.
Manu B M, Shreya B H, Machine Learning [15] Chayakrit Krittanawong, Hafeez Ul Has-
for Parkinson’s Disease Prediction, Interna- sanVirk, Sripal Bangalore, Zhen Wang, Kipp
tional Journal of Engineering Research and W. Johnson, Rachel Pinotti, HongJu Zhang,
Technology (IJERT) ISSN: 2278-0181 Pub- Scott Kaplin, Bharat Narasimhan, Takeshi
lished by, www.ijert.org ICEI - 2022 Confer- Kitai, Usman Baber, Jonathan L. Halperin
ence Proceedings. and W. H. WilsonTang, Machine learn-
[9] K. Arumugam, Mohd Naved Priyanka P. ing prediction in cardiovascular diseases:
Shinde Orlando Leiva-Chauca Antonio a metaanalysis, Scientific Reports, Pubmed
HuamanOsorio Tatiana Gonzales-Yanac, (2020).
Multiple disease prediction using Machine [16] Rudra A. Godse, Smita S. Gunjal, Karan
learning algorithms, Materials Today: Pro- A. Jagtap, Neha S. Mahamuni Suchita
ceedings Elsevier.com, 2021. Wankhade Multiple Disease Prediction
[10] Cameron R. Olsen Robert J. Mentz, Kevin J. Using Different Machine Learning Algo-
Anstrom David Page, Priyesh A. Patel MDc rithms Comparatively, IJRCCE, 2019.
Durham Charlotte, Clinical applications of [17] Priyanka Sonar, K. JayaMalini, Diabetes Pre-
machine learning in the diagnosis, classifica- diction Using Different Machine Learning
tion, and prediction of heart failure, Ameri- Approaches. 2019 IEEE, 3rd International
can Heart Journal Volume 229, Number 0 Conference on Computing Methodologies
Olsen, 2020. and Communication (ICCMC).
[11] Akkem Yaganteeswarudu, Multi Disease [18] H. Abbas, L. Alic, M. Rios, M. Abdul-Ghani
Prediction Model by using Machine Learn- and K. Qaraqe, Diabetes Prediction: A
ing and Flask API, Proceedings of the Fifth Deep Learning Approach. 2019 IEEE 32nd
International Conference on Communica- International Symposium on Computer-
tion and Electronics Systems (ICCES 2020), Based Medical Systems (CBMS), 2019, pp.
2020. 567-570.
[12] A. Sivasangari, Baddigam Jaya Krishna [19] Deepti Sisodiaa, Dilip Singh Sisodiab, Pre-
Reddy, Annamareddy Kiran, P. Ajitha, diction of Diabetes using Classification
Diagnosis of Liver Disease using Machine Algorithms, International Conference on
Learning Models. 2020 Fourth Interna- Computational Intelligence and Data Sci-
tional Conference on I-SMAC (IoT in Social, ence (ICCIDS 2018).
Mobile, Analytics and Cloud) (I-SMAC). [20] Chieh-Chen Wu Wen-Chun Yeh, Wen-Ding
[13] Archana Singh, Rakesh Kumar, Heart Dis- Hsu Md. Mohaimeu Islam, Phung. Anh
ease Prediction Using Machine Learning (Alex) Nguyen Tahmina Nasrin Poly Yao-
Algorithms. 2020 IEEE, International Con- Chin Wang Hsuan-Chia Yang, Yu-Chuan
ference on Electrical and Electronics Engi- (Jack) Li, Prediction of fatty liver disease
neering (ICE3), 2020. using machine learning algorithms, Com-
puter Methods and Programs in Biomedi-
cine 170 (2019) 23–29, 2018.
66 A comprehensive review on tomato plant
disease detection using convolutional
neural network
AQ1:
Rudra Pratap Singha, Sandeep Kumar Mishrab, Samriddhi Please
check and
Chaurasiac, Sameer Thakurd, and Sachin Yadave confirm
whether
Department of Computer Science and Engineering, Babu Banarasi Das Institute of Technology and the
Management, Lucknow, India author
names
are set
Abstract: Tomato plants, a globally essential crop, are susceptible to a wide range of diseases that can impact correctly.
both crop yield as well as quality. The early identification of these diseases is important for maintaining agri-
cultural productivity. Recently, deep learning (DL) models have gained prominence as a powerful tool for
automatic tomato leaf disease detection. This report offers a thorough examination of several DL approaches
used to address this important agricultural problem. An outline of the significance of tomato crops in world
agriculture as well as the numerous dangers they encounter, such as illnesses and pests, opens the review. It cov-
ers important discoveries and illness detection techniques. These studies use a range of DL models, including
transfer learning, Inception modules, and convolutional neural networks, to identify tomato leaf illnesses with
remarkable accuracy rates that frequently surpass 99%. Using publically accessible datasets, like Plant Village,
and optimizing model performance through data augmentation, regularization strategies, and hyper parameter
tuning are recurring themes in these works. In an effort to increase precision and lower computing complex-
ity, a number of research also investigate the application of hybrid models and feature selection techniques.
The analyzed research show that DL-based methods offer a promising way to protect agricultural productivity
worldwide by revolutionizing disease detection for other crops as well as tomato plants. To sum up, this exten-
sive analysis offers a clear grasp of the state of DL applications in tomato leaf disease detection at the moment.
Keywords: convolutional neural network; deep learning; machine learning; transfer learning
1. Introduction on manual visual inspections and chemi-
cal analysis, are not only time-consuming
In recent years, the agricultural industry has
but also prone to human errors [3]. Modern
witnessed a significant transformation in the
technologies have become more popular as
way plant diseases, particularly those affect-
ing tomatoes (Solanum Lycopersium, L.) are a result of these traditional methodologies’
detected and managed. Tomatoes contribute shortcomings, especially in the area of artifi-
significantly to the agricultural economy [1]. cial intelligence (AI). Plant disease detection
The management of tomato diseases is cru- has seen a significant transformation since the
cial to ensure crop quality, reduce production introduction of AI, particularly with regard
costs, and minimize environmental impacts to developments in ML and image process-
associated with chemical treatments [1]. Tra- ing [2]. Leveraging AI technologies, including
ditional methods of disease diagnosis, relying convolutional neural networks (CNNs), has
[email protected]; [email protected]; [email protected];
a
[email protected]; e [email protected]
d
DOI: 10.1201/9781003606635-66
474 Innovations in Computational Intelligence: Trends and Technologies
proven to be a game-changer in the detection Convolutional neural networks are becoming
and classification of tomato leaf diseases [6]. more and more popular, and deep learning is
CNNs, with their deep learning capabilities now the most popular architecture due to DL
and feature extraction prowess, have become models’ ability to learn pertinent features of
the cornerstone of this transformation, ena- input images at various convolutional levels
bling rapid and accurate disease identification that are comparable to how the human brain
[8]. Several studies have explored the appli- functions. AlexNet, VGGNet16, GoogleNet,
cation of CNNs in diagnosing tomato leaf ResNet152, Inception V3, MobileNetV1, and
diseases, showcasing remarkable successes in MobileNetV2 are among the several CNN
disease classification and recognition. These architectures. Transfer learning is a common
advancements have significantly improved Deep Learning method that was used in several
disease management and have the potential studies. Due to its ability to train deep neural
to revolutionize the agricultural landscape. network models with less data, it is now the
Moreover, CNNs have demonstrated their most widely used deep learning method. Using
effectiveness in various other agricultural transfer learning gives an edge that users are
applications, from the classification of dif- not required to train their deep neural net-
ferent crop types to identifying plant parts work from scratch. The data set used for train-
and monitoring crop health. While previous ing and testing purpose by various papers is
studies have achieved impressive results using the standard plant village data set consisting of
CNNs for disease diagnosis, some key chal- 54,305 numbers of healthy as well as infected
lenges remain unaddressed. Many of these plant leaves. There are 14 distinct plant spe-
studies have relied on datasets obtained in cies represented, with 38 classes of leaf images
laboratory environments, which may not representing both healthy and diseased plants.
reflect the complex and variable conditions There are three uses for the plant village data-
of real field settings [5]. Therefore, there is a set. the first images of colorful leaves. Second,
growing need to investigate the application of background-smoothed leaf photos that have
CNN models under actual field conditions, been segmented. The grayscale pictures come
where environmental factors can significantly last. Each kind was divided into testing and
impact disease identification. The goal of this training data sets with various ratios, such as
paper is to provide an in-depth analysis of the 80–20, 70–30, and 60–40, for improved accu-
current state of the art in CNN-based tomato racy and performance.
leaf disease diagnosis, focusing on both labo-
ratory and field conditions by examining the
currently existing literature, we aim to assess 3. Thematic Overview
the strengths and limitations of CNN models It is beneficial to use convolutional neural
in different scenarios and their potential for networks that have already been trained,
revolutionizing disease management in agri- such as Inception ResNet V2 and Inception
culture. We will also explore the EfficientNet, V3. Several studies look into how various
and MobileNet, and their contributions to dropout rates affect the functionality of the
the field. model. Dropout is a regularization technique
used to reduce the possibility of overfitting,
and experimenting with different dropout
2. Methodology rates shows a careful approach to model
The research objective of tomato plant dis- optimization. The study investigates dropout
ease detection is to develop an efficient and rates ranging from 5% to 50%, but there is
accurate system by which instant diagnosis no in-depth analysis or discussion regarding
of disease in tomato plants can take place. the choice of these specific dropout rates [1].
A comprehensive review on tomato plant disease detection using convolutional neural network 475
The incorporation of the Hard Swish acti- tomato early blight and tomato late blight,
vation function, IB Max module, and other both belonging to the same class [17].
enhancements contributes to improved rec-
ognition accuracy and faster convergence.
This approach increases the versatility and 4. Critical Analysis
practical utility of the proposed methodol- The papers share several common themes
ogy [3]. This research proposes a model using that reflect the state of the field. Most of them
Convolutional Neural Network for detecting utilize publicly available datasets like Plant
and classifying tomato plant diseases, aim- Village [6]. This augmentation is essential
ing to improve crop quality and quantity for for enhancing a model’s ability to generalize
farmers’ benefit. The abstract outlines clear to the wide range of conditions and varia-
objectives and methodology. However, future tions encountered in real-world agricultural
improvements lack specificity, and details on settings [5]. For instance, by applying these
optimization techniques for enhancing test techniques, researchers can simulate differ-
accuracy are not provided [4]. ent angles, lighting conditions, and perspec-
The research introduces an innovative tives, thereby ensuring that the DL models
hybrid diagnostic tool, combining CNN with are robust and capable of accurately identi-
an inception module, to enhance diagnostic fying diseases in a wide array of real-world
accuracy and robustness for tomato diseases. scenarios.
The study emphasizes practical application Transfer learning stands out as a com-
in agriculture, addressing the need for timely mon practice among these studies. Research-
and accurate diagnosis. It acknowledges ers leverage pre-trained deep learning models
inconsistent lighting conditions but over- and fine-tune them for the specific task of
looks challenges like plant growth variabil- tomato leaf disease detection [2]. Transfer
ity. The conclusion mentions future work on learning accelerates the model training pro-
the hybrid network but lacks specifics. Add- cess and enhances its performance by lever-
ing details on planned improvements would aging knowledge from unrelated tasks. This
strengthen the conclusion [5]. approach is particularly valuable in agricul-
The research addresses the significant tural disease detection, where it allows mod-
impact of plant diseases on agricultural els to quickly adapt to the intricacies of plant
production, emphasizing the need for effi- pathology without starting from scratch.
cient and timely detection methods. This Researchers experiment with a variety
aligns with the crucial concerns in the field of CNN architectures, including Inception
of agriculture. The focus on recognizing and models, ResNet, ShuffleNet, MobileNet, and
detecting diseases in specific plants, such as VGG, for disease detection [2]. These archi-
strawberries and potatoes, demonstrates the tectures offer different trade-offs in terms of
practical application of proposed Disease computational efficiency and accuracy. The
Recognition Model. It is a traditional pro- choice of architecture is a crucial decision
cedure for plant disease detection is time- in model design, and these studies illustrate
consuming, costly, and imprecise, while this the diverse strategies adopted by research-
mentions the use of Convolutional Neural ers to optimize their DL models. Researchers
Network for image processing, it lacks spe- employ feature selection methods to reduce
cific details about the architecture or con- the dimensionality of the extracted features,
figuration of the CNN [6]. Certain methods improving classification accuracy and mak-
exhibited a limitation wherein they pro- ing the models more efficient. By selecting
duced false predictions, specifically con- the most relevant features, these studies aim
cerning the fine-grained distinction between to enhance the model’s ability to distinguish
476 Innovations in Computational Intelligence: Trends and Technologies
between different disease classes, ultimately impressive outcomes in illness categorization
improving the overall performance of disease assignments [8]. Utilizing CNNs that have
detection [11]. already been trained, such those learned
Wide range of tomato diseases considered on ImageNet, enhances generalization and
in these studies, including Early Blight, Yel- accelerates convergence. Acquiring great
low Leaf Curl Virus, and various other ill- accuracy has made the transfer of informa-
nesses [1]. This diversity reflects the need for tion from one task to another crucial [2].
comprehensive disease classification models The success of disease classification models
that can address the many potential threats to heavily relies on the quality and diversity of
tomato crops. Researchers also emphasize the the datasets used. The availability of large
application of DL models in real field con- and diverse datasets like Plant Village has
ditions, acknowledging the challenges posed enabled the development of robust models
by environmental factors and variability out- [6]. Dataset curation, data augmentation,
side controlled laboratory settings [2]. While and addressing class imbalances [20] have
controlled laboratory settings are valuable been common strategies to enhance model
for initial model training and evaluation, the performance. The patterns and trends in the
ultimate goal is to deploy these models in real literature highlight the increasing promi-
agricultural fields. This practical approach nence of deep learning, the impact of trans-
ensures that the developed models can per- fer learning, and the importance of dataset
form effectively in the unpredictable and quality. Which also emphasize the potential
dynamic environments of agricultural fields, for technology to enhance crop manage-
where factors like varying lighting conditions ment and contribute to global food secu-
and diverse plant growth stages can impact rity. The findings have implications for the
disease identification. field, emphasizing the need for continued
research, innovation, and the development
of user-friendly interfaces to make advanced
5. Synthesis and Implication agricultural tools more accessible to a wider
The synthesis of findings from various stud- audience.
ies on plant disease detection, particularly in The findings from the reviewed studies have
tomato plants, reveals several overarching several important implications: The success-
conclusions, patterns, trends, and relation- ful deployment of deep learning-based disease
ships that have emerged from the literature. detection systems can significantly improve
These findings demonstrate the evolution crop management. Early disease detection
of the field and its potential for significant allows for timely intervention, reducing the
impact. The growing use of CNNs and deep need for chemical treatments and minimizing
learning methods for disease detection in crop damage [5]. By minimizing crop losses
tomato plant leaves is a definite trend in the and improving crop yield, these technologies
literature [3]. In terms of accuracy and dura- can contribute to addressing the challenges of
bility, deep learning models have continu- feeding a growing global population. As such,
ously beaten conventional machine learning they have a pivotal role in ensuring a stable
techniques. In this discipline, using deep food supply [1]. The development of user-
learning models has become the norm, and friendly interfaces facilitates the adoption of
several architectures have been investigated, these technologies by farmers and agricul-
such as Inception, ResNet, VGGNet, Effi- tural researchers [5]. User-friendly interfaces
cientNet, and others [6]. Transfer learning, can bridge the gap between advanced tech-
which involves optimizing pre trained mod- nology and end-users in agricultural settings.
els for particular tasks, has demonstrated Future studies can focus on improving the
A comprehensive review on tomato plant disease detection using convolutional neural network 477
Table 66.1. Accuracy table detection in plants. By understanding the
S Result analysis transferability of knowledge from diverse
No. Model domains to plant pathology, researchers can
Accuracy Loss
unlock new possibilities for leveraging exist-
1. EfficientNetB2 [8] 99.56 0.0091 ing models and datasets, potentially leading
2. VGG 19 [7] 80.24 0.1870 to more accurate and robust detection sys-
tems. The deployment of CNN-based systems
3. ResNet 15 [2] 97.00 NA in agriculture is becoming more prevalent,
4. NAS Net [8] 93.82 NA emphasizing the importance of develop-
ing energy-efficient and sustainable models.
5. Alex Net [4] 80.70 0.0658 Future research should focus on optimizing
the computational and power requirements
6. Deep CNN [7] 88.17 NA
of CNNs to ensure their long-term viability
7. MobileNetV2 [6] 95.60 0.0921 in resource-constrained environments. This
avenue of research is crucial for achieving
8. InceptionResNetV2 99.22 0.0309
environmentally friendly solutions that can
[1]
operate efficiently in remote or off-grid agri-
9. Inception V [8] 98.42 0.0129 cultural areas. Fine-tuning pre-trained CNN
10. Google Net [5] 99.18 NA models specifically for tomato plant diseases
is another promising direction for future
11. EfficientNetB7 [12] 95.62 NA research. This involves tailoring models to
12. ResNet20 93.70 NA the unique characteristics of different plants
AQ1: :
[12] and their associated diseases. Such customi-
Please
zation can significantly enhance the accuracy
mention 13. Effi Mob-net [13] 98.98 NA
‘Table and generalizability of detection systems
111.1’ in
14. XceptionV4 [14] 99.45 NA
the main
text.
15. Attention-based 98.00 NA 7. Conclusion
Residual CNN [15]
AQ3:
The collective exploration of CNN applica-
Please
16. VGG-16 [16] 99.21 NA tions in the detection of tomato leaf diseases,
provide as presented in the reviewed papers, reveals
source for 17. Re Net 50 [19] 93.57 NA
the table both promise and areas for further develop-
Source: ment in this critical domain of agricultural
research. The studies discussed in this review
robustness of models, addressing the chal- have advanced the field by leveraging vari-
lenges of class imbalances, and expanding the ous DL techniques and methodologies, each
application to other crops. with its own unique strengths and challenges.
The strengths identified in these papers are
notable. Several papers have adopted state-
6. Future Scope of-the-art CNN architectures which includes
Research in the field of tomato plant dis- Inception V3, Inception ResNet V2, VGG-19,
ease detection using CNNs can be advanced EfficientNet, and MobileNetV2, showcasing
through the exploration of transfer learn- the importance of advanced model choices.
ing techniques. Specifically, investigating the The application of transfer learning in which
adaptation of CNN models originally trained pre-trained models are finetuned, holds
on non-plant-related datasets holds promise the potential to significantly accelerate the
for improving the performance of disease training process and enhance classification
478 Innovations in Computational Intelligence: Trends and Technologies
accuracy. The utilization of diverse and exten- Technologies, vol. 11, no. 1, p. 10, Jan. 2023,
sive datasets, including data augmentation, doi: 10.3390/technologies11010010.
contributes to improved model generaliza- [6] Nishant Shelar, Suraj Shinde, Shubham
tion and robustness, reflecting the real-world Sawant, Shreyash Dhumal, and Kausar Fakir,
variability of tomato leaf diseases. In conclu- Plant Disease Detection Using Cnn, ITM
Web Conf. 44 03049, 2022, doi: 10.1051/
sion, the reviewed research papers represent
itmconf/20224403049.
significant strides in the application of DL for
[7] G. Sakkarvarthi, G. W. Sathianesan, V. S.
tomato leaf disease detection. They provide Murugan, A. J. Reddy, P. Jayagopal, and
insights into the potential of using advanced M. Elsisi, Detection and Classification of
CNN architectures, transfer learning, and Tomato Crop Disease Using Convolu-
diverse datasets. To further advance this field, tional Neural Network, Electronics, vol. 11,
researchers must prioritize comprehensive no. 21, p. 3618, Nov. 2022, doi: 10.3390/
performance evaluations, detailed technical electronics11213618.
descriptions, and comparisons with existing [8] S. M. Hassan, A. K. Maji, M. Jasiński, Z.
models. This would facilitate the develop- Leonowicz, and E. Jasińska, Identification of
ment of more robust and efficient tools for Plant-Leaf Diseases Using CNN and Trans-
safeguarding global agricultural produc- fer-Learning Approach, Electronics, vol. 10,
tion and addressing the challenges faced by no. 12, p. 1388, Jun. 2021, doi: 10.3390/
electronics10121388.
tomato crops.
[9] Vidyashree Kanabur, Sunil S. Harakanna-
navar, Veena Puranikmath, Pramod Hullole,
References and Dattaprasad Torse, Detection of Leaf
Disease Using Hybrid Feature Extraction
[1] A. Saeed, A. A. Abdel-Aziz, A. Mossad, M. Techniques and CNN Classifier, 2022, doi:
A. Abdelhamid, A. Y. Alkhaled, and M. 10.1007/978-3-030-37218-7_127.
Mayhoub, Smart Detection of Tomato Leaf [10] A. Guerrero-Ibañez and A. Reyes-Muñoz,
Diseases Using Transfer Learning-Based Monitoring Tomato Leaf Disease through
Convolutional Neural Networks, Agricul- Convolutional Neural Networks, Electron-
ture, vol. 13, no. 1, p. 139, Jan. 2023, doi: ics, vol. 12, no. 1, p. 229, Jan. 2023, doi:
10.3390/agriculture13010139. 10.3390/electronics12010229.
[2] O. Attallah, Tomato Leaf Disease Classi- [11] P. Bansal, R. Kumar, and S. Kumar, Disease
fication via Compact Convolutional Neu- Detection in Apple Leaves Using Deep Con-
ral Networks with Transfer Learning and volutional Neural Network, Agriculture, vol.
Feature Selection, Horticulturae, vol. 9, 11, no. 7, p. 617, Jun. 2021, doi: 10.3390/
no. 2, p. 149, Jan. 2023, doi: 10.3390/ agriculture11070617.
horticulturae9020149. [12] M. E. H. Chowdhury et al., Automatic and
[3] R. Zhang, Y. Wang, P. Jiang, J. Peng, and H. Reliable Leaf Disease Detection Using Deep
Chen, IBSA_Net: A Network for Tomato Learning Techniques, AgriEngineering,
Leaf Disease Identification Based on Trans- vol. 3, no. 2, pp. 294–312, May 2021, doi:
fer Learning with Small Samples, Applied 10.3390/agriengineering3020020.
Sciences, vol. 13, no. 7, p. 4348, Mar. 2023, [13] Z. Ullah, N. Alsubaie, M. Jamjoom, S. H.
doi: 10.3390/app13074348. Alajmani, and F. Saleem, EffiMob-Net: A
[4] H.-C. Chen et al., AlexNet Convolutional Deep Learning-Based Hybrid Model for
Neural Network for Disease Detection and Detection and Identification of Tomato Dis-
Classification of Tomato Leaf, Electron- eases Using Leaf Images, Agriculture, vol.
ics, vol. 11, no. 6, p. 951, Mar. 2022, doi: 13, no. 3, p. 737, Mar. 2023, doi: 10.3390/
10.3390/electronics11060951. agriculture13030737.
[5] M. V. Sanida, T. Sanida, A. Sideris, and M. [14] T. Anandhakrishnan and Jaisakthi S. M.
Dasygenis, An Efficient Hybrid CNN Clas- Murugaiyan, Identification of tomato leaf
sification Model for Tomato Crop Disease, disease detection using pretrained deep
A comprehensive review on tomato plant disease detection using convolutional neural network 479
convolutional neural network models. Scal- [18] A. J., J. Eunice, D. E. Popescu, M. K.
able Computing: Practice and Experience, Chowdary, and J. Hemanth, Deep Learning-
vol. 21, no. 4, pp 625–635, 2020. Based Leaf Disease Detection in Crops Using
[15] R. Karthik, M. Hariharan, Sundar Anand, Images for Agricultural Applications, Agron-
Priyanka Mathikshara, Annie Johnson, and omy, vol. 12, no. 10, p. 2395, Oct. 2022, doi:
R. Menaka, Attention embedded residual 10.3390/agronomy12102395.
CNN for disease detection in tomato leaves, [19] H. Orchi, M. Sadik, M. Khaldoun, and E.
Applied Soft Computing, vol. 86, p. 105933, Sabir, Automation of Crop Disease Detec-
2020. tion through Conventional Machine Learn-
[16] S. A. Wagle, H. R, V. Varadarajan, and K. ing and Deep Transfer Learning Approaches,
Kotecha, A New Compact Method Based on Agriculture, vol. 13, no. 2, p. 352, Jan. 2023,
a Convolutional Neural Network for Classi- doi: 10.3390/agriculture13020352.
fication and Validation of Tomato Plant Dis- [20] L. Li, S. Zhang, and B. Wang, Apple Leaf
ease, Electronics, vol. 11, no. 19, p. 2994, Sep. Disease Identification with a Small and
2022, doi: 10.3390/electronics11192994. Imbalanced Dataset Based on Lightweight
[17] P. S. Kanda, K. Xia, A. Kyslytysna, and E. Convolutional Networks, Sensors, vol. 22,
O. Owoola, Tomato Leaf Disease Recogni- no. 1, p. 173, Dec. 2021, doi: 10.3390/
tion on Leaf Images Based on Fine-Tuned s22010173.
Residual Neural Networks, Plants, vol. 11,
no. 21, p. 2935, Oct. 2022, doi: 10.3390/
plants11212935.
67 Building a decentralized non-fungible
token marketplace: Leveraging
blockchain technology for secure digital
asset exchange
Vaibhav Bansala, Shivam Arora, Sonia Deshmukh, and Ankit Saini
KIET Group of Institutions, Delhi-NCR, Ghaziabad, India
Abstract: In the past few years, Non-Fungible Tokens (NFTs) have surfaced as a revolutionary innovation,
captivating the world’s attention by offering a novel way to tokenize and trade unique digital assets and col-
lectibles. This paper presents a comprehensive exploration of the development and implementation of a decen-
tralized NFT marketplace, underpinned by the power of blockchain technology. It harnesses a sophisticated
stack of cutting-edge technologies, including MetaMask, Truffle, Infura, ReactJs, Tailwind CSS, Ganache,
Solidity, Web3js, and Faucet. These technologies have been meticulously integrated to create an ecosystem that
prioritizes security, efficiency, and user-friendliness, redefining the way NFTs are bought and sold in the digital
realm. It delves into the intricate architectural design of the decentralized NFT marketplace, offering in-depth
insights into its underlying infrastructure and the intricate interplay of its constituent technologies. It meticu-
lously outlines the implementation process, illuminating the steps taken to transform this ambitious vision into
reality. The paper also candidly discusses the challenges encountered throughout the development journey,
providing valuable lessons learned.
Keywords: Blockchain technology, collectibles, digital assets, decentralized marketplace
1. Introduction and blockchain is pivotal to appreciating the
disruptive force they represent.
In an era characterized by the rapid digitiza- The paper proceeds by first elucidat-
tion of assets, the emergence of blockchain ing the fundamental concepts of blockchain
technology has ushered in a paradigm shift technology, shedding light on its architecture
in the way we perceive ownership and value. and decentralized nature. A central theme
Non-Fungible Tokens (NFTs) and the decen- throughout this exploration is the concept
tralized digital marketplaces where they are of ownership in the digital age. Traditional
exchanged have emerged as a groundbreaking ownership typically relies on physical mani-
innovation, challenging traditional notions of festations of assets, such as deeds, certificates,
ownership, authenticity, and provenance. By or tangible objects. NFTs, on the other hand,
leveraging the inherent security, transparency, bestow ownership rights to digital entities,
and immutability of blockchain, NFTs have from digital art to in-game items, by anchor-
propelled the digitalization of unique, scarce, ing them to the immutable blockchain. Fur-
and indivisible assets, creating a new dimen- thermore, the decentralized nature of NFT
sion of value in the digital realm. Understand- marketplaces and their global accessibil-
ing the intricate relationship between NFTs ity challenge established intermediaries and
DOI: 10.1201/9781003606635-67
Building a decentralized non-fungible token marketplace: Leveraging blockchain technology 481
gatekeepers, democratizing participation in granting them direct control over their digital
the creation and exchange of digital assets. assets [6].
As we delve deeper into this paper, we will
scrutinize the mechanics of NFTs, exploring 2.2. NFTs and solidity: enabling
the creation, transfer, and provenance track- unique digital ownership
ing of these unique tokens [1].
A unique kind of digital asset known as NFTs
denotes ownership of a particular object or
2. Background piece of material, including digital music, films,
Blockchain technology emerged in 2008 with art, and even in-game goods. NFTs are valu-
the introduction of Bitcoin by an entity named able because of their distinctiveness, particu-
Satoshi Nakamoto, marking the advent of larly in the realm of digital art and collectibles.
decentralized digital currency. This innova- Solidity, a programming language designed
tion combined cryptographic methods with exclusively for Ethereum smart contracts, is
a transparent public ledger, enabling secure essential for crafting NFTs. These smart con-
transactions without centralized control. In tracts are automated agreements where the
subsequent years, blockchain’s applications terms between buyer and seller are encoded
expanded beyond cryptocurrency. Ethereum, directly into lines of code, enabling self-exe-
launched in 2015 by Vitalik Buterin, intro- cution. Solidity allows developers to create
duced smart contracts, empowering devel- custom, secure, and efficient smart contracts,
opers to build decentralized applications including those representing NFTs. By defin-
(DApps) and tailor blockchain functionali- ing the rules for creating, transferring, and
ties. This breakthrough led to the surge of managing NFTs, Solidity ensures that the
Initial Coin Offerings and the creation of digital assets on our platform are not only
diverse blockchain platforms. unique but also secure from potential threats
The technology underwent further refine- and vulnerabilities.
ment, addressing issues related to scalabil-
ity and energy efficiency. Blockchain gained 2.3. MetaMask and Web3.js: user-
prominence across sectors such as finance, friendly access to blockchain
supply chain, healthcare, and voting, prompt-
ing exploration of its potential. Permissioned MetaMask, a popular Ethereum wallet and
and private blockchains gained traction, pro- gateway to the decentralized web, serves as
viding regulated access for businesses and a link between the Ethereum blockchain and
institutions [2–5]. conventional web browsers. Users can use it
to control their Ethereum accounts, including
storing Ether (ETH), the native cryptocur-
2.1. Decentralization rency of Ethereum, and communicating with
Decentralization is a key feature of block- smart contracts and DApps. By integrating
chain technology. Conventional systems MetaMask into our NFT marketplace, we
depend on a central authority, such as a bank empower users to securely store their NFTs,
or government, for transaction verification manage transactions, and participate in the
and validation. In contrast, blockchain func- marketplace with ease [7].
tions through a network of computers called Web3.js, on the other hand, is a JavaScript
nodes, preventing any single entity from hav- library that permits communication between
ing complete control over the system. This web apps and the Ethereum blockchain. By
decentralized approach not only strength- leveraging Web3.js, our platform can seam-
ens security by minimizing the risk of a sin- lessly connect to the Ethereum network with-
gle point of failure but also empowers users, out requiring users to leave their web browsers.
482 Innovations in Computational Intelligence: Trends and Technologies
3. Architecture unique digital assets traded on our platform.
Written in Solidity, this smart contract encap-
Blockchain and IPFS technology are combined sulates the essence of each NFT, including its
in the NFT marketplace system architecture, origin, ownership, and transaction history.
as seen in Figure 67.1, to create a decentral- This smart contract facilitates the creation,
ized, safe, and scalable platform. Blockchain, transfer, and ownership of NFTs. When an
like Ethereum, manages NFT transactions, artist mints their creation, the smart contract
ensuring authenticity and ownership. IPFS generates a unique token, representing the
serves as the data layer, storing NFT assets digital asset. This token is then transferred
in a distributed, tamper-proof manner. Smart between users securely and transparently. The
contracts validate transactions, while IPFS contract enforces the ownership rights, mak-
stores NFT files, such as images and meta- ing each transaction verifiable and tamper-
data, identified by unique CIDs. Backend proof. By deploying this smart contract on
services coordinate interactions between the the Ethereum blockchain, we guarantee the
blockchain and IPFS, enabling seamless data authenticity and provenance of each NFT [9].
uploads and retrieval. Users engage through
secure wallets, signing transactions and inter-
acting with NFTs. This architecture ensures 3.2. Wallet integration: securing
a transparent, user-friendly experience, trans- transactions with MetaMask
forming digital asset trading into a resilient MetaMask integration is instrumental in pro-
and innovative ecosystem [8]. viding a secure and streamlined experience for
users. By connecting their Ethereum wallets
3.1. Smart contract: The heart of to the marketplace, users can manage their
digital ownership funds and NFT collections directly from the
platform. MetaMask acts as a digital wallet,
The Solidity smart contract serves as the storing users’ cryptographic keys and allow-
backbone of our NFT marketplace, embody- ing them to sign transactions securely [10].
ing the rules and standards that define the
3.3. Off-chain data: enhancing
efficiency with Infura
Infura serves as the bridge between our appli-
cation and the Ethereum blockchain, enabling
seamless access to blockchain data without
the need for running a full node. Running
a full node requires substantial computa-
tional resources and storage space, making it
impractical for most users. Infura solves this
challenge by providing reliable and scalable
access to the Ethereum network’s data and
infrastructure. By connecting our application
to Infura, we ensure real-time access to block-
chain data, including NFT ownership records,
AQ1: transaction history, and contract interactions.
Please
provide This off-chain data access enhances the effi-
source ciency of our marketplace, allowing users to
for all the Figure 67.1. System architecture.
figures
retrieve accurate and up-to-date information
Source: instantly.
AUTHOR'S SELF GENERATED
Building a decentralized non-fungible token marketplace: Leveraging blockchain technology 483
4. Experimentation
• ‘User’ be the class representing individu-
als engaging with the platform.
• ‘NFT’ be the class representing non-fun-
gible tokens.
• ‘InfuraService’ be the class serving as an
intermediary for secure interactions with
IPFS via Infura infrastructure.
The process of creating a new NFT can be
expressed as a method ‘createNFT()’ within
the ‘User’ class: Figure 67.2. Working diagram.
createNFT() → NFT(id, metadataCID, Source:
AUTHOR'S SELF GENERATED
imageCID)
Each instance of the NFT class carries essen- Function handleSubmit(e):
tial attributes: e.preventDefault()
• id: Identifier for the NFT. If price is null or price <= 0 then return
• metadataCID: Content Identifier for SetGlobalState(‘modal’, ‘scale-0’)
metadata stored on IPFS. SetLoadingMsg(‘Initializing price
• imageCID: Content Identifier for the update...’) Try:
image file stored on IPFS. SetLoadingMsg(‘Price updating...’)
The ‘InfuraService’ class facilitates the SetGlobalState(‘updateModal’, ‘scale-0’)
interaction with IPFS: Await updateNFT(id: nft.id, cost: price)
SetAlert(‘Price updated...’)
InfuraService → IPFS (metadata, image) Reload the window
Where, metadata represents associated meta- Catch error:
data uploaded to IPFS and image represents Log ‘Error updating price:’, error
the associated image file uploaded to IPFS. SetAlert(‘Update failed...’, ‘red’)
In our NFT marketplace experimentation, Function closemodal():
each unique image undergoes a meticulous SetGlobalState(‘updateModal’, ‘scale-0’)
process. First, the image is captured, ensur- ResetForm()
ing high quality and authenticity. Next, it is
meticulously described, providing valuable Function resetForm(): SetPrice(‘ ‘)
context and information for potential buyers. Return handleSubmit, closeModal,
Subsequently, the image is minted onto the resetForm
IPFS storage system, guaranteeing decentrali- End Function
zation and security.
The working diagram as shown in Figure 5. Results
67.2. can be explained as:
Algorithm: Procedure to update the prices In the analysis of our NFT marketplace lev-
of minted NFT images dynamically eraging blockchain technology, the categori-
Function updateNFT(): cal distribution of minted images is expressed
modal = useGlobalState(‘updateModal’) through a pie chart, providing a mathematical
nft = useGlobalState(‘nft’) depiction of the platform’s artistic landscape.
price = nft?.cost Let H, C, P, M, and A represent the percent-
ages of Handmade, Computer-Generated,
484 Innovations in Computational Intelligence: Trends and Technologies
Photographic, Mixed Media, and Collage Art
categories, respectively. The pie chart shown
in Figure 67.3 Illustrates H=30%, C=25%,
P=20%, M=15% and A=10%.
The dominance of Handmade artworks at
30% signifies a substantial presence of manu-
ally crafted pieces, while Computer-Gener-
ated images contribute 25%, showcasing the
precision achievable through computational
processes. Photographic NFTs, capturing real-
world moments, account for 20%, followed
by Mixed Media at 15%, and Collage Art at
10%. This mathematical representation not
only quantifies the diverse artistic content but
also underscores the platform’s commitment
to fostering a varied and inclusive ecosystem
for both creators and collectors in the vibrant Figure 67.4. Trends in transaction.
world of NFTs. Source:
The graphical representation shown in AUTHOR'S SELF GENERATED
Figure 67.4 serves as an invaluable tool for
discerning and understanding the nuanced
Images and Collage Art exhibit unique trends
dynamics of transactions across distinct
with 3 and 5 transactions respectively, indi-
image categories. The transactional trends
cating a specialized yet engaged community
manifest in a manner that underscores the
appreciating the richness and diversity of
diverse preferences within our ecosystem.
these artistic categories.
Handmade Images, boasting 15 transactions,
The implementation of the NFT market-
stand out as a focal point, reflecting a substan-
place has yielded significant results, showcas-
tial appreciation for artisanal craftsmanship.
ing the potential of decentralized platforms
Computer-Generated Images, with 8 transac-
for digital asset trading. Through the seam-
tions, signify a discernible demand for digi-
less integration of various cutting-edge tech-
tally curated artworks. Photographic Images
nologies, our work has successfully addressed
closely follow with 12 transactions, signaling
key challenges and achieved substantial mile-
a keen interest in capturing and owning real-
stones, emphasizing security, scalability, and
world moments. Meanwhile, Mixed Media
user experience.
5.1. Security enhancement
By utilizing smart contracts written in Solid-
ity, rigorous testing methodologies, and
external security audits, the work ensured
robust security protocols. Integrating secure
user wallet management through MetaMask
and implementing encryption techniques bol-
stered the overall security infrastructure. The
platform successfully prevented unauthorized
Figure 67.3. Image categories minted. access and protected user assets, ensuring a
Source:
AUTHOR'S SELF GENERATED
safe environment for NFT transactions.
Building a decentralized non-fungible token marketplace: Leveraging blockchain technology 485
5.2. Real-time data access User wallets, especially when integrated
with financial applications, represent another
Integration with Infura facilitated real-time
significant security concern. Safeguarding
data access from the Ethereum blockchain.
users’ private keys and sensitive data is critical
Users could verify transaction statuses, view
to prevent unauthorized access and potential
NFT ownership records, and monitor auction
loss of funds. Utilizing industry best practices
progress without delays. This real-time data
such as end-to-end encryption, multi-factor
access enhanced the transparency of the plat-
authentication, and hardware wallet support
form, allowing users to make informed deci-
significantly enhances the security posture of
sions and reinforcing trust in the marketplace.
user wallets.
6. Challenges 7. Conclusion: Pioneering the
The development of a decentralized NFT Future of Digital Asset Trading
marketplace presented several challenges:
This paper stands as a testament to the trans-
formative power of blockchain technology,
6.1. Scalability: overcoming particularly within decentralized NFT mar-
Ethereum’s limitations ketplaces. Through the adept use of advanced
Scalability stands as a paramount challenge tools like MetaMask, Truffle, Infura, ReactJs,
in the world of decentralized applications, Tailwind CSS, Ganache, Solidity, Web3js, and
particularly for Ethereum-based works. The Faucet, we have meticulously crafted a pio-
Ethereum network, while robust, faces limi- neering platform. This innovation transcends
tations regarding transaction throughput and the traditional concept of mere transactions;
processing speed. As the popularity of NFTs it signifies a paradigm shift in our digital
surged, the network experienced congestion, interactions. These technologies harmonize
leading to high gas fees and slower transac- to establish a seamless and highly secure
tion times. By adopting techniques like Opti- environment, fundamentally reshaping our
mistic Rollups, transactions can be batched relationship with digital assets. This paper
off-chain and then settled on- chain, sig- illuminates the profound impact of techno-
nificantly increasing the number of transac- logical integration, revolutionizing how we
tions the platform can handle. This solution trade, perceive, and value digital commodi-
not only reduces gas fees but also speeds up ties in our swiftly evolving digital landscape.
the transaction process, ensuring a smoother In an era where digital transactions are ubiq-
experience for users. uitous, this research not only showcases the
present capabilities of blockchain technology
6.2. Security: fortifying smart but also points towards an exciting future
where the boundaries of digital ownership
contracts and user wallets and interaction are redefined, fostering a new
In the decentralized landscape, security is par- era of possibilities in the digital realm.
amount. Smart contracts, being the backbone
of the NFT marketplace, are particularly vul-
nerable to attacks. Ensuring the integrity and References
safety of these contracts is non-negotiable. [1] Wang, Z., L. Yang, Q. Wang, D. Liu, Z. Xu,
Thorough testing, including extensive unit and S. Liu (2019). Artchain: Blockchain-ena-
tests, integration tests, and stress tests, was bled platform for art marketplace. In 2019
implemented throughout the development IEEE International Conference on Block-
process. chain (Blockchain), pp. 447–454. IEEE.
486 Innovations in Computational Intelligence: Trends and Technologies
[2] Nakamoto, S. (2008). Bitcoin: A peer-to- proof-of-work based blockchain systems.
peer electronic cash system. Decentralized IEEE Network 34 (6), 266–271.
business review. [7] Frauenthaler, P., M. Sigwart, C. Spanring,
[3] Somy, N. B., K. Kannan, V. Arya, S. Hans, A. M. Sober, and S. Schulte (2020). Eth relay:
Singh, P. Lohia, and S. Mehta (2019). Own- A cost-efficient relay for Ethereum-based
ership preserving ai market places using blockchains. In 2020 IEEE International
blockchain. In 2019 IEEE international Conference on Blockchain (Blockchain), pp.
conference on blockchain (Blockchain), pp. 204–213. IEEE.
156–165. IEEE. [8] Kumar, R. and R. Tripathi (2019). Implemen-
[4] Kuzlu, M., M. Pipattanasomporn, L. Gurses, tation of distributed file storage and access
and S. Rahman (2019). Performance analy- framework using IPFS and blockchain. In
sis of a hyperledger fabric blockchain frame- 2019 Fifth International Conference on
work: throughput, latency and scalability. Image Information Processing (ICIIP), pp.
In 2019 IEEE international conference on 246–251. IEEE.
blockchain (Blockchain), pp. 536–540. IEEE. [9] Abuhashim, A. and C. C. Tan (2020). Smart
[5] Liang, S., H. Baozhong, L. Yang, Y. Han, contract designs on blockchain applications.
Z. Song, M. Hao, and F. Jingang (2020). In 2020 IEEE Symposium on Computers
Blockchain-based power grid data asset and Communications (ISCC), pp. 1–4. IEEE.
management architecture. In 2020 Interna- [10] Jian, Z., Q. Ran, and S. Liyan (2021). Secur-
tional Conference on Computer Science and ing blockchain wallets efficiently based on
Management Technology (ICCSMT), pp. threshold ECDSA scheme without trusted
207–211. IEEE. center. In 2021 Asia-Pacific Conference on
[6] Chen, R., I.-P. Tu, K.-E. Chuang, Q.-X. Lin, Communications Technology and Computer
S.-W. Liao, and W. Liao (2020). Endex: Science (ACCTCS), pp. 47–51. IEEE.
Degree of mining power decentralization for
68 CollegeGo: leveraging alumni data for
improved college placements
Avneesh Agraharia, Subha Mishrab, Rudrendra Bahadur Singhc,
Kumar Ronitd, and Pahun Ratnae
Department of Computer Science and Engineering, BBDITM, Lucknow, Uttar Pradesh, India
Abstract: This paper is designed to harness the valuable information and experiences of college alumni to
enhance the placement process. By analyzing alumni data, this project aims to provide students, the placement
cell, and the college administration with critical insights to improve future placements. In a world where tech-
nology advances at an unprecedented pace, complexity often accompanies its benefits. Our platform directly
addresses this challenge by seamlessly integrating the invaluable experiences of former students to guide current
ones toward successful careers. It is a comprehensive solution for today’s information and interaction needs. The
project will gather and centralize data from college alumni who have successfully transitioned into their careers.
This data will include employment details, career trajectories, and post-graduation experiences.
Keywords: Placement automation, Android application, Java SE, review
1. Introduction meet the diverse needs of students and the
dynamic demands of the job market. This
“Our paper stands as a beacon, democ-
disconnect between student aspirations and
ratizing platform, breaking barriers, and
the available opportunities necessitates the
nurturing a realm where alumni limit-
development of a comprehensive and innova-
lessly guide their juniors towards their
tive solution that leverages placement infor-
better future.”
mation, data analytics, and machine learning
It is a pioneering solution that addresses this to revolutionize the way students engage with
question head-on, revolutionizing the way and excel in the college placement process.
students interact with the college placement In our relentless pursuit of overcoming the
process. With its focus on leveraging place- intricate challenges posed by the complex-
ment information, our application is a com- ity, accessibility hurdles, and financial bar-
prehensive platform designed to empower riers within the sphere of communication,
students with the tools, insights, and resources our proposed approach embodies a vision-
they need to navigate the intricate maze of ary paradigm that redefines the very essence
college placements successfully. of technological accessibility. At its core, our
approach is not just about developing a cut-
ting-edge tool.
2. Methodology The proposed work titled “Leveraging
The traditional approach to college placement Alumni Data for Improved College Place-
processes often lacks the precision, personali- ments” outlines the specific steps and objec-
zation, and data-driven insights required to tives to be undertaken in the project.
a
[email protected]; [email protected]; [email protected]; dkumarronit599@
gmail.com; [email protected]
DOI: 10.1201/9781003606635-68
488 Innovations in Computational Intelligence: Trends and Technologies
Platform development: We will initiate Real-time feedback mechanism: Imple-
the development of the Guidance Hub. This ment a system for collecting real-time feed-
platform will serve as the epicenter of Alumni back from companies after placement drives.
driven information and experiences, offer- Seamless output integration: security and
ing a harmonized environment for users to privacy: Prioritize the security and privacy of
access a diverse range of Placement regarding student and company data.
information. Implement robust authentication and
API integration: A pivotal aspect of our authorization mechanisms to ensure only
work involves the creation of robust appli- authorized users can access sensitive
cation programming interfaces (APIs) for information.
each user request. These APIs will facilitate By following this comprehensive approach,
seamless communication between the various the application will enable students to make
tools. informed decisions, enhance their employ-
User authentication: To enhance both ability, and streamline the placement process
security and convenience, we will implement within the college. It will become a valuable
a single sign-on mechanism. This feature will tool for students, the placement cell, and the
empower users to access the information Hub college administration in achieving successful
without the need for multiple logins, stream- placements and fostering a culture of data-
lining the user experience. driven decision-making.
User guidance: Recognizing the varying
skill levels among users, we will provide com-
prehensive tutorials, guides, and tips within 3. Thematic Overview
the platform. These resources will empower The system is now in a more challenging,
both beginners and experienced users to automated solution. Online education and
unlock the full potential.
Proposed workflow approach for applica-
tion: The envisioned workflow for Applica-
tion is meticulously designed to be intuitive,
seamless, and empowering, ensuring users
can harness the power of information effort-
lessly. Here’s an overview of our proposed
workflow approach.
Data collection and integration: Begin
by gathering comprehensive placement data
from the college, including past placement
records, company details, student profiles,
and feedback from companies.
Create a structured database to store this
data, ensuring it is up-to-date and easily AQ1:
accessible. Please
mention
Personalized student recommendations: Figure
Develop algorithms that provide personal- 68.1 in
the main
ized recommendations to students based on text.
their profiles and model predictions. Rec-
ommendations may include skill develop- AQ2:
ment, interview preparation, and specific job Figure 68.1. Proposed approach for application. Please
provide
opportunities. Source: source for
AUTHOR'S SELF GENERATED
the figure
CollegeGo: leveraging alumni data for improved college placements 489
site management can solve these problems by Cross-cultural adaptability: For colleges
improving the registration process, reducing with diverse student populations, there may
manual impact, increasing accuracy, ensur- be a research gap in how an application
ing data security, and producing and deliver- adapts to the unique needs of students from
ing real-time updates to students but it is not different cultural backgrounds.
time efficient and also have no user feedback User engagement and long-term impact:
system [1]. Measuring user engagement and assessing the
Java or Kotlin can be used to develop long-term impact of application on students’
Android applications; Java is the most pop- careers could be a research gap. Understand-
ular language. Android mobile applications ing how frequently students use the platform
can also be developed using frameworks such and its effects on their career trajectory is
as Xamarin and React Native [2]. essential for continuous improvement.
The fog computing paradigm uses dis- Privacy and data security concerns: The
tributed, heterogeneous, and resource- con- project can further explore privacy and data
strained devices at the edge of the network security concerns, ensuring that student and
to efficiently implement latency- and band- company data is rigorously protected while
width-sensitive. MSA shows great potential making the system as transparent and secure
in utilizing fog and cloud resources and creat- as possible.
ing new paradigms such as osmotic comput-
ing but third-party apps include which can
threaten User’s privacy [3]. 5. Synthesis and Implications
The working of the application must be short
and easy so that every user feels comfortable
4. Critical Analysis and found it interesting. In our application,
While the application strives to revolutionize the workflow is mentioned below.
college placements and equip students with User interaction: Users interact with the
valuable insights, several research gaps exist Application web application built using
that merit further exploration. XML. They input prompts or select options
Personalized industry-specific recommen- through the intuitive user interface.
dations: Application must provide personal- Backend processing: PHP handles these
ized recommendations based on a student’s user requests and communicates with the
profile and historical data, but there is backend server. Java facilitates interactions
room to delve deeper into industry-specific with the database, managing user accounts,
recommendations. preferences, and content requests.
Impact of skill development on placement Payment processing: When users opt for
success: The project emphasizes skill develop- premium features, PHP communicates with
ment recommendations, but the direct impact APIs securely handles the payment processing
of skill acquisition and enhancement on place- and subscription management tasks, ensuring
ment success remains an area for research. a smooth user experience during the upgrade
Continuous integration of alumni feed- process.
back: While application encourages alumni
involvement, there is a research gap in terms
of how consistently and effectively alumni 6. Module
feedback is integrated into the platform.
Real-time industry insights: Application 6.1. Student module
must offer historical placement data, but the Students can find information on each place-
project could explore ways to provide real- ment that stimulates their interest in the Stu-
time industry insights and job market trends. dent Module.
490 Innovations in Computational Intelligence: Trends and Technologies
By visiting the login page, students can Integration with social media and network-
access the app by providing their personal ing: Allow students to link their application
and academic information. profiles with professional networking sites
Students can read the training materials to enhance their job search and networking
and other announcements after logging in. capabilities.
The student has access to notifications Collaboration with companies: Foster
about companies who visit the campus and stronger partnerships with companies to
different training events. ensure they have easy access to the platform
for recruitment purposes.
6.2. Coordinator module International student support: Develop
resources specifically tailored to international
Upon registering for the application, each students, including visa application guidance
student’s details will be displayed. The stu- and cultural adaptation assistance.
dent cannot use it until the coordinator has
verified those facts.
In addition, the coordinator has access 8. Conclusion
to all other administrative announcements, In conclusion, the application represents a sig-
company information, and specifics on the nificant leap forward in the realm of college
campus drive. placements for students. The journey from
inception to completion has been a testament
6.3. Administrator module to innovation, data-driven decision-making,
and the pursuit of a more student-centric
With total authority over this application,
approach to career development.
the Placement Officer assumes the role of
administrator.
All that the administrator can do is access References AQ3:
the database. Please
[1] Amit Kumar Singh, Ayush Kaushik, G. V. provide
The administrator updates the campus reference
Chandana, A. Chitra, and M. Mala, 2023,
drive details, posts the training materials, and Training and Placement Cell Automation,
citation
through-
arranges placements. Volume 6, ISSN (Online): 2581–5792, Inter- out
national Journal of Research in Engineering,
6.4. Company module Science and Management.
[2] Manisha Patil and Dhanya Pramod, 2023,
With total authority over this application, Enhancing Android Framework Used to
the Placement Officer assumes the role of Detect Unexpected Permission Authorization
administrator. of Mobile Application, Vol. 44, No. 3, Tuijin
All that the company can do is to update Jishu/Journal of Propulsion Technology.
their information. Several mock tests may be [3] Samodha Pallewatta, Vassilis Kostakos,
added, updated, and deleted by the company. and Rajkumar Buyya, 2023, Placement of
Microservices-based IoT Applications in
Fog Computing: A Taxonomy and Future
7. Future work Directions, Vol. 55, No. 321pp, 1–43, ACM
The Application has laid a strong foundation Computing Surveys
[4] Kapil Wagh, Dnyaneshwari Tilekar, Bram-
for enhancing college placements, but there
hesh Chaugule, and Pradip Gorde, 2023,
are numerous avenues for future work and development in the web application for
expansion to make it even more effective. training and placement cell, Vol. 05, Interna-
Here are some areas of future work. tional Research Journal of Modernization in
Engineering Technology and Science.
CollegeGo: leveraging alumni data for improved college placements 491
[5] Mohammad Mainul Islam, Fahimeh [6] Samrudhi Padwal, Samruddhi Ghorpade,
Ramezani, Hai Yan Lu, and Mohsen Nad- P. R. Patil, Manasi Patil, Shraddha Biraje,
erpour, 2023, Optimal placement of applica- Sapana Salunkhe, 2022, E-training and
tions in the fog environment: A systematic placement management system, Vol. 04, No.
literature review, Vol. 174, Journal of Paral- 06, International Research Journal of Mod-
lel and Distributed Computing. ernization in Engineering Technology and
Science.
69 An in-depth review of AI/ML techniques
in emotion-preserving multilingual
video translation for practical cross-
cultural communication
Diwakar Yagyasena, Akshit Kumar Tiwarib, Abhishek Kevin
Gomesc, Aditya Yadavd, and Harsh Anand Guptae
Department of Computer Science and Engineering, Babu Banarasi Das Institute of Technology and
Management, Lucknow, Uttar Pradesh, India
Abstract: Our review paper thoroughly examines utilization of AI/ML algorithm in Emotion-Preserving
Multilingual Video Translation as basis for operationalized Cross-Cultural Communications. Incorporating
recent advancements in speech-to-speech translation, audio-video transformation, and AI/ML based language
automation, we contemplate whether the automated machine can retain emotional tone of language amidst
inter-lingual exchange without considering S2MU and multi-ling Through critical benchmarking, this study
examines these translation models’ quality and effectiveness in translating text from English to any target lan-
guage arguing that cross-cultural communication depends on such tools.
Keywords: Multilingual video translation, emotion-preserving translation, AI-driven language processing,
cross-cultural communication, speech-to-speech translation technology
1. Introduction Most of the existing methods for multilin-
gual video translation (MVT) [6,15] focus on
The world is becoming more interconnected
translating from multiple source languages into
and diverse, and video communication is a
one target language, using a multi-stage pipe-
powerful way to share information and emo-
line of automatic speech recognition (ASR),
tions across cultures. However, language is
not the only barrier to effective cross-cul- machine translation (MT), and text-to-speech
tural communication. Emotions and cultural synthesis (TTS). However, these methods have
nuances also play a vital role in conveying limitations in terms of quality, efficiency, and
the meaning and tone of the message [7]. This scalability. Moreover, they do not address the
paper reviews the challenges and opportuni- issue of emotion preservation, which is crucial
ties of using AI/ML techniques to preserve for authentic and engaging communication [7].
emotions in multilingual video translation This paper proposes a novel approach for
[8]. The goal is to create video content that emotion-preserving multilingual speech-to-
can be understood and appreciated by audi- speech translation (S2ST), which can gener-
ences from different linguistic and cultural ate multiple target languages from a single
backgrounds, without losing the original source language. The approach is based on
emotional and sub textual content [12,13]. recent advances in direct speech-to-masked
[email protected]; [email protected]; [email protected];
a
[email protected]; [email protected]
d
DOI: 10.1201/9781003606635-69
An in-depth review of AI/ML techniques in emotion-preserving multilingual video translation 493
unit (S2MU) and multilingual vocoders, Multilingual video translation for practice
which can reduce the interference from other cross cultural communication. This section
languages and improve the performance of examines the phases and processes necessary
the translation process [1]. The paper also when undertaking a literature review.
discusses the challenges of choosing the Searching process was planned in such a
appropriate unit vocabularies for differ- manner so as to reach every relevant pub-
ent multilingual scenarios, and the potential lication about Multilingual video transla-
applications of the proposed approach for tion. Completely, we had a thorough search
cross-cultural communication. of research databases like ICMPC, IJRPR,
The paper aims to provide a comprehensive IRJET, Google Scholar, Research Gate, iJRA-
overview of the state-of-the-art AI/ML tech- SET among others. Special care was taken
niques for emotion-preserving multilingual to conduct these searches during that spe-
video translation, and to highlight the gaps cific period of time ranging from July 2019
and opportunities for future research. The to July 2023 which makes sure that only the
paper also demonstrates the practical value of latest and recently conducted research on
the proposed approach for creating video con- this particular issues are involved in the final
tent that can transcend language and cultural document. The search strings included appro-
barriers, and convey emotions effectively. The priate phrases/keywords for instance; “Text-
paper is intended for researchers, practition- To Video,” “extract audio from video” etc.
ers, and enthusiasts who are interested in the slightly modified for this purpose. To ensure
intersection of AI/ML, video communication, completeness we used filters that narrowed
and cross-cultural communication. the search results and referenced relevant
The paper argues that emotion preserva- sources.
tion is essential for achieving a competitive Following guidelines were used to select
edge in the global market and fostering mutual papers for inclusion in this review:
understanding and cooperation among dif-
ferent cultures. • The conference focused on the application
Most of the previous works have dealt of the most advanced technologies in the
with translating from multiple source lan- context of Multilingual Video Translation.
guages to one target language [6,15], but this • In this case, it is about articles, conference
paper explores the possibility of expanding papers, publications, and reliable findings.
the multilingual scenarios to support multiple • It was wholly in the English.
target languages [9–11]. • Release them from the period stated, from
Nowadays the ability to watch video July 2019 until July 2023
without loss of feeling, especially in terms
of cross-cultural communication becomes Initially, the titles and abstracts of identi-
vitally important because nowadays people fied papers were carefully analyzed in order
live and work in a world unchained by geo- to assess their relevance to the research topic.
graphical boundaries. Following, the full-length versions of these
With a global village where people have relevant papers were thoroughly investigated
been interlinked, there is a need for successful to see if they met the established criteria for
cross-cultural communication so as to pro- acceptance. When conflicts arose, a process
mote mutual understanding. of consensus was used to reach an agreement
among everyone. Because the literature was
descriptive in nature, quality assessment and
2. Methodology risk of bias analysis were deemed unneces-
An organized approach to identification sary because the primary goal was to conduct
and assessment of required research on an in-depth evaluation of available research
494 Innovations in Computational Intelligence: Trends and Technologies
3. Thematic Analysis
Emotive speech translation, therefore, prom-
ises enriching cross-lingual communication
encapsulated in a creative and fascinating
atmosphere. This paper presents a holistic
approach for building and examining the
key steps in this fascinating procedure. Build-
ing on what has been learnt from previous
studies, we have provided a rigorous basis
for subtitling of video content. The proposed
framework stresses that emotions should be
retained in speech. First, we start by selecting
a video with multiple emotions expressions
AQ1: and audible speech content. This video pro-
Please
mention vides the foundation for the extraction of the
table and audio that serves as the basis for the remain-
figure
citations
ing processing stages. Finally, we dispose of
in the the files that were created on a temporary
main text. basis while observing the ethical rules associ-
ated with resource management. As a result,
AQ2:
Please
we get rid of the messy machine and have a
provide straightforward and productive procedure. In
source Figure 69.1. Methodological Process Flow our humane approach that is incorporated in
for all the Chart.
figures our complete framework, speech translation
AUTHOR'S SELF GENERATED
Source: will mean more than just words in the future.
What it endeavors to do is to include and
rather than analyze the rigorous approach of express the whole scope of human feelings,
individual studies Another peer reviewer was so as to facilitate and deepen cross-cultural
consulted if necessary to provide a different relationships.
perspective.
The core principles, breakthroughs, proto-
types, models, datasets, and any noteworthy
discussions or consequences associated with
the conversion processes between video-to-
audio, audio-to-text, and vice versa have been
compiled from selected papers. To maintain
coherence and extract relevant details for
future research, the data was organized in a
structured manner.
Another peer reviewer was consulted when
necessary to provide an additional perspec-
tive on the research findings. This meticulous
process ensured the inclusion of pertinent
and high-quality literature in the final docu-
ment, contributing to a robust foundation for Figure 69.2. Metric analysis.
further research in the field. Source: AUTHOR'S SELF GENERATED
An in-depth review of AI/ML techniques in emotion-preserving multilingual video translation 495
4. Critical Analysis of Previous Works
Table 69.1. Critical analysis comparison
S.No Paper Analysis
1 Hongyu Gong, Ning Dong, Sravya Popuri, Multilingual vocoders, when combined
Vedanuj Goswami, Ann Lee, and Juan Pino, with language embedding and LID
Multilingual Speech-to-Speech Translation losses, demonstrated the most effective
into Multiple Target Languages, doi. performance in speech resynthesis.
org/10.48550/arXiv.2307.08655, July 2023
[1].
2 Samruddhi Sasavade, Trupti Sutar, Karan This paper looks at how to convert video
Barale, and Digvijay Kambale, Extract the into audio in Python with the MoviePy
Audio from Video by using python, Vol 10, library. Elaborates on the application of
No 06, e-ISSN: 2395-0056 p-ISSN: 2395- Tkinter in developing the graphic user
0072, Jun 2023 [2]. interface for the converter.
3 M. Saraswathi, V. V. S. V. Ronit, and S. The proposed system is an audio and video
Sai Pranav, Vol 4, No 5, pp. 1204–1208, to text conversion system implemented
Implementation of Video and Audio to Text using Tkinter as GUI, MoviePy for audio
Converter, ISSN 2582-7421, May 2023 [3]. extraction, speech recognition for speech
recognition, and open CV and Tesseract for
image recognition.
4 Sanjeev Kumar S., Preksha C., and Pooja The paper proposed a transformer-based
M., Text to Speech Converter Using approach for emotion recognition in
Python, International Conference on audio recordings. Datasets were used for
VLSI, Communications and Computer evaluation to make sure that the model
Communication, Advances in Intelligent operated properly.
Systems and Technologies, doi.org/10.53759/
aist/978-9914-9946-1-2_24, June 2023 [4].
5 Aman Sharma, Vibhor Sharma, Language Study used Seq2Seq approach to train
Translation Using Machine Learning, Volume: its model for translating. The model
03, Issue: 06, e-ISSN: 2582-5208, June 2021 performed well however it had difficult
[14]. cases too.
6 Sagar Nimbalkar, Tekendra Baghele, This paper discusses the development of
Shaifullah Quraishi, Sayali Mahalle, Monali a speech translation system using Google
Junghare, Vol 4, no 5, pp 1204–1208, Speech API and Microsoft Trans API,
Personalized Speech Translation using Google implemented on a Raspberry Pi, utilizes
Speech API and Microsoft Translation API, voice commands via a headset.
International Research Journal of Engineering
and Technology (IRJET), e-ISSN: 2395-0056,
Vol 07, No 05, p-ISSN: 2395-0072, May
2020 [16].
7 B. P. Mishra, Laxman Singh, Spoorti J. Jainar, Researchers employed an electro-acoustic
AQ3: and Nagaraja B. G. A Research of Noise system with active noise control (ANC) to
Please
provide Estimation and Removal Techniques for diminish surrounding noise using adaptive
source Speech Signal, ISSN: 2277-3878, Vol 8, No filtering in MATLAB 2015.
for all the 2S3, July 2019 [17].
tables
Source:
496 Innovations in Computational Intelligence: Trends and Technologies
5. Analyzing Social Parameters in usefulness in real multi-language intercultural
communication. To pave the way for future
Prior Studies
research and advancements, the following
Table 69.2. Parametric analysis comparison recommendations are proposed:
S.No Paper Parametric A. Enhanced emotional intelligence:
analysis One potential area of future study to
1 Hongyu Gong, Ning Languages investigate is the refinement of emotional
Dong, Sravya Popuri, Supported- comprehension and expression capabili-
Vedanuj Goswami, 16 ties of AI/ML translation algorithms. The
Ann Lee, and Juan k-means
proposed approach would involve the use
Pino, Multilingual Values
of advanced sentiment analysis and emo-
Speech-to-Speech - 800
Translation into Multiple - 1000 tions recognition techniques, requiring
Target Languages, - 1500 accurate and contextually relevant com-
doi.org/10.48550/ - 2000 munication tailored to specific cultural
arXiv.2307.08655, July contexts.
2023 [1]. B. Real-time multimodal integration:
2 Sanjeev Kumar S., Audio Real-time multimodal integration is a
Preksha C., Pooja channels: topic that warrants more exploration.
M., Text to Speech - Mono (1) Specifically, the development of a system
Converter Using - Stereo (2.1) that combines artificial intelligence and
Python, International Sample-rates machine learning techniques for the pur-
Conference on VLSI, supported: pose of integrating multi-lingual emotion
Communications -16000 Hz retaining video translation with other
and Computer -20000 Hz modalities, such as augmented reality
Communication, and virtual reality, is of academic inter-
Advances in est. This facilitates the enhancement of
Intelligent Systems
cross-cultural communication encoun-
and Technologies, doi.
org/10.53759/aist/978-
ters by promoting greater depth and
9914-9946-1-2_24, June effectiveness.
2023 [4]. C. Robust privacy and security measures:
3 Aman Sharma and Tokenizer
The increasing prevalence of these tech-
Vibhor Sharma, Factors: nologies necessitates a comprehensive
Language Translation - [:, 1] approach to addressing concerns related
Using Machine Learning, - [:, 0] to privacy and security. There exists a
Vol 03, No 06, e-ISSN: necessity to improve secure methodolo-
2582-5208, June 2021 gies for safeguarding sensitive consumer
[14]. information and ensuring privacy in
Source: cross-cultural interactions.
D. User-centric design:
User-centric design is a crucial aspect of
6. Recommendations for Future development, as it prioritizes the user
Work experience and usability. The subsequent
phase of future research should focus on
Based on the detailed investigation into the user-centric design concepts to ensure the
AI/ML methods and emotion preservation accessibility and usability of such systems
in multilingual video translation, it becomes for a wide range of user demographics.
apparent that this domain has a great potential
An in-depth review of AI/ML techniques in emotion-preserving multilingual video translation 497
E. Scalability and resource efficiency: ideas knows no linguistic or technological
The objective of this research is to investi- barriers. The transformative journey towards
gate approaches for reducing the resource a more connected world is underway, with
requirements of AI/ML techniques in mul- AI and technological progress as the driving
tilingual automated video subtitling. The forces behind this paradigm shift.
aim is to ensure that this technology is
accessible to all possible users.
F. Continual training and adaptation: References
It is imperative to design AI/ML models [1] Hongyu Gong, Ning Dong, Sravya Popuri,
that incorporate ongoing learning and Vedanuj Goswami, Ann Lee, and Juan Pino,
adaptation mechanisms to effectively Multilingual Speech-to-Speech Transla-
address increasing language and cultural tion into Multiple Target Languages, doi.
variances. org/10.48550/arXiv.2307.08655, July 2023.
[2] Samruddhi Sasavade, Trupti Sutar, Karan
Barale, and Digvijay Kambale, Extract the
7. Conclusion Audio from Video by using python, Vol 10,
No 06, e-ISSN: 2395-0056 p-ISSN: 2395-
In conclusion, the comprehensive explora- 0072, Jun 2023.
tion of artificial intelligence (AI) and techno- [3] M. Saraswathi, V. V. S. V Ronit, and S. Sai
logical innovations in this paper illuminates Pranav, Implementation of Video and Audio
their transformative impact on communica- to Text Converter, ISSN 2582-7421, Vol 4,
tion dynamics. The highlighted applications, No 5, pp. 1204–1208, May 2023.
ranging from sentiment analysis to video [4] Sanjeev Kumar S., Preksha C., and Pooja
indexing, demonstrate the multifaceted ways M., Text to Speech Converter Using Python,
in which AI is reshaping our interaction with International Conference on VLSI, Commu-
information and each other. The nuanced nications and Computer Communication,
capabilities of AI, such as emotion detection, Advances in Intelligent Systems and Tech-
nologies, doi.org/10.53759/aist/978-9914-
not only contribute to a deeper comprehen-
9946-1-2_24, June 2023.
sion of human communication but also pave [5] Ali A. AL-Bakhrani, Gehad Abdullah Amran,
the way for more empathetic and responsive Aymen M. Al-Hejri, S. R. Chavan, Ramesh
interactions. Manza, and Sunil Nimbhore, Development
Moreover, the incorporation of AI-pow- of Multilingual Speech Recognition and
ered text extraction from videos signifies a Translation Technologies for Communica-
leap forward in information accessibility. tion and Interaction, International Confer-
By automating the extraction of valuable ence on Advances in Computer Vision and
insights from multimedia content, AI stream- Artificial Intelligence Technologies, doi.
lines workflows and ensures that the wealth org/10.2991/978-94-6463-196-8_54, 2023.
of information embedded in diverse media [6] Polepaka Sanjeeva, Vanipenta Balasri Nitin
formats is readily available. This not only Reddy, Jagirdar Indraj Goud, Aavula Guru
enhances efficiency but also contributes to a Prasad, and Ashish Pathani, TEXT2AV –
Automated Text to Audio and Video Con-
more informed and connected society.
version, E3S Web of Conferences 430,
In essence, the discussed advancements 01027, 2023.
underscore the pivotal role of AI and tech- [7] Al-onazi, B. B., M. A. Nauman, R. Jahan-
nology in fostering a globally connected com- gir, M. M. Malik, E. H. Alkhammash, E.
munity. By breaking down communication M. Elshewey, Transformer-Based Multilin-
divides, promoting mutual understanding, gual Speech Emotion Recognition Using
and augmenting accessibility, these innova- Data Augmentation and Feature Fusion.
tions herald a future where the exchange of
498 Innovations in Computational Intelligence: Trends and Technologies
Appl. Sci. 2022, 12, 9188. doi.org/10.3390/ [13] Rupayan Dirghangi, Koushik Pal, Sujoy
app12189188, September 2022. Dutta, Arindam Roy, Rahul Bera, Manosijo
[8] Sadil Chamishka, Ishara Madhavi, Rash- Ganguly, Dipankar Pariary, and Karan
mika Nawaratne, Damminda Alahakoon, Kumar, Language Translation Using Artifi-
Daswin De Silva, Naveen Chilamkurti, cial Intelligence, Ijraset Journal for Research
Vishaka Nanayakkara, A voice-based real- in Applied Science and Engineering Technol-
time emotion detection technique using ogy, doi.org/10.22214/ijraset.2022.46380,
recurrent neural network empowered feature 2022.
modeling. 1174: Futuristic Trends and Inno- [14] Aman Sharma and Vibhor Sharma, Lan-
vations in Multimedia Systems Using Big guage Translation Using Machine Learning,
Data, IoT and Cloud Technologies (FTIMS). Vol 03, No 06, e-ISSN: 2582-5208, June
Multimed Tools Appl 81, 35173–35194. 2021.
doi.org/10.1007/s11042-022-13363-4, June [15] Philipp Harzig, Moritz Einfalt, and Rainer
2022. Lienhart, Vol 4, No 5, pp. 1204–1208,
[9] Najeema Afrin, G. Aditi Sai, K. Gopi Krishna, Synchronized Audio-Visual Frames with
K. Rathan, Language Convertor Using Fractional Positional Encoding for Trans-
Python, International Research Journal of formers in Video-to-Text Translation, doi.
Modernization in Engineering Technology org/10.48550/arXiv.2112.14088: 2395–
and Science, Vol 04, No 06, June 2022. 0072, December 2021.
[10] Sonali Wangikar, Anjali Bobade, Nam- [16] Sagar Nimbalkar, Tekendra Baghele, Shaiful-
rata Pawar, and Akshay Dhanmani, Text lah Quraishi, Sayali Mahalle, and Monali
to Speech Conversion using Python, Inter- Junghare, Vol 4, No 5, pp. 1204–1208, Per-
national Journal of Advanced Research in sonalized Speech Translation using Google
Science, Communication and Technology Speech API and Microsoft Translation API,
(IJARSCT), ISSN (Online) 2581-9429, Vol International Research Journal of Engineer-
2, No 1, July 2022. ing and Technology (IRJET), e-ISSN: 2395-
[11] Abdul Afzal Pasha and M. D. Irshad Hus- 0056, Vol 07, No 05, p-ISSN: 2395-0072,
sain, Speech to Text Conversion Using May 2020.
Python, International Journal of Scientific [17] B. P. Mishra, Laxman Singh, Spoorti J. Jai-
Research in Engineering and Management, nar, and B. G. Nagaraja, A Research of
doi: 10.55041/IJSREM14934, June 2022. Noise Estimation and Removal Techniques
[12] Rajat Saini, Speech Recognition System, for Speech Signal, ISSN: 2277-3878, Vol 8,
International Research Journal of Moderni- No 2S3, July 2019.
zation in Engineering Technology and Sci-
ence, e-ISSN: 2582-5208, January 2022
70 Data analysis for revenue optimization
in the hospitality sector
Pankaj Kunekara, Parul Patleb, and Anagha Gajaralwarc
Department of Information Technology Vishwakarma Institute of Technology, Pune, India
Abstract: This paper overhauls revenue management and boosts guest experiences in hospitality. Power BI,
Power Query, and DAX analyze varied data sources like booking trends and seasonal patterns. Data cleaning
leads to a solid star schema data model. DAX creates crucial performance metrics. The data becomes an inter-
active dashboard via Power BI, empowering agile revenue strategies and guest experience improvements. This
highlights data’s key role in navigating the competitive hospitality world and fostering guest loyalty.
Keywords: Hospitality sector, revenue management, Power BI, data integration, visualizations, guest experiences
1. Introduction 2. Literature Review
The hospitality industry, at the nexus of tech- This paper [1] employs sales data analysis
nology and guest satisfaction, demands data- with Power BI, SQL, and showcasing the
driven insights for revenue management and effectiveness of data-driven approaches for
exceptional experiences. Amidst evolving cus- sales success.
tomer expectations and fierce competition, This paper [2] emphasizes data warehous-
precision in revenue strategies and personal- ing and online transaction processing (OLTP).
ized guest journeys is paramount. Suggests broader database inclusion for more
However, a critical gap persists while data comprehensive analysis.
from various sources exists, its fragmented This paper [3] focuses on predicting prod-
nature hampers effective decision-making. uct sales in a store, outperforming traditional
Traditional approaches lack efficiency and ML algorithms. Offers insights into optimiz-
fail to unveil nuanced patterns crucial for ing sales predictions for retail companies.
strategic choices. The solution lies in a unified This paper [4] provides five key findings
platform that integrates diverse data, empow- on using big data to enhance customer expe-
ering decision-makers. riences, operational efficiency, and revenue.
Addressing this, the project aims to lever- This paper [5] focuses on how Big Data
age Power BI to bridge this gap, using booking and data analytics contribute to various
trends, customer feedback, and seasonal data aspects of the hospitality industry.
to reshape revenue analytics. AtliQ Grands, This paper [6] summarizes RM research
facing market challenges, seeks to integrate in the hotel industry between 2001 and
“Business and Data Intelligence” through a 2013. Limited by available databases. Rec-
third-party provider to revitalize its opera- ommends investing more in training revenue
tions. The objectives: develop key metrics, managers.
align a dashboard, and offer insightful analy- This study [7] investigates factors driv-
ses, enhancing decision-making capacities for ing BDA adoption by hotels and its impact
AtliQ Grands. on performance. Provides insights for hotel
[email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-70
500 Innovations in Computational Intelligence: Trends and Technologies
managers to leverage BDA for enhanced busi- 3. Using DAX for insights: We use DAX,
ness value and performance. a language for data analysis, to create
This paper [8] discusses how big data ana- important benchmarks for revenue man-
lytics has transformed the sector. Showcases agement. This involves complex calcu-
real-world examples of how big data has lations and aggregations on the data,
benefited various aspects of the hospitality helping us spot trends and patterns cru-
industry. cial for decision-making.
This article [9] evaluates the literature 4. Visualizing data with Power BI: Next, we
regarding the adoption of artificial intelli- import the prepared data into Power BI, a
gence (AI) in the hospitality industry. Empha- strong tool for business analytics. Here, we
sizes potential risks of AI adoption, including turn the data into visuals like dashboards
job displacement and concerns about safety, and reports, making it easier to understand
security, and privacy. and act upon the revenue insights.
This study [10] presents a comprehensive 5. Dashboard creation: Finally, we design
survey on the application of Computational interactive dashboards and reports that
Intelligence (CI) in the Hotel and Travel showcase key indicators and trends in a
Industry. Provides guidelines and recommen- visually engaging way. The goal is to pro-
dations for future research directions and vide decision-makers with an easy-to-use
potential applications in the field. interface to make informed choices based
on the insights we’ve generated.
3. Methodology
1. Data collection: We collect vital infor- 4. Calculations
mation from diverse sources in the hos- We use DAX formulas for key performance
pitality sector—such as booking trends, metrics, employing Week-on-Week (WoW)
customer reviews, and seasonal pat- calculations like Revenue, Occupancy, ADR,
terns. This information is meticulously RevPAR, Realization, and DSRN percentages.
organized into Excel files, including five This ensures precise analysis and facilitates
crucial CSV files: dim_date (date-related strategic decision-making in the hospitality
details), dim_hotels (hotel specifics), sector.
dim_rooms (room types), fact_aggre-
gated_bookings (hotel booking stats),
and fact_bookings (customer-specific 5. Data model
booking info). These files form a com- The project employs a star schema data
prehensive dataset, allowing for detailed model, characterized by a central fact table
reports to enhance revenue management (“fact_bookings”and “fact_aggregated_
and guest experiences. bookings”) capturing key transactional
2. Data preparation and structure: After details. This table is surrounded by dimen-
gathering the data, we clean it thoroughly sional tables (“dim_dates”, “dim_rooms”,
using Power Query to fix errors and “dim_hotels”, and “dim_customers”) pro-
ensure high quality. Then, we build a solid viding context and facilitating exploration.
data model following industry standards, Relationships between tables are represented
creating tables for customer info, book- by lines, enabling efficient data retrieval. The
ings, and timeframes, while also tracking star schema’s structured design enhances
vital metrics like revenue and customer query performance and simplifies data analy-
satisfaction. sis for effective decision-making.
Data analysis for revenue optimization in the hospitality sector 501
Table 70.1. Key metrics list top of form
Sno Metrics Description/Purpose DAX FORMULAE
1 Revenue To get the total revenue_realized. Revenue =
SUM(fact_bookings[revenue_realized])
2 Total Total number of bookings Total Bookings =
Bookings happened. COUNT(fact_bookings[booking_id])
3 Occupancy Total successful bookings Occupancy % = DIVIDE([Total Succesful
% happened to total rooms Bookings],
available. [Total Capacity],0)
4 Average Average ratings given by the Average Rating =
Rating customers. AVERAGE(fact_bookings[ratings_given])
5 Cancellation calculating the cancellation Cancellation % = DIVIDE([Total
% percentage. cancelled bookings],
[Total Bookings])
6 ADR ADR (Average Daily rate) It is the ADR = DIVIDE([Revenue], [Total
ratio of revenue to the total rooms Bookings],0)
booked/sold. It is the measure of
the average paid for rooms sold in
each time.
7 Realization Successful “checked out” Realization % = 1- ([Cancellation
% percentage over all bookings %]+[No Show rate %])
happened.
8 RevPAR RevPAR (Revenue Per Available RevPAR = DIVIDE([Revenue],[Total
Room) It is the revenue generated Capacity])
per available room, occupied or
not.
9 DBRN DBRN (Daily Booked Room DBRN = DIVIDE([Total Bookings], [No
Nights) Average of how many of days])
rooms is booked for a day
considering a period.
10 DSRN DSRN (Daily Sellable Room DSRN = DIVIDE([Total Capacity], [No of
Nights) Average of how many days])
rooms are ready to sell for a day
considering a period.
11 DURN DURN (Daily Utilized Room DURN = DIVIDE([Total Checked Out],
AQ1: Nights) measures the average [No of days])
Please
provide number of rooms successfully
source used by customers per day within
for all the a specific period.
tables
Source:
AUTHOR'S SELF GENERATED
502 Innovations in Computational Intelligence: Trends and Technologies
AQ2:
Please
provide
source
for all the Figure 70.1. Data Model (Star Schema).
figures
Source:
AUTHOR'S SELF GENERATED
Figure 70.2. Final dashboard created.
Source: AUTHOR'S SELF GENERATED
Figure 70.3. Revenue tooltip for trend by week.
Source:
AUTHOR'S SELF GENERATED
Data analysis for revenue optimization in the hospitality sector 503
6. Results
Table 70.2. Insights from dashboard
Filtering Options
City Filter Dynamically filter data for Hyderabad, Mumbai, Delhi, and Bangalore. AQ3:
Please
Month Filter Focus on specific months: May, June, and July 2022. mention
Week Filter Narrow down data by selecting weeks from W19 to W31. table and
figure
Platform Filter Explore data based on booking platforms: Direct Offline, Direct Online, citations
Journey, LogTrip, MakeYourTrip, Others, Tripster. in the
main text.
Property Filter Analyze data specific to AtliQ properties: Bay, Blu, City, Exotica, Grands,
Palace, Seasons.
Room Class Filter Refine analysis by room classes: Elite, Standard, Presidential, Premium.
Visualization Enhancements
Week on Week % Track percentage changes with arrows: Green for increase, red for decrease,
Change yellow for no change.
Hover Tooltip Tooltip shows trends in revenue, RevPAR, occupancy, and realization based
Insights on weekdays and weekends, and ADR, DSRN shows trend based on week
number.
Detailed Report Reports display % changes in RevPAR, occupancy, ADR, and realization
based on weekends/weekdays, and a chart illustrating realization % and
ADR by booking platforms.
Revenue Breakdown Donut chart visually represents revenue percentages for luxury and business
hotels.
Trend Analysis Line chart depicts trends in occupancy %, ADR, and RevPAR.
Key Insights
Occupancy Constant changes suggest dynamic pricing strategies, especially on peak
and RevPAR days.
Fluctuations
ADR Consistency Consistent ADR implies potential for revenue optimization through dynamic
pricing differentials.
Opportunities for Recommend dynamic pricing for peak days and higher rates on weekdays
Revenue Increase and weekends for business and luxury hotels respectively.
Property by Key Categorizes properties based on cities, showcasing key metric changes and
Metrics Table totals for a comprehensive view.
Unfiltered Results from Key Performance Metrics
Note: These insights, inclusive of various performance metrics, provide a comprehensive
understanding of the dataset before applying specific filters or constraints.
Top Revenue Atliq Exotica Mumbai: Highest revenue at 117 million; Atliq Grands Delhi:
Lowest at 36 million.
City Occupancy Atliq Palace Delhi: Impressive 66.3%; Bangalore: Lowest at 44.3%.
Leaders
RevPAR Atliq Exotica Mumbai: Highest at 10,629; Atliq Exotica Hyderabad: Lowest
at 4,061.
Total Bookings Atliq Exotica Mumbai: Highest at 7,251; Atliq Grands Delhi: Least at 3,114.
(continued)
504 Innovations in Computational Intelligence: Trends and Technologies
Table 70.2. continued
ADR Atliq Seasons Mumbai: Highest at 16,597; Atliq Blu Hyderabad: Lowest at
8,676.
DSRN Atliq Exotica Hyderabad: Highest at 128; Atliq Grands Delhi: Lowest at 52.
DBRN Atliq Exotica Mumbai: Highest at 80; Atliq Grands Delhi: Least at 24.
DURN Atliq Exotica Mumbai: Highest at 56; Atliq Grands Delhi: Lowest at 24.
Realization % Ranged from approximately 69% to 71% for all properties.
Cancellation % Consistent range of approximately 24% to 26% for all properties.
Average Rating Atliq Exotica Mumbai: Highest at 4.32; Atliq Seasons Mumbai: Lowest at
2.30.
Total Revenue Cumulative revenue across all hotels in three months: Substantial 1.69
billion.
Source:
AUTHOR'S SELF GENERATED
7. Conclusion [2] Capris, T., Garcia, N. M., Takagi, Y., Pires, I.
M., Melo, P., and Gonçalves, N. J. (February
This research underscores the transformative 2023). Data Analysis for the Development
potential of advanced analytics tools in the of a Power BI Solution for Sales. In 2022
hospitality sector. By leveraging Power BI, International Conference on Data Analytics
alongside Power Query and DAX language, for Business and Industry (ICDABI).
a comprehensive methodology is established [3] Shivankar, S., Mehetar, S., Darade, N., Bhi-
to revolutionize revenue management and manpalli, S., and Dafale, D. (2023, April).
guest experiences. The integration of diverse Global Superstores Sales Prediction and
data sources, meticulous data cleansing, and Data Visualization Using Power BI. Interna-
shaping, and the application of DAX lan- tional Journal of Research in Engineering,
Science and Management, Volume 6, Issue 4.
guage collectively form the cornerstone of
[4] Mnyakin, M. (2023). Big Data in the Hos-
this approach. The resultant interactive dash- pitality Industry: Prospects, Obstacles, and
boards and reports provide decision-makers Strategies. International Journal of Business
with invaluable insights, enabling them to Intelligence and Big Data Analytics.
navigate the competitive landscape with data- [5] Chamoli, V., and Sangeeta. (2019, June). The
driven precision foster guest loyalty through Relevance of Big Data and Data Analytics
enhanced experiences. for the Hospitality Industry. JETIR, Volume
6, Issue 6.
[6] Erdem, M., and Jiang, L. (August 2016).
Acknowledgment An overview of hotel revenue management
We are grateful to Prof. Pankaj Kunekar research and emerging key patterns in the
for his guidance and Prof. Dr. Premanand third millennium. Journal of Hospitality and
Ghadekar, Head of the IT Department, for Tourism Technology, 7(3), 300-312. DOI:
10.1108/JHTT-10-2014-0058.
their support in completing this research.
[7] Carneiro, T., Picoto, W. N., and Pinto, I.
(2023). “Big Data Analytics and Firm Per-
References formance in the Hotel Sector.” Tour. Hosp.,
4(2), 244–256. https://doi.org/10.3390/
[1] Patil, R. C., Desai, A., Jadhav, P., Jamadar, M., tourhosp4020015.
Koli, S., and Shirole, S. (2022). Sales Analy- [8] Author(s) not provided (2023, June 27). Big
sis Using Power BI. International Research Data: The Game Changer in the Hospitality
Journal of Modernization in Engineering, Industry. Blog.
Technology and Science, 04(04), 1561.
Data analysis for revenue optimization in the hospitality sector 505
[9] Limna, P. (2022). Artificial Intelligence (AI) future crisis: using DEA-based performance
in the hospitality industry: A review article. analysis to assess initial pandemic responses
International Journal of Computing Sciences in the Taiwanese hotel industry. Interna-
Research. Advance online publication. doi: tional Hospitality Review. ISSN: 2516-8142.
10.25147/ijcsr.2017.001.1.103. Open Access. Article publication date: 23
[10] Guerra-Montenegro, J., Sanchez-Medina, May 2023.
J., Laña, I., Sanchez-Rodriguez, D., Alonso- [13] Wang, X. L., Heo, C. Y., Schwartz, Z., Leg-
Gonzalez, I., and Del Ser, J. (2021). Com- ohérel, P., and Specklin, F. (2015). Revenue
putational Intelligence in the hospitality Management: Progress, Challenges, and
industry: A systematic literature review Research Prospects. Journal of Travel and
and a prospect of challenges. Applied Soft Tourism Marketing, 32(7), 797–811. https://
Computing, 102, 107082. https://doi. doi.org/10.1080/10548408.2015.1063798.
org/10.1016/j.asoc.2021.107082. [14] Peterlin, J., Meško, M., Dimovski, V., and
[11] Tong-On, Panithan, Siripipatthanakul, Supa- Roblek, V. (2021). Automated content
prawat and Phayaphrom, Bordin, The Imple- analysis: The review of the big data sys-
mentation of Business Intelligence Using temic discourse in tourism and hospitality.
Data Analytics and Its Effects Towards on Sustainable Development, 38(3), 377–385.
Performance in the Hotel Industry in Thai- https://doi.org/10.1002/sres.2790.
land (October 17, 2021). International Jour- [15] Remy, D. (2019). Data Analytics Dilemma at
nal of Behavioral Analytics, 1(2), No. 9, pp. Alpen Hotel. Journal of Information Tech-
1–16, September 2021, Available at SSRN: nology Teaching Cases. SAGE Publications.
https://ssrn.com/abstract=3944077.
[12] Huang, Y.-H. (John), Meyer, B., Connolly,
D., and Strader, T. (2023). Preparing for a
71 Technological solutions for E-waste
management and disposal
Aditya Pratap Singha and Akshat Chaturvedib
Computer Science and Engineering, Babu Banarasi Das Institute of Technology and Management,
Lucknow, India
Abstract: There is a rapidly growing push to combat e-waste a global dilemma having severe impacts on the
environment and human health. So, we are providing a solution to this using the MERN Stack: e-waste facil-
ity locator With this web application, users can search for e-waste facilities with criteria such as location and
e-waste type to recycle at a distance they are willing to travel. Developed at the Triple-I Lab, this locator will
help users easily recycle their e-waste and minimize the environmental impact of e-waste. This research is also
valuable for e-waste recycling centers, government authorities, and environmentalists. This e-waste facility
locator has to be optimized for such applications and will be a unique and original method in addressing the
problem of Disposal of e-waste can contribute significantly to reducing the environmental impact of e- waste
as a whole.
Keywords: e-waste, recycling, web application, location based services, sustainability, user-friendly, environ-
ment, human health
1. Introduction E waste can be managed better which include
collection, transportation and treatment.
Electronic waste (e-waste) is one of the fast- Smart e-waste management is an ICT-
est-growing waste streams in the world, with enabled way of collecting and recovering
over 53 million metric tons generated glob- resources in the most sustainable fashion; it
ally in 2019 [1]. E-waste contains a number involves pretty much everything from internet
of hazardous materials, such as lead, mercury, of things (IoT) [4], big data, artificial intel-
and cadmium, which can leach into the envi- ligence (AI), etc.; all have been around long
ronment and pose a risk to human health and enough to deduce that using these technologies
the environment [2]. can very well optimise the efficiency as well as
Improper disposal of e-waste can lead to the effectiveness of e-waste management. One
a number of environmental problems [3], promising approach to smart e-waste manage-
including: ment is the use of web-based e-waste facility
locators and pickup schedulers. These systems
• Contamination of soil and water allow users to easily find and schedule e-waste
• Release of toxic pollutants into the air pickup from their doorstep [5].
• Damage to ecosystems Web-based e-waste facility locators and
• Increased risk of human exposure to haz- pickup schedulers offer a number of benefits
ardous materials [6], including:
E-waste is a big challenge from the man- • Convenience: Users can easily find and
agement point, but with different technologies schedule e-waste pickup from their doorstep.
[email protected]; [email protected]
a
DOI: 10.1201/9781003606635-71
Technological solutions for E-waste management and disposal 507
• Efficiency: E-waste collection and disposal attention due to its potential to enhance trans-
can be made more efficient by optimizing parency, traceability, and security in e-waste
the routes of pickup vehicles. management systems.
• Transparency: Users can track the status
of their e-waste pickup and disposal. 2.1. Blockchain-based e-waste
• Security: E-waste data can be stored and management systems
transmitted securely using the MERN
stack. Several studies have explored the application
of blockchain technology in e-waste manage-
In addition to these benefits, web-based ment. In a study titled “A Blockchain-Based
e-waste facility locators and pickup schedul- IoT-Enabled System for E-waste Manage-
ers can also help to [5]: ment and Recycling,” researchers proposed a
system that utilizes blockchain to track the
1. Reduce the environmental impact of movement of e-waste from generation to dis-
e-waste: By making it easier for users to posal [4]. The system employs smart contracts
dispose of their e-waste properly, these to automate transactions and ensure compli-
applications can help to reduce the amount ance with regulations.
of e-waste that is landfilled or incinerated. Another study, “Blockchain-Based E-waste
2. Promote a circular economy: Web-based Management: A Framework for Secure and
e-waste facility locators and pickup Traceable E-waste Disposal,” presents a
schedulers can help to promote a circular framework for a blockchain-based e-waste
economy by making it easier for e-waste management system. The framework empha-
to be recycled and reused. sizes secure data storage, transparent trans-
3. Employment Generation: The e-waste actions, and auditable tracking of e-waste
management sector can provide employ- throughout the supply chain [4].
ment opportunities through the creation
and deployment of web-based facility 2.2. IoT-enabled e-waste
locators and web-based pickup schedulers.
management systems
In conclusion, this website-embedded The Internet of Things (IoT) has also emerged
e-waste facility locator-and-pickup scheduler as a promising technology for e-waste man-
is promising in terms of its ability to support agement. IoT devices can be deployed to
efficient and effective and sustainable man- monitor e-waste collection, transportation,
agement of e-waste [7]. and recycling processes, providing real-time
This paper reviews previous work on data and insights to improve efficiency and
smart e-waste management systems and pro- reduce environmental impact [8].
poses a web-based e-waste facility locator In a study titled “IoT-Enabled Smart Waste
and pickup scheduler using the MERN stack Management for Sustainable E-waste Recy-
as a solution for better e-waste management cling,” researchers developed an IoT-enabled
using a web based solution. system for monitoring and managing e-waste.
The system utilizes sensors to collect data on
e-waste volume, location, and composition,
2. Related Work enabling optimization of collection routes
E-waste management and disposal are essen- and recycling processes [5].
tial environmental concerns, and various Another study, “IoT-Based E-waste Man-
technological solutions have been developed agement System for Efficient E-waste Collec-
to address these challenges. Blockchain tech- tion and Recycling,” presents an IoT based
nology, in particular, has garnered significant system for e-waste management. The system
508 Innovations in Computational Intelligence: Trends and Technologies
utilizes RFID tags to identify and track e-waste ultimately contributing to environmental
items, providing realtime information to opti- protection efforts [10].
mize collection routes and reduce transporta- Taking inspiration from the proposed web
tion costs [7]. application for land protection, an e-waste
application could revolutionize how we man-
2.3. AI-powered e-waste management age our electronic waste [11]. Imagine a plat-
form where users can:
systems
Artificial intelligence (AI) is also being • Access educational resources: Learn about
explored for enhancing e-waste management the environmental impact of e-waste,
practices. AI algorithms can analyze large responsible disposal methods, and the
datasets of e-waste data to identify patterns, importance of recycling.
predict waste generation trends, and optimize • Locate e-waste facilities: Easily find
resource allocation [5]. nearby recycling centers and donation
In a study titled “AI-Powered E-waste points, eliminating geographical barriers
Management: Using Machine Learning to and encouraging proper disposal.
Enhance E-waste Recycling,” researchers • Connect with like-minded individuals:
developed an AI-based system for e-waste Share experiences, participate in e-waste
classification and sorting. The system uti- collection drives, and collaborate on com-
lizes machine learning algorithms to accu- munity-driven initiatives.
rately identify and categorize e-waste items, • Support the circular economy: Buy and
improving recycling efficiency and reducing sell refurbished electronics or products
contamination [7]. made from recycled materials, extend-
Another study, “AI-Driven E-waste Man- ing the lifespan of electronic devices and
agement for Predictive Maintenance and reducing the need for raw materials.
Resource Optimization,” presents an AI-
driven system for predictive maintenance By fostering a community-driven approach
of e-waste processing equipment and opti- and leveraging the power of social media, this
mization of resource utilization. The system e-waste application can achieve several key
analyzes historical data and sensor readings goals:
to predict equipment failures and optimize
waste processing schedules [2]. • Increased awareness: Inform users about
the urgency of e-waste management and
2.4. Web-based e-waste management inspire them to take action.
• Improved accessibility: Provide convenient
platforms
and accessible information about e-waste
This research investigates the potential of disposal options, regardless of location.
social media web applications in addressing • Enhanced engagement: Encourage user
life-on-land sustainability objectives. It pro- participation through interactive fea-
poses a web application that educates users tures, fostering a sense of community and
about land protection, helps resolve human- responsibility.
wildlife conflicts, and facilitates the sale of • Promoted sustainability: Support the circu-
eco-friendly products [9]. The MERN stack lar economy by encouraging reuse, repair,
was chosen for development, and testing and responsible recycling of e-waste.
results confirm the application’s function-
ality and browser compatibility. This user- This proposed social media approach
centric platform allows users to post ideas, offers a dynamic and engaging solution to
join projects, and sell eco-friendly products, the growing e-waste problem. By adapting
Technological solutions for E-waste management and disposal 509
the user-centric features and readily accessi- and in future add more features. MongoDB serve
ble platform like the land protection applica- for all E-Waste Collection Centers· Express.
tion, e-waste management can move beyond React executes the core UI components, makes
traditional methods and embrace the power changes when variables change-state and js
of digital communication to achieve environ- manages those server-side operations. js for
mental sustainability [12]. user interfaces and Node. js streams. Server-
side execution is also possible with node. js The
location lies on the Haversine formula, which
3. Methodology calculates the distances by users’ location and
collection centers. Google Maps API aids in
3.1. System development showing the collection centers on map which
methodology gives a rich user experience [15].
In this research project, the development
methodology employed is Feature Driven 3.4. Get user approximate location
Development (FDD), recognized for its effec-
To identify the user’s location, the Google
tiveness and adaptability in system devel-
Maps Geocoder class is employed to convert
opment. FDD offers iterative development,
user-entered addresses into coordinates. The
emphasizing industry best practices and
coordinates assist in marking the user’s loca-
ensuring a focus on quality throughout the
tion on the Google Map interface. Users are
project lifecycle. It involves five sequential
then presented with options to self-drop their
processes: developing an overall model, build-
e-waste or request a pick-up [16].
ing a feature list, planning by feature, design-
ing by feature, and building by feature. The
core of FDD lies in designing and building the 3.5. Self-drop off follow the direction
application incrementally based on identified sign in E
features, fostering an iterative process to meet
Creating directions for self-drop-offs is done
evolving requirements [13].
with the help of DirectService class from the
Google Maps API. If the user has GPS ena-
3.2. Methodology flow chart bled on their phone, these driving directions
The overall development workflow, utiliz- provide turn-by-turn instructions from their
ing the MERN stack (MongoDB, Express.js, current location to the selected e-waste col-
React.js, Node.js), is depicted in the method- lection center.
ology flow chart. The project begins with the
development of an e-waste collection center 3.6. One-click request pick up
locator, serving as the communication element
in the system. The subsequent steps involve One-click request pick-up: The function
locating the user’s approximate location, offer- requires user registration and records some key
ing options for self-drop or pick-up, suggesting information in the database. Users can option-
routes for self-drop-offs, and implementing a ally request a pick-up and enter information
one-click request pickup feature. The flow such as address, phone number and desired
chart provides a comprehensive overview of time. The data is then sent to the e-waste drop-
the system’s sequential processes [14]. off outlet chosen by the system via an email.
This MERN stack based end-end approach
is logical and procedural framework that serves
3.3. E-Waste collection center locator the ultimate purpose of envisioning and develop-
You can get the E-Waste Collection Center App ing an effective and long run e-waste manage-
(E-C3) pr/rrc_ui, also play with FirstMilestone ment solution for Indian e-win scenario [17].
510 Innovations in Computational Intelligence: Trends and Technologies
4. Dataflow 4.3. Get user approximate location
The data flow in the proposed web-based Coordinate Transmission:
e-waste management system is crucial for Coordinates of the user’s approximate
its seamless operation. The MERN (Mon- location are transmitted to the backend server
goDB, Express.js, React.js, Node.js) stack is built using Node.js and Express.js.
employed to manage and transmit data effi- The server processes the incoming data
ciently throughout the system [18]. The data and triggers subsequent actions based on user
flow encompasses various stages, from user preferences [19].
interaction to backend processing and com-
munication between components. Below is 4.4. Suggest route for self-drop off
an extensive overview of the data flow in the
proposed system: User Preference Recognition:
Based on the user’s choice to self-drop
their e-waste, the system utilizes the Direc-
4.1. User interaction tion Service class from the Google Maps
The data flow commences with user inter- API to suggest a route. Route Information
action through the web interface developed Transmission:
using React.js. Users interact with the system The suggested route information is trans-
by accessing the E-Waste Collection Center mitted back to the user interface, providing
Locator, entering their addresses, and choos- step-by-step directions for self-drop-off [7].
ing between selfdrop and pick-up options.
4.5. One-click request pick up
4.2. E-waste collection center locator User Registration:
User Input Processing: Users who choose the pick-up option are
required to register within the system.
1. User-entered addresses are processed by Registration details are securely stored in
the Google Maps Geocoder class to con- the MongoDB database.
vert them into coordinates. Pick-Up Request Processing:
2. The coordinates are then utilized to locate The user’s pick-up request, including
the user’s approximate position on the address, contact number, and preferred time,
Google Map interface. Collection Center is captured by the backend.
Selection: The information is sent via email to the
1. The Haversine formula is employed to selected ewaste collection center, ensuring
calculate distances between the user’s efficient communication.
location and available e-waste collec-
tion centers stored in MongoDB. 4.6. Database interactions
2. Relevant collection centers within a MongoDB Integration:
specified radius are identified and pre-
sented to the user. 1. MongoDB serves as the database for stor-
ing essential information, including col-
Google Maps Integration: lection center details, user registration
The Google Maps API facilitates the data, and pick-up requests.
visualization of collection centers on 2. The data is accessed and updated as
the map, enhancing user interaction and needed, ensuring real-time responsive-
decision-making. ness [14].
Technological solutions for E-waste management and disposal 511
4.7. System Output The E-Waste Collection Center Locator,
central to the system, empowers users to
User Feedback:
make informed decisions about the nearest
Throughout the data flow, the system
collection centers based on their locations.
provides feedback to users, such as sug-
The integration of geolocation technologies,
gested routes, collection center information,
route suggestion for self-drop-offs, and effi-
and confirmation of pick-up requests. Email
cient communication channels for pick-up
Notifications:
requests underscore the system’s commitment
Email notifications are generated to
to user convenience and environmental sus-
inform users about the status of their pick-
tainability [20].
up requests and provide details for successful
As the paper concludes, it is evident that
scheduling.
the proposed web based e-waste management
This comprehensive data flow ensures
system holds significant potential in address-
a smooth and interconnected operation of
ing the current challenges in e-waste disposal.
the web-based e-waste management sys-
By leveraging modern technologies and a sys-
tem, leveraging the MERN stack’s capa-
tematic development approach, the system
bilities for effective data processing and
not only facilitates proper e-waste disposal
communication.
but also contributes to environmental conser-
vation, job creation, and the promotion of a
5. Conclusion circular economy.
In the ever-evolving landscape of waste
The review paper has delved into the com-
management, the proposed system stands as
plexities of e-waste management, addressing
a testament to the power of technology in
the challenges posed by the rapidly growing
driving positive change. Future work could
global e-waste stream. Recognizing the envi-
involve real-world implementations, user
ronmental and health risks associated with
feedback studies, and continuous refinement
improper e-waste disposal, the paper empha-
to ensure the system’s effectiveness in diverse
sized the need for innovative solutions to
e-waste management scenarios. The journey
enhance the efficiency and sustainability of
towards sustainable e-waste management is
e-waste management systems.
ongoing, and this research contributes a valu-
In response to this need, the proposed web-
able step forward in achieving that goal.
based e-waste management system, devel-
oped using the MERN (MongoDB, Express.
js, React.js, Node.js) stack, emerges as a References
promising solution. The adoption of Feature [1] A. A. Acquah et al., “Musculoskeletal disor-
Driven Development (FDD) as the method- der symptoms among workers at an informal
ology ensures an iterative and quality-centric electronic waste recycling site in agbogb-
approach to system development [17]. loshie, ghana,” Int. J. Environ. Res. Public
The data flow analysis outlined in the paper Health, vol. 18, no. 4, pp. 1– 20, 2021, doi:
provides a detailed understanding of how 10.3390/ijerph18042055.
user interactions, geographical data process- [2] A. Kush and A. Arora, “Proposed Solution
ing, and backend communication seamlessly of eWaste Management,” Int. J. Futur. Com-
integrate within the proposed system. The put. Commun., vol. 2, no. 5, pp. 490–493,
utilization of the Google Maps API, Haver- 2013, doi: 10.7763/ijfcc.2013.v2.212.
sine formula, and advanced features such as [3] S. Sivaramanan, “E-Waste Management,
Disposal and Its Impacts on the Environ-
one-click request pick-up demonstrates the
ment,” Univers. J. Environ. Res. Technol.
sophistication and user-friendliness of the
www.environmentaljournal.org, vol. 3, no.
proposed solution.
512 Innovations in Computational Intelligence: Trends and Technologies
5, pp. 531–537, 2013, [Online]. Available: [13] P. Nguyen, “Applying HTTP Protocol of
http://www.environmentaljournal.org/3-5/ TCP communication layer to build login sys-
ujert-3-51.pdf tem by modern web technology,” no. Octo-
[4] A. U. R. Khan and R. W. Ahmad, “A ber, 2022.
[5] Blockchain Based IoT-Enabled E-Waste [14] B. Nguyen, “Improving web development
Tracking and Tracing System for Smart process of MERN stack,” Metrop. Univ. Appl.
Cities,” IEEE Access, vol. 10, no. August, Sci., no. May, p. 52, 2021 [Online]. Avail-
pp. 86256–86269, 2022, doi: 10.1109/ able: https://www.theseus.fi/bitstream/han-
ACCESS.2022.3198973. dle/10024/146232/thesis.pdf?sequence=1
[6] R. Kahhat, J. Kim, M. Xu, B. Allenby, [15] P. Baral, “Role-Based User Access Control in
E. Williams, and P. Zhang, “Exploring Mern Stack,” Bachelor Eng. Inf. Commun.
e-waste management systems in the United Technol., p. 55, 2020.
States,” Resour. Conserv. Recycl., vol. 52, [16] Samikshya Aryal, “Mern Stack With Mod-
no. 7, pp. 955–964, 2008, doi: 10.1016/j. ern Web Practices-Developers Connecting
resconrec.2008.03.002. Application,” 2020.
[7] T. Shevchenko, M. Saidani, Y. Danko, I. [17] “MdAlamin a social platform for software
Golysheva, J. Chovancová, and R. Vavrek, developers: using modern web stack mern,”
“Towards a smart Ewaste system utilizing no. December, 2022.
supply chain participants and interactive [18] M. Keinanen, “Creation of a web service
online maps,” Recycling, vol. 6, no. 1, pp. using the MERN stack,” Metrop. Univ.
1–14, 2021, doi: 10.3390/recycling6010008. Appl. Sci., no. August, 2018, [Online].
[8] M. Ikhlayel, “An integrated approach to Available: https://www.theseus.fi/bitstream/
establish ewaste management systems for handle/10024/153461/Markus_Keinanen.
developing countries,” J. Clean. Prod., vol. pdf?sequence=1
170, no. January, pp. 119–130, 2018, doi: [19] W. A. I. A. Wickramasinghe, K. M. P. M.
10.1016/j.jclepro.2017.09.137. [20] Karunarathna, V. O. V Rajapakse, and D. I.
[9] I. Technology and N. Delhi, “REST- De Silva, “Designing and Developing a Web
ful IoT Service Using the MERN Stack,” Application Using The MERN Stack to Sup-
vol. 3, no. 6, pp. 2609– 2616, 2021, doi: port Biodiversity Conservation and Sustain-
10.35629/5252-030626092616. able Use of Natural Resources,” pp. 1–23,
[10] M. Nguyen, “Minh Nguyen full-stack crud 2023.
application User management system Thesis [21] H. Tran, “Developing a social platform based
centria university of applied sciences Degree on MERN stack Title: Developing a social
Programme March 2023,” no. March, p. 42, platform based on MERN stack Number of
2023. Pages: 41 pages + 2 appendices,” no. March,
[11] M. Bawane, “A Review on Technologies 2021.
used in MERN stack,” Int. J. Res. Appl. Sci. [22] T. Atiemo, S., Faabeluon, L., Manhart, A.,
Eng. Technol., vol. 10, no. 1, pp. 479–488, Nyaaba, L., Schleicher, “Baseline assessment
2022, doi: 10.22214/ijraset.2022.39868. on e-waste management in Ghana,” no.
[12] S. Raju, S. Soundararajan, and V. Loga- June, pp. 7–12, 2016, [Online]. Available:
nathan, “MERN Stack Web Application,” http://sustainable-recycling.org/sustaina-
R.S.C.B, vol. 25, no. 6, pp. 6325–6332, ble-recycling/wpcontent/uploads/2016/07/
2021, [Online]. Available: http://annalso- Sampson_2016_SRI-Ghana.pdf
frscb.ro
72 Comparative study of various
transformation techniques in image
fusion
Kunal Kishor Jhaa, Sumit Kumar Jhab, Hari Om Shanker Mishrac,
Amit Dhawand, Manish Tiwarie
Electronics and Communication Engineering, Motilal Nehru National Institute of Technology,
Allahabad, Prayagraj, India
Abstract: The basic purpose of picture fusion is combining complimentary information from multiple pictures
into a single picture while maintaining the significant and necessary elements from each of the original picture.
The final picture is more suited for both human and machine perception and offers a more precise depiction
of scene than any of individual source pictures. Fusion at pixel level, one of three stages of picture fusion, is
typically used in practice due to preserving the original measured values. A transform-based approach is widely
used to do pixel level picture fusion; hence, the best picture fusion method to maximize productivity of the fused
picture is transform-based picture fusion. The most effective transform methodologies are Discrete Wavelet
Transforms (DWT), Discrete Cosine Transforms (DCT), and hybrid transforms (DWT with NSCT - Non-Sub -
Sampled Contourlet Transform). The merged picture of the deconstructed sub bands is produced using inverse
transforms. In this paper, the various transform techniques have been applied to perform the picture fusion
process. A relatively powerful transform technique, known as curvelet Transform, which is capable of multi-
resolution and multi-direction flexible breakdown and capture the essence of the picture for sub-sampled data,
has been also used for picture fusion. A detailed comparison of these transform techniques has been carried out
with respect to the following performance parameters: The Correlation coefficients, MSE (Mean Square Error)
and PSNR (Peak Signal to Noise Ratio), to evaluate the quality of a fused picture. Low MSE and PSNR ratio
are criterion of a high-quality fused picture. To analyze various transform techniques, fusions of two mode of
pictures, i.e., multimodel and multi-view, have been carried out and detailed comparison has been provided.
Keywords: Picture/image fusion, Discrete Wavelet Transform (DWT), Mean Square Error (MSE), Discrete Co-
sine Transform (DCT), Peak Signal to Noise Ratio (PSNR), Curvelet Transform
1. Introduction and treatment [1]. The primary goal of pic-
ture fusion is to combine significant elements
The picture fusion methodology can be
from several source pictures into a single fused
applied in many different fields. Picture fusion
technology is utilized to enhance the quality picture that includes all pertinent data from
of the data from many picture modalities. This the original picture. Another benefit is that it
approach is used in a variety of industries and can store just a single fused picture instead of
has become popular in those that deal with many source pictures, which contains all the
robotics, computer vision, remote sensing, data available in the given multiple pictures,
microscopic imaging, and medical diagnosis which lowers the cost of storage.
[email protected]; [email protected]; [email protected]; [email protected]; emtiwari@
a
mnnit.ac.in
DOI: 10.1201/9781003606635-72
514 Innovations in Computational Intelligence: Trends and Technologies
The primary uses of picture fusion tech- Transform which is capable of multi-resolu-
nology are the multi-focus imaging system tion and multi-direction flexible breakdown
and the support system for medical diagno- and capture the essence of the picture for sub-
ses and offer fresh knowledge for medical sampled data and the performance is supe-
diagnosis and treatment. The picture fusion rior in the transform domain compared to
technique is employed for a variety of pur- the spatial domain method [12]. Owing to
poses, including segmentation and resolu- these powerful characteristics, the curvelet
tion enhancement [9–11]. The picture fusion transform has been used to fuse images and
technique can make use of the pixel, decision, compare the outcomes with other transform-
feature, and data levels of picture representa- based picture fusion methods. In this paper,
tion. A fused picture in which each pixel rec- picture fusion is performed by applying vari-
ognizes as a collection of pixels from many ous transform approaches. The obtained
source pictures can be produced by apply- fused pictures are analyzed and compared in
ing the pixel level picture fusion approach. the transform domain. It has been observed
The fused picture has the advantage over the that, among the available transform tech-
input picture in that it is more informational niques, the fused pictures resulting from
and contains original information. Even in applying the curvelet transform-based picture
cases where decision level-based fusion and fusion accomplish better compared to results
feature levelbased fusion are in opposition obtained using other transform domain tech-
to one another, picture fusion at pixel level niques. The structure of this document is as
is more useful and simpler to employ. Using follows: Transformation-based photo fusion
the picture fusion technique, the average of methods have been covered in Section II. Sec-
the source picture is calculated on a pixel-by- tions III and IV cover several picture fusion
pixel basis [6], [7], [8]. Spatial and transform techniques and performance metrics, respec-
domains can be used to categories the picture tively. The detailed simulation results and a
fusion process. Some fusion techniques that fit comparison of various picture fusion algo-
within the nature of the spatial domain are the rithms are provided in Section V. The conclu-
Brovey method, Intensity and Hue Saturation sion of the research work done in this paper
(IHS), Principal Component Analysis (PCA), is discussed in Section VI.
and high pass filtering-based approaches
[14], [24–26]. Spatial domain, however, will
result in a notable spatial distortion in the 2. Transform Based Picture Fusion
combined pictures. One way to overcome the Methods
spatial domain limitation is to combine trans- In this particular section, a discussion on dif-
form-based methods with pixel-level fusion. ferent transform-based picture fusion meth-
Examples of transform-based techniques are ods has been presented. Among the various
NSCT, Discrete Cosine Transforms (DCT) available transform techniques, this section
and Discrete Wavelet Transforms (DWT) examines the Non-Sub Sampled Counterlet
[1–5]. The literature review above leads to Transform (NSCT), Curvelet Transform, Dis-
the conclusion that picture fusion has been crete Wavelet Transform, and Discrete Cosine
accomplished using a variety of transform Transform.
techniques. Various transform methods per-
form differently for a required picture fusion
process. Every transform technique has cer- 2.1. DWT-based picture fusion
tain advantages and drawbacks in particular DWT is a mathematical process frequently uti-
applications. However, a relatively power- lized in digital image processing to find local
ful transform technique, known as curvelet features in the image [2]. A two-dimensional
Comparative study of various transformation techniques in image fusion 515
(2-D) picture, such as a 2-D grayscale pic- cosines [17]. Frequency of low block, fre-
ture, can be broken down into a variety of quency of middle block, and frequency of
frequency components known as sub bands high block are three frequencies that make-
using the wavelet transform. It contains two up DCT. Low- frequency block is in first
sub bands, a band of low-frequency, and a block. The information in second block is of
band of high-frequency, when it is applied moderate frequency, while that in the third
to columns [2]. Four sub bands are created block is of high- frequency. In the low- Fre-
when DWT is applied to rows: three sets of quency block, the information on approxi-
accurate coefficients and one set of approxi- mative coefficients can be seen [17], [18]. The
mation coefficients. Each of a deconstructed DCT-based picture fusion technique is vividly
picture’s horizontal, vertical, and diagonal, or illustrated by the schematic representation in
approximate, components are represented by Figure 72.2.
a detailed coefficient [3]. The output of the
low-low (L L), low-high (L H), high-low (H 2.3. Non-sub sampled counterlet
L), and high-high (H H) sub bands is matched transform (nsct)-based image
by four subpictures that include the estimated
and detailed coefficients. Low frequency fusion
decomposed sub band picture incorporates of A variation of the contourlet transform is the
observable components, as opposed to high NSCT. A realistic 2D sparse portrayal of sig-
frequency decomposed sub band picture, nals is provided by this transform, which has
which is composed of the image’s finer ele- several scales anddirections [17]. More effec-
ments, such as texture, lines, and edges [5], tively than other transformations, contour-
[6]. The sub band pictures will break down lets capture abrupt transitions like edges and
the frequency bands at the appropriate levels. geometrical shapes. In compressed sensing,
The schematic representation in Figure 72.1 a fusion method based on non-subsampled
provides a clear illustration of DWT-based contourlet transform (NSCT) was suggested
picture fusion technique. [18]. Technique employs NSCT to decompose
two or more original images. It then obtains
2.2. DCT-based image fusion the sparse matrix using the sparse representa-
tion of the NSCT coefficients and fuses the
The picture is divided into 8x8 blocks by sparse matrices using the absolute valuemaxi-
using DCT. The DCT is used to break down mum strategy.
an image’s spatial frequency into different
AQ1:
Please
provide
source
for all the Figure 72.1. DWT based image fusion. Figure 72.2. DCT based image fusion.
figures
and tables Source: Source:
AUTHOR'S SELF GENERATED
AUTHOR'S SELF GENERATED
516 Innovations in Computational Intelligence: Trends and Technologies
sub picture and the high frequency sub pic-
ture. Some of well-known fusion techniques
are subsequently discussed in the below
paragraphs:
Averaging method
Applying low-frequency (LF) bands to
Figure 72.3. NSCT based image fusion. the average fusion approach. By averaging
Source: AUTHOR'S SELF GENERATED the pixel values from the two input images,
it is accomplished. The (x, y) pixel values of
2.4. Curvelet transform-based image all pictures are taken and added. To find the
fusion average, divide this total by two [3]. Aver-
age value is applied to corresponding pixel in
It has directed parameters, in contrast to the
final picture, and this procedure is replicated
wavelet transform, the curvelet pyramids
for each pixel value. It is employed to remove
have components with a very high level of
precise coefficients and tiny details from an
directional specificity [7]. Additionally, unlike
image. For more fusion method like Gradi-
the isotropic (non-parabolic) scaling of wave-
ent method, LDP, SF method please refer the
lets, curvelet transform is generally based on
literature [3], [5], [7–9], [17].
principle of anisotropic scaling. The follow-
ing Figure 72.4 are the primary steps in pic-
ture fusion using the Curvelet transform, final 4. Performance Parameters
fused picture is generated by carrying out an
Parameters such as MSE, PSNR, and corre-
inverse Curvelet transform on fused coeffi-
lation coefficient are used to assess accuracy
cient after the coefficients have been treated
of fusion. These performance parameters are
according to fusion rules [17]. The curvelet
described below:
flow graph is displayed in Figure 72.4.
4.1. Mean square error
3. Fusion Methods
To compare the input picture with the fused
Different fusion strategies are utilized to picture, MSE is utilized. The MSE of errors,
perform the coefficient of low frequency or MSE, is calculated using the input image’s
mean square multiplied by 255^2.
1
MSE = SMi=i SNj=i (aij – bij)2(1)
M×N
where, ai j is value of pixel at location (i, j) in
input picture and bi j is value of pixel at loca-
tion (i, j) in fused picture.
4.2. Peak signal to noise ratio
The PSNR method measures error. PSNR is
the ratio of highest value that a signal can
have to quantity of noise that could signifi-
cantly distort it and reduce the accuracy with
which it is represented. This is communicated
Figure 72.4. Curvelet transform flow graph. using a logarithmic decibel scale.
Source:
PSRN = 10 log10(MSE
255
)(2)
2
AUTHOR'S SELF GENERATED
Comparative study of various transformation techniques in image fusion 517
4.3. Correlation co-efficient
It identifies the structural similarities between
the input and the combined picture. More
information is preserved when the correla- a b
tion value is higher. The definition of the cor-
relation coefficient is:
SMi=i SNj=i (aij × bij)
CC = (3)
SMi=i SNj=i (aij)2
c d
where ‘a’ represents the variance between the
input picture and its average value overall,
and ‘b’ represents the variance between the
fused picture and its average value overall.
5. Simulation Results e f
The results of image fusion utilizing trans- Figure 72.6. (a) CT Picture, (b) MRI Picture,
form- based approaches are described in this (c) DCT based fusion, (d) DWT based fusion,
segment. Additionally, performance of vari- (e) NSCT based fusion, (f) Curvelet based fusion.
ous picture fusion methods is assessed based Source:
AUTHOR'S SELF GENERATED
on different performance measure parame-
ters. and the experimental findings are shown where the image in Figure 72.5(a) focus on
below. To carry on the above-mentioned left half face and image in Figure 72.5(b)
analysis, two separate examples are consid- focus on right half face of the object. After
ered (Figures 72.5–72.6). performing various transform techniques, e,
In Figure 72.5(a)–(b), two multi-focus g., DCT, DWT, NSCT and Curvelet trans-
input pictures of same object are considered form, the corresponding results in the form
of fused pictures are shown in Figure 72.5(c)–
(f), respectively. The Table 72.1 provides val-
ues of various performance measures, such as,
for PSNR, MSE, and Correlation Co-efficient
(CC) for the different applied transform tech-
a b niques on the multi-focus pictures shown in
Figure 72.5(a)–(b). Ideally, the PSNR and CC
Table 72.1. Performance metrics for transform-
based algorithms’ multi-focus pictures
c d
Transform Co-relation
MSE PSNR
used coefficient
DCT 61.9966 25.4359 0.986
DWT 73.0322 24.7244 0.9849
e f
NSCT 60.9632 25.5089 0.0592
Figure 72.5. (a) Picture 1, (b) Picture 2, (c) DCT
based Fusion, (d) DWT based Fusion, (e) NSCT Curvelet 57.0456 25.7974 0.9874
based fusion, (f) Curvelet based fusion. Transform
Source: Source:
AUTHOR'S SELF GENERATED AUTHOR'S SELF GENERATED
518 Innovations in Computational Intelligence: Trends and Technologies
should be higher and MSE should be lower the fusion rule controls how well the image
for a better-quality fused picture. From the is fused. Theclassic fusion approaches have
Table 72.1, it can be concluded that the bet- two significant drawbacks: picking the coef-
ter fused picture produced by the Curvelet ficients individually and selecting them using
transform has a higher PSNR value and less the same rule. Low and high frequency bands
MSE values. are treated independently by Discrete Wavelet
Similarly, another example of medi- Transforms and Discrete Cosine Transform-
cal image of CT and MRI is considered as based picture fusion method. Compared to
shown in Figure 72.6(a)–(b). The results in alternatives, the LDP and SF based fusion
the form of the fused pictures by applying technique performs better. Using the same
DCT, DWT, NSCT and Curvelet Transform, fusion method to choose low and high fre-
are subsequently shown in Figure 72.6(c)–(f). quency coefficients and decoding process to
Table 72.2 display the values for performance produce a higher-quality fused image. It has
parameters PSNR, MSE, and Correlation Co- been observed that DCT performed better for
efficient (CC). multi-model images and multi-focus images.
It can be observed that, by fusing the CT By referring to the detailed comparative anal-
and MRI images, DWT creates a merged ysis done in this paper, it can be concluded
image that better reveals the soft tissue and that the overall most effective transformation
bone tissue. The DCT-based picture fusion technique is curvelet transform.
technology produces distorted combined
images.
However, the Curvelet transform has been References
shown to be efficient and to deliver superior [1] Saranya, C., and Shoba, S. (2015). Compari-
information in fused images when compared son of Image Fusion Technique by Various
to DWT, DCT, NSCT. Transform based Methods. International
Journal of Engineering Research and Tech-
nology, 4(9).
5. Conclusion [2] Ellmauthaler, A., Pagliari, C. L., and Da
Silva, E. A. (2012). Multiscale image fusion
Various transform techniques have been uti- using the undecimated wavelet transform
lized to perform the picture fusion process with spectral factorization and nonorthogo-
for multi-focus and multi-model pictures. nal filter banks. IEEE Transactions on image
Among these transform techniques, the Cur- processing, 22(3), 1005–1017.
velet transform has been shown to be more [3] Li, S., and Yang, B. (2010). Hybrid multires-
effective in transform-based picture fusion olution method for multisensor multimodal
techniques. In wavelet-based approaches, image fusion. IEEE Sensors Journal, 10(9),
1519-1526.
Table 72.2. Performance metrics for transform- [4] Nunez, J., Otazu, X., Fors, O., Prades, A.,
based approaches’ multi-model images Pala, V., and Arbiol, R. (1999). Multires-
Transfonn MSE PSNR Correlation olution-based image fusion with additive
Used Co-efficient wavelet decomposition. IEEE Transactions
on Geoscience and Remote sensing, 37(3),
DCT 44.8977 16.877 0.8173
1204–1211.
DWT 87.6285 15.6686 0.8071 [5] Kumar, M., and Dass, S. (2009). A total var-
NSCT 87.7 16.9096 0.8105 iation-based algorithm for pixel-level image
fusion. IEEE Transactions on Image Pro-
Curvelet 50.51 16.9832 0.8296 cessing, 18(9), 2137–2143.
Transform [6] Starck, J. L., Fadili, J., and Murtagh, F. (2007).
Source: The undecimated wavelet decomposition
AUTHOR'S SELF GENERATED
Comparative study of various transformation techniques in image fusion 519
and its reconstruction. IEEE transactions on on circuits and systems for video technology,
image processing, 16(2), 297–309. 23(9), 1491–1501.
[7] Petrovic, V. S., and Xydeas, C. S. (2004). Gra- [17] Cao, L., Jin, L., Tao, H., Li, G., Zhuang, Z.,
dient-based multiresolution image fusion. and Zhang, Y. (2014). Multi-focus image
IEEE Transactions on Image processing, fusion based on spatial frequency in discrete
13(2), 228–237. cosine transform domain. IEEE signal pro-
[8] Li, S., and Yang, B. (2010). Hybrid multires- cessing letters, 22(2), 220–224.
olution method for multisensor multimodal [18] Zhou, J., Civco, D. L., and Silander, J. A.
image fusion. IEEE Sensors Journal, 10(9), (1998). A wavelet transform method to
1519–1526. merge Landsat TM and SPOT panchromatic
[9] Pajares, G., and De La Cruz, J. M. (2004). A data. International journal of remote sens-
wavelet-based image fusion tutorial. Pattern ing, 19(4), 743–757.
recognition, 37(9), 1855–1872. [19] Wald, L., Ranchin, T., and Mangolini, M.
[10] Nair, T. G., and Sharma, R. (2013, Febru- (1997). Fusion of satellite images of differ-
ary). Accurate merging of images for predic- ent spatial resolutions: Assessing the quality
tive analysis using combined image. In 2013 of resulting images. Photogrammetric engi-
International Conference on Signal Process- neering and remote sensing, 63(6), 691–699.
ing, Image Processing and Pattern Recogni- [20] Mallat, S. G. (1989). A theory for multires-
tion (pp. 169–173). IEEE. olution signal decomposition: the wave-
[11] Liu, S. S., Zhang, X. H., and Zheng, A. (2013, let representation. IEEE transactions on
March). Image fusion algorithm based on pattern analysis and machine intelligence,
wavelet sparse represented compressed sens- 11(7), 674–693.
ing. In Conference of the 2nd International [21] Zhu, R., Li, X., Zhang, X., and Wang, J.
Conference on Computer Science and Elec- (2021). HID: the hybrid image decomposi-
tronics Engineering (ICCSEE 2013) (pp. tion model for MRI and CT fusion. IEEE
1214–1217). Atlantis Press. Journal of Biomedical and Health Informat-
[12] Saranya, C., and Shoba, S. (2015). Compari- ics, 26(2), 727–739.
son of Image Fusion Technique by Various [22] Abdallatif, M. H., Eissa, H., and Benedress,
Transform based Methods. International L. G. (2022, May). Wavelet transform based
Journal of Engineering Research and Tech- image fusion for brain tumor detection. In
nology, 4(9). 2022 IEEE 2nd International Maghreb
[13] Lallier, E., and Farooq, M. (2000, July). A Meeting of the Conference on Sciences and
real time pixel-level based image fusion via Techniques of Automatic Control and Com-
adaptive weight averaging. In Proceedings of puter Engineering (MI-STA) (pp. 283–287).
the third international conference on infor- IEEE.
mation fusion (Vol. 2, pp. WEC3-3). IEEE. [23] Xi, W., Wang, X., and Wang, B. (2022, July).
[14] Liu, S. S., Zhang, X. H., and Zheng, A. (2013, An Image Fusion Method of Wavelet Trans-
March). Image fusion algorithm based on form and Principal Component Analysis
wavelet sparse represented compressed sens- Based on EIT. In 2022 4th International
ing. In Conference of the 2nd International Conference on intelligent control, Measure-
Conference on Computer Science and Elec- ment, and signal processing (ICMSP) (pp.
tronics Engineering (ICCSEE 2013) (pp. 905–909). IEEE.
1214–1217). Atlantis Press. [24] Noor, A., Gaffar, S., Hassan, M. T., Junaid,
[15] Saranya, C., and Shoba, S. (2015). Compari- M., Mir, A., and Kaur, A. (2020, November).
son of Image Fusion Technique by Various Hybrid image fusion method based on dis-
Transform based Methods. International crete wavelet transform (DWT), principal
Journal of Engineering Research and Tech- component analysis (PCA) and guided filter.
nology, 4(9). In 2020 First International Conference of
[16] Lee, I. H., and Choi, T. S. (2013). Accurate Smart Systems and Emerging Technologies
registration using adaptive block processing (SMARTTECH) (pp. 138–143). IEEE.
for multispectral images. IEEE transactions
520 Innovations in Computational Intelligence: Trends and Technologies
[25] Malhotra, R., Kaur, K., and Singh, P. (2021, Hybrid Network with a Transformer Mod-
July). Wavelet based image fusion tech- ule for Medical Image Fusion. IEEE Journal
niques: a comparison-based review. In 2021 of Biomedical and Health Informatics.
6th international conference on communica- [28] Aravinth, J. (2023, May). Alzheimer’s Dis-
tion and electronics systems (ICCES) (pp. ease Prediction by Spatio-Temporal Feature
1148–1152). IEEE. Fusion for MRI Data. In 2023 Third Inter-
[26] Nalwaya, P., Allapakam, V., Karuna, Y., and national Conference on Secure Cyber Com-
Saladi, S. (2023, April). CT-PET Based Image puting and Communication (ICSCCC) (pp.
Fusion Algorithm for Medical Applications. 580–585). IEEE.
In 2023 IEEE 12th International Confer- [29] Candes, E. J., and Donoho, D. L. (1999).
ence on Communication Systems and Net- Curvelets: A surprisingly effective nonadap-
work Technologies (CSNT) (pp. 391–397). tive representation for objects with edges
IEEE. (pp. 1–10). USA: Department of Statistics,
[27] Liu, Y., Zang, Y., Zhou, D., Cao, J., Nie, R., Stanford University.
Hou, R., ... and Mei, J. (2023). An Improved
73 Blockchain transaction processing:
challenges and resolutions
Kamal Kant1,a, Sarvesh Pandey2,b, and Udai Shanker1,c
1
Department of Computer Science and Engingeering; M. M. M. University of Technology,
Gorakhpur, India
2
Computer Science-MMV, Banaras Hindu University, Varanasi, India
Abstract: This paper delves into the transformative shift from traditional databases to blockchain technology
motivated by challenges such as tampering and central control inherent in databases. Traditional databases
(e.g., relational databases) are structured in rows and columns while blockchain is structured as a chain of
blocks secured through cryptographic primitives. Blockchain technology has paved the way for developing more
practical solutions with its features, which include decentralization, security, and immutability etc. We intro-
duce traditional databases and blockchain technology focusing on the shift towards the later. A emphasis is also
placed on the pivotal role of consensus, a notable divergence that renders blockchain unique. Furthermore, we
discuss key consensus algorithms, their challenges, and the opportunities they present.
Keywords: Blockchain, Leader Selection, Privacy, Cryptography, Transparency, Consensus
1. Introduction a desire for solutions that are not based on
relational databases and their underlying
Mission-critical software systems include algorithms—commit processing (Shanker,
database management systems (DBMSs) that Misra, and Sarje, 2006) (Pandey and Shanker,
are complicated and time-sensitive. Today’s 2022) as well as concurrency control (Pandey
database management systems resulted from and Shanker, 2018)—have proven to be an
decades of scholarly and industrial research essential aspect in searching for alternatives
and intensive business software creation. to relational databases and their underlying
As some of the earliest widely used online algorithms.
server systems and database systems helped In previous decade, blockchain has grown
to develop the first design solutions for net- in popularity and matured. Initially, block-
worked services, data management, applica- chains were not designed with workloads for
tions, and operating systems. Early DBMSs online transactions in mind when they were
were among the most significant software sys- first introduced. It is, nevertheless, on the
tems in computer science because they were verge of becoming a general-purpose transac-
often adopted and reinvented, and because tional system. As transactional management
they solved implementation problems that systems, their adaptability is rapidly improv-
were previously exclusive to DBMSs. Rela- ing. Blockchains are adopting the same
tional databases, first introduced by E. F. design methodologies and ideas created and
Codd in 1970 and enabling ACID semantics used for traditional database systems with
and SQL, have been popular for a long time particular adjustments based on the users’
and are still in use today (Shanker, Misra, needs. As a result, one may compare the
and Sarje, 2008). Recently, there has been
[email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-73
522 Innovations in Computational Intelligence: Trends and Technologies
blockchain system to a safe classical data- Blockchain is an emerging technology and
base system. The growth of blockchain has it is classified into three categories: private, in
stimulated demand for decentralized appli- which nodes cannot enter without permission
cations across several industries including and is basically for one organization; public,
supply chain management, insurance, and in which nodes can enter without permission;
financial markets. Many problems about and consortium, which is a mixture of private
the blockchain have been raised since the and public blockchain.
invention of Bitcoin (El-Hindi, et al., 2019); This is how the remainder of the paper
examples include “blockchain as a data- is structured. Section II discusses the transi-
base” or “contrast between blockchain and tion from database to blockchain. Section
database” (Nakamoto, Bitcoin). Blockchain III depicts the classification of the consensus
is unlike traditional databases for several algorithm. Section IV describes the technical
reasons including decentralization, chained difficulties and recent advancements in this
hash cryptographic security, lack of admin- field, and Section V concludes the paper.
istration control, immutability, and the abil-
ity to transfer without requiring permission
from a central authority. To preserve these 2. Mutual Influence and
differences, a lot of business applications Development from Database to
have added blockchain technology to their Blockchain
traditional database storage option. This
A database is currently integrated into many
has made their implementation more secure
blockchain platforms. The blockchain and
and necessitated less industry-party trust.
the database may achieve various functions
Despite the preceding characteristics, block-
and features by cooperating. We may exam-
chain needs some functionalities present in
ine how blockchain differs from actual data-
traditional databases. Combining features of
base systems if we position it as a database
a blockchain with a database enhances the
that provides a storage mechanism. Here are
program’s effectiveness and security..
some primary areas where the properties of
Any two parties in a multi-party situation
blockchain and databases differ; however,
may conduct transactions utilising blockchain
both can be benefitted from and strengthen
technology without the need for a middleman
one another. Compared to database transac-
or third party. By agreement or consensus, it is
tions, the latency of blockchain transactions
altered and kept safe using encryption. Data
is typically relatively high. With the usage of
is stored in blocks rather than rows and col-
a database, the latency can be reduced to the
umns as in conventional databases with each
required level.
block holding the information for the current
Sequential isolation is required in block-
block and the previous block’s hash values.
chain transactions, which can be achieved via
The initial block on the blockchain is called
consensus methods that provide robust con-
the genesis block. A miner is a node that is
sistency. There is a well-known approach for
responsible for mining the block and can
concurrency control and two-phase locking
perform compute-intensive operations. Like-
(Kung and Robinson, Jun. 1981). New block-
wise addresses in the linked list data struc-
chain databases, nevertheless, like Blockchain
ture, hash values chain all blockchain blocks.
DB (El-Hindi, et al., 2019) due to MongoDB
When a new block is required and also there
(MangoDB, “Building enterprise-grade block-
is unanimity among all nodes using different
chain databases with mon-godb,” 2017 nov),
kinds of consensus algorithms, the new node
are beginning to offer novel blockchain-based
is attached to the chain’s tail.
transaction processes. Most databases enable
complicated inquiries, but Multi- Version
Blockchain transaction processing: challenges and resolutions 523
Table 73.1. Comparison of traditional and is faulty (no reply) or malicious (reply with the
blockchain wrong answer), the whole system collapses or
Domain Database Blockchain is going to be damaged. More problems have
Features domain domain not been rectified in the traditional database
High throughput yes To be until now; so, we are moving from the con-
carried out ventional database to blockchain technology.
In blockchain technology, it is the distributed
Scalability yes To be
carried out
electronic ledger whose exact copy is kept
to every single network node that is there.
Transactions Low High
It allows the platform to transition between
Decentralization Blockchain- yes two properties without using the third party,
inspired which means that it permanently eliminates
databases the third party. Blockchain has many more
Immutability preventing yes characteristics that separate it from the dif-
deletions ferent other ones.
and keeping
track of
updates 3. Consensus Algorithm
transferring digital New yes Classification
assets databases
AQ1: distributed Consensus algorithms are to make a final
Please
provide in the decision to achieve a consensus. First, there
source blockchain is the category of proof-based consensus
for all the manner algorithms, which necessitates that a node
tables
Source: compete with other nodes and demonstrate
AUTHOR'S SELF GENERATED a higher level of qualification to commit the
transactions. Proof of Work (PoW) (Naka-
concurrency control (Bernstein and Good- moto, Bitcoin: Review, 21260, 2008) (Mishra,
man, 1983) can support provenance queries Kushwaha, and Misra, 2010) (Buterin and
on historical data. Griffith, 2017), Proof of Stake (PoS) (Kiayias,
Decentralization is now possible with the Russell, and David, 2017) (Li, Andreina, and
introduction of new blockchain-style data- Bohli, 2017), Proof of Burn (PoB) (Kiayias,
bases which have a wide range of potential and Zindros, 2020) etc. belong to this cate-
applications for which it could be used. The gory of algorithms. Because the commitment
immutability of transactions or their resist- is decided by which committed result obtains
ance to tampering, is another outstanding the majority of votes, the other type of algo-
aspect of blockchain. Database systems can rithm is voting based. Paxos (Lamport, 2012)
achieve this tamper-resistance by imple- (Shi and Wen-Cheng, 2012), Raft (Biryukov,
menting measures that prevent database Alex, Khovratovich, and k, 2014), PBFT and
deletes and modifications (Dwyer, 2015) its variants (Kotla, Alvisi, Dahlin, Clement,
(Conti, Kumar, Lal, and Ruj, Fourth Quar- and Wong, 2009) (Konnov, Makarov, Pozdn-
ter 2018) (Berentsen and F.Schar, 2018) yakova, Safin, and Salagaev, February, 1999)
(Maurer, 2016). Table 73.1 shows the sum- (Bessani, Sousa, and Alchieri, 2014) (Aublin,
marized comparison of the database and Mokhtar, and Quéma, 2013) (Stellar, 2015,
blockchain domain. [Online]) (Yin, Malkhi, Reiter, Gueta, and
In the traditional database, there are many Abraham, 2018) etc. belong to this category.
problems or drawbacks. It is organized or The following algorithms are discussed in
managed by a single entity. If this single entity Table 73.2 based on their characteristics.
524 Innovations in Computational Intelligence: Trends and Technologies
Table 73.2. Comparative analysis of consensus algorithms based on characteristics
Algorithms PoW [5,15,16] PoS PoB [20] Paxos [21,22] Raft [23] BFT and
[17–19] its Variants
Characteristics [24–29]
Model of Trust trustless trustless trustless Semitrusted Semitrusted Semitrusted
Governance Permissionless Both Both Permissioned Permissioned Permissioned
Transaction Have chance Have Have Instant Instant Instant
Finality chance chance
Decentralization ↑ ↑ ↑ ↓ ↓ ↓
Scalability ↑ ↑ ↑ ↓ Moderate ↓
Reward 1 1 1 Mostly 0 Mostly 0 Mostly 0
Realization of a Bitcoin, Peercoin, Slimcoin NA NA Byzcoin,
network Ethereum 808Coin Tendermint
↑=high; ↓=low; NA = Not Available; 1 = Reward Based Consensus Method; Mostly 0 = Mostly No Award
Source: AUTHOR'S SELF GENERATED
4. Research Challenges and blocks. A small number of transactions may
be delayed due to the block size (Zheng, Xie,
Opportunities
Dai, Chen, and Wang, 2018).
One of the innovative developments is block- Privacy Leakage. Since any user on the net-
chain technology, which is being acclaimed. work may see the balances and specifics of
Despite various worries about the technology public keys, the principal weakness of block-
preventing its mainstream acceptance, many chain is the disclosure of transaction privacy.
sectors are implementing blockchain technol- One of the methods that has been proposed is
ogy because of the listed factors below. obtaining anonymity in the blockchain. Mix-
Initial Cost. Blockchain technology’s open- ing solution and anonymous solution are the
ness promises long-term advantages. The terms that apply to this (Bastin, 2018).
software needed to set up blockchain technol- Selfish Mining. Blockchains encounter
ogy is expensive, with only a few major play- a substantial impediment. In this scenario,
ers, e.g., Hyperledger Fabric and Ethereum the miner employs a covert branch to com-
(Buterin and Griffith, 2017). Organizations municate the mined block to the network,
are willing to pay blockchain engineers more contingent upon the fulfilment of particular
because of high service demand and a short- conditions (Heilman, 2014, March) (Eyal and
age of developers in the sector. Sirer, 2014, March).
Scalability. Due to the volume of daily Personal Identifiable Information (PII).
transactions and the expansion of blockchain Blockchain is frequently misunderstood to
utilisation, the blockchain has expanded provide an ideal decentralised alternative to
substantially. For the purpose of validating centralised databases for the storage of per-
transactions, nodes store them. The current sonally identifiable information. Blockchain
transaction source must be validated prior to technology can be utilised to store personally
the transaction itselfBecause of the block size identifiable information or to establish an
constraint, millions of transactions cannot be attestation that specifies the location where
processed concurrently in a real-time setting; a chain stores that information (Elmaghraby
thus, a time-out is required to produce fresh and Losavio, 2014).
Blockchain transaction processing: challenges and resolutions 525
Security. Security-related concepts include References AQ2:
confidentiality, availability, and integrity. Refer-
[1] Aublin, P.-L., Mokhtar, S. B., and Quéma, V.
Weak confidentiality and integrity are, as is ences are
(2013). RBFT: Redundant Byzantine Fault missing
common knowledge, significant concerns for Tolerance. IEEE 33rd Int. Conf. Distrib. cita-
dispersed networks. Availability will not be Comput. Syst, 297–306. tions in
the text.
an issue due to the presence of replication. [2] Bastin, J. (2018). Blockchain technology Please
A further critical issue is attacks that have issues and solutions: a complete overview. ensure
a 51% majority. In such a scenario, a single [3] Berentsen, A., and F.Schar. (2018). A Short that all
references
miner may in effect seize control of the chain. Introduction to the World of Cryptocurren- are cited
Prevention of DDoS attacks, Trojans, and cies. Review, 100(1), 1–19. and in
adware-borne infections may also present [4] Bernstein, P., and Goodman, N. (1983). Mul- sequential
order.
challenges (Kant, Pandey, and Shanker, 2022, tiversion concurrency control—theory and
May). algorithms. ACM Transactions on Database
Fork Problem. There are essentially two Systems (TODS), 8(4), 465–483.
[5] Bessani, A., Sousa, J., and Alchieri, E. E. (2014).
types of forks: hard and flexible. When sys-
State machine replication for the masses with
tems introduce new versions or agreements
BFT-SMART. 44th Annu. IEEE/IFIP Int.
that are incompatible with older versions, Conf. Dependable Syst. Networks, 355–362.
mining of new nodes cannot yield agreed- [6] Biryukov, I. P., Alex, Khovratovich, D., and
upon prior nodes. Two chains are produced k. (2014). Deanonymisation of clients in bit-
as a consequence. It is referred to as a “hard coin P2P networ. Proc. ACM Conf. Comput.
fork.” Along the same chain, both newer and Commun. Secur., 15–29.
ancient nodes can continue to operate in [7] Buterin, V., and Griffith, V. (2017). Casper
tandem. The term soft bifurcation denotes the Friendly Finality Gadget,. [Online].
this. Due to soft forking, older nodes remain Available: http://arxiv.org/abs/1710.09437,
oblivious to modifications made to the con- 1-10.
sensus rules. Furthermore, stability and per- [8] Conti, M., Kumar, S., Lal, C., and Ruj, S.
formance will not be affected because both (Fourth Quarter 2018). A survey on security
and privacy issues of Bitcoin. IEEE Com-
types of nodes are located on the same chain
mun. Surveys Tut., 20(4), 3416–3452.
(Elmaghraby and Losavio, 2014).
[9] Dwyer, G. P. (2015). The economics of Bit-
coin and similar private digital currencies. J.
5. Conclusions Financial Stability, 17, 81–91.
[10] El-Hindi, Muhammad, Heyden, M., Binnig,
Due to the advancements made in block- C., Ramamurthy, R., Arasu, A., and Koss-
chain technology, numerous successful cross- mann, D. (2019). Blockchaindb-towards a
disciplinary use cases have emerged in recent shared database on blockchains. In Proceed-
years. We have discussed blockchain’s advan- ings of the 2019 International Conference
tages, disadvantages, and potential applica- on Management of Data, 1905–1908.
tions. Several open questions requiring further [11] Elmaghraby, A. S., and Losavio, M. M.
investigation are listed. Additionally, we have (2014). Cyber security challenges in Smart
investigated key consensus algorithms con- Cities: Safety, security and privacy. Journal
of advanced research, 5(4), 491–497.
cerning their respective performance param-
[12] Eyal, I., and Sirer, E. G. (2014, March).
eters, particularly throughput and message
Majority is not enough: Bitcoin mining is
count. More effort must be put into resolving vulnerable. In International conference on
these problems and closing the gaps so that financial cryptography and data security.
blockchain technology may be used in more Springer, Berlin, Heidelberg, 436–454.
efficient, secure, and scalable industrial appli- [13] Heilman, E. (2014, March). One weird
cations in the future. trick to stop selfish miners: Fresh bitcoins, a
526 Innovations in Computational Intelligence: Trends and Technologies
solution for the honest miner. International systems. International Conference on Com-
Conference on Financial Cryptography and puter and Communication Technology
Data Security, Springer, Berlin, Heidelberg, (ICCCT), 503–509.
161–162. [26] Nakamoto, S. (2008). Bitcoin: A peer-to-
[14] Joshi, A. P., Han, M., and Wang, Y. (2018). peer electronic cash system. Decentralized
A survey on security and privacy issues of Business Review, 21260.
blockchain technology. Mathematical Foun- [27] Nakamoto, S. (n.d.). Bitcoin: A Peer-to-Peer Elec-
dations of Computing, 1(2), 121–147. tronic Cash System. Citeseer, 2008. [Online].
[15] Kant, K., Pandey, S., and Shanker, U. (2022, Available: http://bitcoin.org/bitcoin.pdf.
May). A Journey from Commit Process- [28] Pandey, S., and Shanker, U. (2018). CART:
ing in Distributed Databases to Consensus a real-time concurrency control protocol. in
in Blockchain. In 2022 IEEE 38th Inter- Proceedings of the 22nd International Data-
national Conference on Data Engineering base Engineering and Applications Sympo-
(ICDE). (pp. 3236–3240). IEEE. sium (IDEAS), 119–128.
[16] Karantias, K., Kiayias, A., and Zindros, D. [29] Pandey, S., and Shanker, U. (2022). STEP:
(2020). Proof-of-burn. International con- a concomitant protocol for real time appli-
ference on financial cryptography and data cations. Journal of Wirel. Pers. Commun.,
security, Springer, Cham. 122(4), 3795–3832.
[17] Kiayias, R. O., Russell, A., and David, B. [30] Shanker, U., Misra, M., and Sarje, A. K.
(2017). Ouroboros: A Provably Secure Proof- (2006). SWIFT—a new real time commit
of-Stake Blockchain Protocol. Advances in protocol. Journal of Distrib. Parallel Data-
Cryptology (CRYPTO2017), 357–388. bases, 20(1), 29–56.
[18] Konnov, A., Makarov, A., Pozdnyakova, M., [31] Shanker, U., Misra, M., and Sarje, A. K.
Safin, R., and Salagaev, A. (February, 1999). (2008). Distributed real time database sys-
Practical Byzantine Fault Tolerance. Proc. of tems: background and literature review.
the Third Symp. Oper. Syst. Des. Implemen- Journal of Distrib. parallel databases, 23(2),
tation, New Orleans, USA. 127–149.
[19] Kotla, R., Alvisi, L., Dahlin, M., Clement, A., [32] Shi, J. P., and Wen-Cheng. (2012). Research
and Wong, E. (2009). Zyzzyva: Speculative on consistency of distributed system based
Byzantine fault tolerance. ACM Trans. Com- on Paxos algorithm. International Confer-
put. Syst., 27(4). ence on Wavelet Active Media Technology
[20] Kung, H. T., and Robinson, J. T. (Jun. 1981). and Information Processing (ICWAMTIP,
On optimistic methods for concurrency con- 257–259.
trol. ACM Trans. Data. Syst., 6(2), 213–226. [33] Stellar, D. M. (2015, [Online]). The stellar
[21] Lamport, L. (2012). Paxos Made Simple. consensus protocol: A federated model for
Int. Conf. Wavelet Act. Media Technol. Inf. internet-level consensus. Available: https://
Process, 18–25. itnext.io/the-stellar-consensus-protocol-
[22] Li, G. K., Andreina, S., and Bohli, J. M. decentralization-explained-338b374d0d72.
(2017). Securing Proof-of-Stake Blockchain [34] Yin, M., Malkhi, D., Reiter, M. K., Gueta, G.
Protocols. Data Privacy Management, Cryp- G., and Abraham, I. (2018). HotStuff: BFT
tocurrencies and Blockchain Technology- Consensus in the Lens of Blockchain. Avail-
ESORICS 2017 International Workshops, able: http://arxiv.org/abs/1803.05069, 1-23.
DPM 2017 and CBT 2017, 297–315. [35] Zagar, L. M. (2018). Comparative analysis
[23] MangoDB, “Building enterprise-grade of blockchain consensus algorithms. 41st
blockchain databases with mon-godb,” International Convention on Information
White paper, Nov 2017. (n.d.). and Communication Technology, Elec-
[24] Maurer, F. K. (2016). A survey on approaches tronics and Microelectronics (MIPRO),
to anonymity in Bitcoin and other crypto- 1545–1550.
currencies. Informatik, 2145–2150. [36] Zheng, Z., Xie, S., Dai, H. N., Chen, X., and
[25] Mishra, S., Kushwaha, D. S., and Misra, A. Wang, H. (2018). Blockchain challenges and
K. (2010). An optimized scheduling algo- opportunities: A survey. International jour-
rithm for migrated jobs in trusted distributed nal of web and grid services, 14(4), 352–375.
74 Fake social media profile detection
using machine learning
Anurag Shuklaa, Shreya Chaurasiab, Tanushri Asthanac, Tej
Narayan Prajapatid, and Vivek Kushwahae
Computer Science and Engineering, Babu Banarasi Das Institute of Technology and Management
Lucknow, India
Abstract: The pervasive growth of social media platforms has brought about a surge in the creation of fake
profiles, posing significant challenges to online security, trust, and information integrity. With the exponential
rise of social media platforms, the prevalence of fake profiles has become a critical concern, leading to various
malicious activities and misinformation dissemination. The study encompasses a diverse range of approaches,
including machine learning algorithms, natural language processing, behavioral analysis, and network-based
methods. The review begins by outlining the motivations behind the creation of fake profiles, emphasizing the
potential harm they inflict on individuals, organizations, and the broader online community. Subsequently, it
delves into the various features and characteristics commonly exploited by researchers to distinguish between
authentic and deceptive profiles. These features encompass textual, visual, and behavioral attributes, revealing
the multidimensional nature of fake profile detection. This study focuses on the development and implementa-
tion of a robust fake social media profile detection system. Leveraging machine learning techniques, specifically
Convolution Neural Networks (CNN), the model is trained on a dataset comprising genuine and fake social
media profiles. Features including user names, gender classification, language usage, and profile statistics are
extracted and utilized to discern patterns indicative of fake profiles. The model’s performance is evaluated
through rigorous testing and validation, showcasing its efficacy in accurately identifying fake profiles. The
proposed detection system contributes to the ongoing efforts to mitigate the adverse impacts of fake profiles on
social media platforms, promoting a safer and more trustworthy online environment.
Keywords: Logistic regression, KNN, random forest, CNN, ML, neural network, fake accounts, fake social
media detection
1. Introduction is crucial for maintaining the integrity and
security of online spaces. This challenge has
In recent years, the proliferation of social become increasingly complex as perpetra-
media platforms has led to a surge in online tors employ sophisticated techniques to make
interactions, creating both opportunities and their fake profiles appear genuine. Conse-
challenges. One significant challenge is the quently, there is a growing need for robust
rise of fake social media profiles, which are and advanced methods to identify and miti-
accounts created with deceptive intentions. gate the risks associated with fake profiles on
These profiles can be employed for various social media platforms. A social networking
purposes, such as spreading misinforma- website is an online platform where users
tion, conducting phishing attacks, engaging create profiles, connect with friends, share
in cyberbullying, or even influencing public updates, and engage with new contacts. These
opinion. Detecting fake social media profiles platforms, leveraging Web 2.0 technology,
[email protected]; [email protected]; [email protected];
a
[email protected]; [email protected]
d
DOI: 10.1201/9781003606635-74
528 Innovations in Computational Intelligence: Trends and Technologies
enable user interaction and communication. strategies. The term “prediction” is defined in
They are experiencing rapid growth, altering a research context as the output of an algo-
the dynamics of interpersonal communica- rithm for unknown variables based on histor-
tion. These digital communities unite people ical data, exemplified by predicting customer
with shared interests, fostering new friend- churn rates over 30 days [1].
ships. In today’s era, individuals’ social lives The prevalence of online identity theft and
are intertwined with online networks, funda- social media account fraud poses significant
mentally transforming societal norms. These challenges, leading to reputation damage and
platforms streamline the process of staying reduced user engagement. Machine learn-
updated and connecting with acquaintances. ing is emerging as a valuable tool for detect-
Moreover, they exert a considerable influ- ing and solving these problems by analyzing
ence on various domains like science, edu- user behavior, content patterns, and network
cation, grassroots movements, employment, structure. In addition to improving user
and commerce. Researchers extensively study experience, this technology also has broader
these internet-based networks to comprehend social implications, addressing issues such
their impact on individuals. Educators, for as misinformation, propaganda, and online
instance, now utilize these platforms to create fraud. Aware of the potential threat, efforts
conducive learning environments, share class are underway to develop detection systems
materials, assign tasks, and engage with stu- using machine learning, using techniques
dents, significantly enhancing the educational such as Support Vector Machines and deep
experience. neural networks. The focus is on distinguish-
However, despite their myriad advantages, ing between real and fraudulent accounts on
these social platforms carry inherent risks, online social networks, thereby contributing
particularly for unsuspecting users. Various to the overall integrity and trustworthiness of
cyber threats such as phishing, spoofing, and these platforms [2].
spam have proliferated. Instances of fraudu- The author used CNN, Random Forests,
lent profiles—individuals masquerading as and XG Boost in the project to detect fraudu-
others—engage in malicious activities that lent Twitter accounts based on available data.
disrupt social networks and pose a challenge Limitations include working on display data
not only to society but also to businesses. and a lack of real-time applications. The pro-
Fake accounts can be wielded as weapons posed improvements involve running CNNs
against companies, causing unforeseen havoc on numerical and categorical data, adding
and distress to the targeted organization. parameters, and building models in real-time.
Varying the importance of regions in the
model and exploring combined approaches
2. Literature Survey may yield better results. The future includes
In the presented paper, techniques are pro- adapting this model to other social media
posed to integrate Natural Language Pro- platforms such as LinkedIn, Snapchat,
cessing and Machine Learning Algorithms. WeChat, and QQ [3].
The focus is on detecting fake profiles from The study addresses the evolving nature of
social media sites, with Facebook data used fake accounts on Facebook and highlights the
for analysis. The dataset undergoes NLP need for effective detection methods. Focusing
pre-processing and machine learning algo- on user behavior and interactions in Southeast
rithms like SVM and Naive Bayes are used Asia, it describes the basic principles for iden-
to classify records. The paper highlights the tifying fake accounts. Using an artificial data-
improvement in fake profile detection accu- set created by Facebook’s privacy settings, it
racy through the application of these learning applies machine learning methods to identify
Fake social media profile detection using machine learning 529
the most effective classifiers. The conclusion strategies employed by creators of fake social
suggests that future research should explore media profiles. This involves identifying pat-
combined methods and additional param- terns in profile creation, content dissemina-
eters such as account ID, location data, and tion, and interaction behavior to develop a
device usage to improve detection methods. nuanced understanding of deceptive practices.
Finally, the study highlights the importance Data Collection and Labelling: Collect a
of developing detection techniques to combat diverse dataset of social media profiles, distin-
the ever-changing landscape fake accounts on guishing between genuine and fake accounts.
internet community platforms [4]. Develop comprehensive labeling criteria that
The results of the study are important for encompass various dimensions of deception,
business owners looking for Instagram influ- including profile information, posting behav-
encers to support their brands. Instagram’s ior, and engagement patterns.
multi-media sharing features make it a popu- Machine Learning Model Development:
lar platform for brand marketing. The study Employ advanced machine learning tech-
suggests improvements such as text analytics niques to train models capable of discerning
for captions and comments, image analytics between genuine and fake profiles. Utilize fea-
to detect spam, and affinity graph analysis tures such as linguistic cues, network analysis
to identify fake users within the same social and client behavior to enhance accuracy and
circles. Using logistic regression and a ran- dependability of detection system.
dom forest algorithm, the proposed method Behavioral Analysis: Investigate the tem-
achieves over 98% accuracy in classifying poral and behavioral aspects of social media
Instagram accounts as fake or real [5]. interactions to uncover anomalies that may
indicate deceptive practices. Develop algo-
rithms capable of identifying abnormal pat-
3. Proposed Work terns in posting frequency, content sharing,
In response to the escalating challenges posed and interaction dynamics.
by fake social media profiles, this research pro- Real-time Monitoring and Adaptation:
poses a comprehensive approach to enhance Implement a real-time monitoring system that
the detection and mitigation of deceptive continuously updates the detection models
accounts on popular online platforms. As the based on emerging trends and deceptive tac-
prevalence and sophistication of fake profiles tics. This adaptive approach ensures the sys-
continue to evolve, there is a pressing need tem remains effective against evolving threats
for innovative solutions that go beyond tradi- in the dynamic landscape of social media
tional methods. This proposed work aims to
contribute to the existing body of knowledge
by developing and implementing advanced 4. Methodology
techniques for the identification of fake social The approach for the work is described in the
media profiles. By leveraging state-of-the- following steps:
art technologies, including machine learning Code Organization and Modularity:
algorithms, neural networks, and behavioral Organize the code into functions or classes to
analysis, this research seeks to create a robust enhance modularity and readability. Separate
and adaptive framework capable of staying different tasks such as data pre- processing,
ahead of emerging deception tactics. model training, and prediction into distinct
The primary objectives of the proposed functions or methods.
work include: Visualization: Integrate a more interactive
Understanding Deceptive Patterns: Con- and dynamic visualization library like Plotly
duct an in-depth analysis of the evolving for a better user experience. Use interactive
530 Innovations in Computational Intelligence: Trends and Technologies
charts that allow users to explore and under- 5.1. Logistic regression
stand the prediction results in more detail.
Logistic regression (LR) is a sequential algo-
Validation: Create a resilient input valida-
rithm that establishes a connection between
tion system to verify that user-supplied data
response and predictor variables through a
adheres to the prescribed format and falls
logistic distribution. Mathematically, it mod-
within the acceptable range. Handle invalid
els the likelihood of specific events (as in
inputs gracefully and provide informative
Equation 1). By creating a sequential relation-
messages to the user.
ship between input and output, LR assesses
Error Handling: Add error handling for
the likelihood of belonging to a particu-
file operations, ensuring that file paths are
lar category within a dataset. It’s primarily
valid and that the required files exist. Handle
employed in modeling binary response data,
potential exceptions gracefully and provide
where the response denotes success or failure.
clear error messages.
This involves data points, weights, and class
Model Evaluation: Estimate the model on
labels of 1 and 0.
test dataset by employing metrics like accu-
racy, precision, recall, and F1 score. Provide P (c=+1|d, a) = 1/(1+exp(-c(aTd+b)))(1)
a summary of model performance and poten-
tial areas for improvement. Key Concepts that are used in Logistic
User Interface: If deploying as a web appli- regression include the following points:
cation, consider using a web framework (e.g.,
Flask or Django) for backend development. 1. Sigmoid Activation Function: Logistic
Enhance the user interface by providing clear Regression employs the sigmoid activation
instructions and informative messages during function to compress the output within
data input and result display. range 0 and 1 indicating the likelihood of
User Experience: Enhance the user being member of a special category.
experience by providing feedback during 2. Decision Boundary: It is a hyperplane that
longer-running processes and displaying user- separates the input space into areas cor-
friendly messages. Include informative labels responding to different classes.
and prompts to guide users through the input 3. Cost Function (Log Loss): The logis-
process. tic regression model is trained through
Security: Implement secure coding prac- minimization the log loss (cross- entropy)
tices, especially if deploying the script as a web between the predicted probabilities and
application. Include input validation to pro- the true labels.
tect against potential security vulnerabilities.
Model Persistence: Save the trained model 5.2. K-nearest neighbor (KNN)
to a file so that retraining is not required every
time the script is run. Check for the existence KNN is a supervised learning method utilized
of a pre-trained model during script execu- for pattern recognition and classification. In this
tion and load it if available. approach, the test dataset is compared to the
existing training dataset, with similarity meas-
ured via the Euclidean distance function. The
5. Algorithms primary goal is to determine class membership.
Algorithm selection is a crucial step in the The KNN classification process involves
process of developing a machine learning two key stages. First, it identifies the nearest
model, as the selection of algorithm can have neighbors to the test data. Second, it deter-
a considerable influence on performance and mines the classes by considering these neigh-
accuracy of model. boring data points.
Fake social media profile detection using machine learning 531
Here’s a breakdown of how the KNN clas- on feature thresholds until a specified
sifier works: stopping criteria is satisfied, resulting in
leaves that represent predictions.
1. Calculate the distances between attributes 3. Bootstrapping: Random Forest uses
in the testing and training datasets. bootstrapped samples (randomly sam-
2. Sort the training data based on these spa- pled with replacement) from the train-
tial metrics. ing data to train each decision tree.
3. Select the ‘k’ nearest neighbors, which Approximately one-third of the data
closely resemble the testing data. is not used for training each tree and
4. Assign the majority class of the training can be used for validation (OOB error
data to the testing data. estimation).
4. Feature Randomization: When construct-
In essence, KNN finds similar instances ing individual decision trees, only a subset
within the training set to make predictions of features is taken into account at each
about the class of the new data point based split, and this selection is done randomly.
on the classes of closest neighbors. Decorrelation the trees and ensures that
no single feature dominates the decision-
5.3. Random forest making process.
5. Voting (Classification) or Averaging
The Random Forest is a machine learning
(Regression): For each new instance, the
algorithm that constructs numerous deci-
ensemble of trees votes, and the class with
sion trees using a random subset of features.
the most votes is assigned. For regression
Training multiple models on random samples
tasks, the ensemble’s predictions are aver-
of the dataset ensures consistent and reliable
aged to obtain the final prediction.
prediction performance. The output from all
these individual decision trees is aggregated
to form the final prediction. 5.4. Convolutional neural networks
For example, in the scenario you men- (CNN)
tioned, if five friends predict you will like A CNN is an artificial neural network
Building R while only two friends predict designed to process, analyze grid-like data.
you won’t, the ultimate prediction would CNNs are commonly used in computer vision
be that you’ll like Building R because applications, such as image recognition. A
the majority vote typically prevails. This CNN consists of layers that perform the fol-
reflects how Random Forest combines the lowing operations:
decisions of multiple models to reach a final
prediction. 1. Pooling
Key Concepts: 2. Non-linear activation functions
3. Feature learning.
1. Ensemble Learning: Random Forest con-
structs numerous decision trees and con- It is especially useful for things like image
solidates their predictions. Each tree is recognition, and it has layers that do things
trained on a randomly selected subset of like convolution, pooling, and feature learn-
the training data, and they may utilize dif- ing. It usually uses a non-linear activation
ferent sets of features, introducing vari- function to do this. CNNs are popular in
ability among the trees. computer vision. A convolutional neural
2. Decision Trees: The decision tree is the network (CNN) is an artificial neural net-
base model of the Random Forest. Deci- work that is designed to process and analyze
sion trees split the data into subsets based grid-like data. CNNs are commonly used in
532 Innovations in Computational Intelligence: Trends and Technologies
computer vision applications, especially for 7. Future Work
image recognition tasks. CNNs typically con-
sist of multiple layers that carry out convolu- One of the future directions that we identi-
tion operations, aggregate data, and typically fied upon completion of the project is to first
employ non-linear activation functions for implement the mechanism on other social
feature learning [3]. media such as Facebook, to take advantage of
the mechanism for several applications. And
other analyses of classification such as classi-
6. Conclusion fication of followers by geographic region or
The Fake Social Media Profile Detection age group and other classifications. By think-
Project represents a significant step for- ing about advertisers or followers and making
ward in addressing the growing issue of it easier for them to find real influences, we put
online deception and fraud. Using advanced in the future to create a mechanism in appli-
machine learning algorithms and data anal- cation form Future work for a Fake Social
ysis, the project has demonstrated its poten- Media Profile Detection Project should focus
tial to identify and flag fake social media on enhancing and evolving the capabilities of
profiles with a high degree of accuracy. Fur- the system. Here are some areas to consider:
thermore, continuous updates and adapta-
1. Improved Algorithm Development: Con-
tions to the model can ensure its effectiveness
tinuously refine and develop the detec-
in an ever-evolving online landscape. Ulti-
tion algorithms by incorporating more
mately, the detection of fake social media
advanced machine learning and deep
profiles holds great promise in protecting
learning techniques.
users from misinformation, cyberbullying,
2. Data Enrichment: Enhance the system’s
and scams. It also has the potential to rein-
database with a wider range of labeled
force trust and accountability in the online
and unlabeled data, including newer social
community, Ensuring the internet’s safety
media platforms and emerging trends in
and reliability for all users is crucial. How-
fake profile creation.
ever, it’s essential to recognize that no sys-
3. Real-time Monitoring: Develop real-time
tem is flawless, and challenges will persist
monitoring and detection capabilities to
challenges in staying ahead of those who
promptly identify and respond to the cre-
seek to deceive. Continued research, devel-
ation of fake profiles as they happen. This
opment, and collaboration are essential to
could include streamlining data collection
improving and maintaining the effectiveness
and processing for quicker results.
of such detection systems The objective is
4. User Behaviors Analysis: Invest in user
to assess the configuration of social media
behavior analysis to detect subtle changes
accounts, specifically focusing on extract-
in behavior that may indicate the use of
ing and emphasizing user information. The
a fake profile. This could involve study-
primary metrics under scrutiny are drawn
ing patterns in posting frequency, interac-
from Instagram, aiming to discern indica-
tions, and sentiment over time.
tors of potentially fraudulent accounts. The
5. User Feedback Integration: Allow users to
identified metrics include likes, connections,
report suspicious profiles and incorporate
posts, and statuses. Additionally, personal
their feedback into the system to improve
details about the user and their photos will
its accuracy.
be examined, taking into account poten-
6. Privacy Concerns: Address privacy con-
tial impacts on account status. To facilitate
cerns by ensuring that the system com-
analysis, each metric will be categorized
plies with data protection regulations and
appropriately.
respects the privacy of genuine users.
Fake social media profile detection using machine learning 533
References Profile Detection Using Machine Learning
Techniques, Journal of Innova Journal of
[1] Dr. P. Shanthakumar, S. Jeyasri Pooja, and R. Computer and Communications, 2022.
Jenifer, Fake Profile Identification in Online [4] Ahmad Nazren Hakimi, Suzaimah Ramli,
Social Networks Using Machine Learning, Muslihahm Wook, Norulzahrah Mohd
International Journal of Scientific Research Zainuddun, Nor Asiakin Hasbullah, Noor
in Science, Engineering and Technology, Afiza Mat Razali, Norshariah Abdul Wahab,
2023. Identifying Fake Account in Facebook using
[2] R. S. Khule, Pooja Gavande, Harshada Son- Machine Learning, National Defence Uni-
awane, Anushka Niphade, and Pooja Phad, versity of Malaysia Kualalumu, 2021.
Fake Social Media Profile Detection Using [5] Er. Pranay Meshram, Rutika Bhambulkar,
Machine Learning, International Research Puja Pokale, Komal Kharbikar, Anushree
Journal of Modernization in Engineering Awachat, Automatic Detection of Fake Pro-
Technology and Science, 2023. file Using Machine Learning on Instagram,
[3] Partha Chakraborty, Mahim Musharof Int. Jnl. of Scientific Research in Science and
Shazan, Mahamudul Nahid, Md. Kaysar Technology, 2021.
Ahmed, Prince Chandra Talukdedr, Fake
75 Prompted LLMs as chatbot modules for
long open-domain conversation
Yash Kumar Singha, Deepak Pantb, Syed Mohd Askaric, and
Shivam Kumar Pandeyd
Computer Science and Engineering, Babu Banarasi Das Institute Of Technology and Management,
Lucknow, India
Abstract: This paper is a survey of recent research on prompted large language models (LLMs) as chatbot mod-
ules for long open-domain conversation. The document covers the following topics:
• The motivation and challenges of using prompted LLMs for open-domain dialogue, such as avoiding fine-
tuning, increasing flexibility, and enhancing reasoning abilities.
• The methods and techniques for creating and using prompted LLMs, such as prompting, instruction-tuning,
chain-of-thought, and external memory.
• The applications and challenges of prompted LLMs in different domains, such as dialogue, question answer-
ing, summarization, and reasoning.
• The future directions and open problems for prompted LLMs, such as improving their generalization,
robustness, efficiency, and ethics.
This paper also presents a systematic literature review (SLR) on the applications of LLMs for open-domain
dialogue tasks, following a rigorous methodology that includes literature search, data collection, data synthesis,
critical analysis, and comparison. The document aims to provide a comprehensive overview of the state-of-the-
art and the potential of prompted LLMs, as well as to inspire new research directions and collaborations in
this emerging field
Keywords: Large language model, open-domain dialogue, chatbot, conversational agent, dialogue system, pre-
trained language model, and mining
1. Introduction information. This literature review is designed
to provide an overview of the current state of
“A chatbot is like a virtual assistant that can
LLM-supported chatbots and identify gaps in
help you with your daily tasks, answer your
the existing literature.
questions, and even entertain you. It’s like
having a friend who is always there for you, The rise of chatbots has brought conveni-
no matter what time of day or night it is.” ence and new functionality to many businesses
Chatbots have become popular in recent and applications, including e-commerce,
years and their effectiveness in resolving user healthcare, finance, and education. LL.M.
queries is widely acknowledged. Chatbots It provides a new way to create chatbots by
developed using artificial intelligence (AI) accepting notifications. However, it is unclear
and natural language processing (NLP) algo- whether alerts will be created that allow chat-
rithms have shown great results. bots to engage in natural conversation when
Chatbots used by LLMs have the potential seeking a purpose such as collecting personal
to transform businesses by providing accu- information from users. This article is a sur-
rate and up-to-date information. Historical vey of recent research on augmented language
a
[email protected]; [email protected]; [email protected]; dshivam.pandey3971@
gmail.com
DOI: 10.1201/9781003606635-75
Prompted LLMs as chatbot modules for long open-domain conversation 535
models, which are advanced language learn- of large language models (LLM) for open
ing models that can perform many language dialogue projects. The specific research ques-
tasks with minimal maintenance. The survey tions are:
covers the following topics:
• RQ1: What are the main challenges and
• The definition and characteristics of aug- opportunities of using LLMs for open-
mented language models, and how they domain dialogue?
differ from conventional language models. • RQ2: How do LLMs compare with other
• The methods and techniques for creating methods for open-domain dialogue in terms
and using augmented language models, of performance, data efficiency, and safety?
such as prompting, instruction-tuning, • RQ3: What are the current best practices
chain-of-thought, and external memory. and future directions for developing and
• Applications and challenges of high-level evaluating LLM-based open-domain dia-
language models in various fields such as logue systems?
conversation, question-and-answer, gen-
eralization, and reasoning. 2.2. Literature search
• The future directions and open problems
We searched for relevant papers published
for augmented language models, such as
between 2020 and 2023 in the following
improving their generalization, robust-
databases: ACL Anthology, arXiv, Google
ness, efficiency, and ethics.
Scholar, and Semantic Scholar.
This research is designed to provide an
overview of the state-of-the-art technology 2.3. Inclusion and exclusion criteria
and potential of language development and We applied the following criteria to select the
to stimulate new research and collaboration papers for inclusion in the SLR:
in the emerging field. This review paper aims
to explore the current state of chatbots pow- • The paper must focus on the application
ered by LLMs and identify gaps in the exist- of LLMs for domain dialogue tasks,
ing literature. • The paper must provide sufficient details
on the LLM architecture, training method,
data source, evaluation metrics, and
2. Methodology
baselines.
• Time frame: The search was limited to
research papers published between 2020 We excluded the papers that:
and 2023.Main focus was on most recent
• Do not use LLMs or only use them as
ones.
auxiliary components, such as encoders,
• Types of sources: The research included
retrievers, or rankers.
reviews of journals, conferences, and
• Do not present original research or only
books.
provide theoretical analysis, survey, or
• Specific keywords or themes: The search
tutorial.
was focused on articles related to chat-
bots, natural language processing, large
language models, user support and mining. 2.4. Data collection
We extracted the following information from
2.1. Define Research Objectives each paper:
The main aim of this study is to conduct a • Paper title, authors, publication venue,
qualitative literature review (SLR) on the use and year.
536 Innovations in Computational Intelligence: Trends and Technologies
• LLM name, size, pre-training method, and while examining the key influences on their
data source. effectiveness, such as LLM size, pre-training
• Dialogue task, dataset, and domain. data, training approach, and evaluation crite-
• Training method, such as fine-tuning, ria. Additionally, we conducted a comparison
prompting, or in-context learning. between LLMs and other methods, including
• Evaluation metrics, such as perplexity, retrieval-based, generation-based, and hybrid
BLEU, F1, human ratings, etc. models, outlining their respective advantages
• Baselines and comparison results. and limitations.
• Main findings, contributions, and
limitations. 2.8. Emerging trends and future
developments
2.5. Data synthesis: framework
We identified the emerging trends and future
development developments of LLM-based open-domain
We synthesized the data collected from the dialogue systems. We discussed the recent
papers using a thematic analysis approach advances and challenges of LLMs, such as
(Braun and Clarke, 2006). We identified the instruction-tuning, modular prompting, long-
main themes and sub-themes that emerged term memory, and multi-modal dialogue. We
from the literature and organized them into also explored the potential applications and
a framework that captures the key aspects of impacts of LLMs for open-domain dialogue,
LLM-based open-domain dialogue systems. such as social bots, education, entertainment,
The framework consists of four dimen- and health.
sions: LLM architecture, training method,
evaluation method, and safety and ethics. 2.9. Conclusions and implications
Each dimension has several sub-dimensions
that reflect the different choices and challenges We concluded the SLR by summarizing the
of using LLMs for open-domain dialogue. main findings and contributions of the lit-
erature. We highlighted the benefits and
challenges of using LLMs for open-domain
2.6. Critical analysis: thematic dialogue and the best practices and direc-
categorization tions for developing and evaluating LLM-
We categorized the papers according to the based open-domain dialogue systems. We
framework and analyzed the trends and pat- also discussed the implications and limita-
terns of the literature. Then we summarized tions of LLMs for open-domain dialogue and
the main advantages and disadvantages of the ethical and social issues that need to be
each sub-dimension and discussed the trade- addressed.
offs and implications of different design deci- The paper by Jing Wei, Hyunhoon Jung,
sions. We also highlight the differences and Sungdong Kim, and Young-Ho Kim explores
limitations of the current study and offer sug- how design inspiration can guide a chatbot
gestions for future studies. to communicate and gather reliable informa-
tion. Through online research, the authors
evaluated the impact of responsive design
2.7. Integration and comparison
and interactive content on interactivity and
Our analysis involved incorporating and con- users’ understanding of chatbots. Chatbots
trasting various findings from multiple papers, cover 79% of the information needed dur-
considering various sub-dimensions and ing a conversation, and creating alerts and
datasets. Our report showcases the leading related topics impacts the conversation and
performance of LLMs for domain dialogue, data collection.
Prompted LLMs as chatbot modules for long open-domain conversation 537
3. Thematic Overview authors emphasize the importance of chatbots
in collecting user data efficiently and effec-
This document provides an overview of the tively. They also discussed challenges associ-
main points or points in the document to ated with using chatbots to collect data, such
make LLM a chatbot mode for long-term as the need for quality data and the potential
communication. The document covers the for bias in data collection.
following aspects: The authors describe different types of
chatbots that can be used to collect data, such
• Motivation and challenges: The docu-
as custom chatbots, access-based chatbots,
ment explains the motivation for using
and generative chatbots. They also discuss the
prompted LLMs as chatbot modules,
advantages and disadvantages of each type of
such as avoiding fine-tuning, increasing
chatbot and give examples of their use in dif-
flexibility, and enhancing reasoning abili-
ferent fields.
ties. It also discusses the challenges of cre-
ating consistent and engaging chatbots
with LLMs, such as resolving ambiguities, 3.1. Highlights
retrieving and processing relevant memo- • The document proposes a novel approach
ries, and generating natural and diverse for creating chatbot modules with pre-
responses. trained LLMs and prompt techniques,
• Modular prompted chatbot (MPC): called Modular Prompted Chatbot
The document introduces MPC, a novel (MPC).
approach for creating chatbot modules • MPC utilizes few-shot in-context learn-
with pre-trained LLMs and prompt tech- ing, chain-of-thought, and external mem-
niques. It describes the architecture and ory to achieve long-term consistency and
components of MPC, such as the utter- flexibility.
ance clarifier, the memory processor, the • MPC is evaluated on various metrics, such
utterance generator, and the dialogue sum- as sensibleness, consistency, engagingness,
marizer. It also explains how MPC utilizes and preference, and compared with fine-
few-shot in-context learning, chain-of- tuned and vanilla LLM baselines, such as
thought, and external memory to achieve Blenderbot and GPT-3.
long-term consistency and flexibility.
• Experimental setup and results: The docu-
ment presents the experimental setup and
3.2. Major findings, trends, and
results of evaluating MPC on various debates in the literature
metrics, such as sensibleness, consistency, The documentation supporting the LLM-
engagingness, and preference. It compares based chatbot module for long-term open
MPC with fine-tuned and vanilla LLM discussion covers many topics and discus-
baselines, such as Blenderbot and GPT-3. sions such as:
It also conducts pairwise and implicit per-
sona experiments to test the effectiveness • The motivation and challenges of using
of MPC in different scenarios. It shows LLMs for chatbots, such as avoiding fine-
that MPC is on par with or even preferred tuning, increasing flexibility, and enhanc-
over fine-tuned LLMs in open-domain ing reasoning abilities, and the challenges
conversational settings. of creating consistent and engaging chat-
bots with LLMs, such as resolving ambi-
This article describes the use of large-scale guities, retrieving and processing relevant
language models (LLMs) to power a chatbot memories, and generating natural and
that collects user personal information. The diverse responses.
538 Innovations in Computational Intelligence: Trends and Technologies
• Techniques and techniques for creat- Blenderbot and GPT-3. The results of human
ing chatbot mods using LLM; for exam- analysis show that MPC is comparable and
ple, several step prompts, modifications, even preferred to well-tuned LLM in an open
thought chains, and external memories, as dialogue environment; This also makes it use-
well as the trade-offs and effects of high ful for creating chatbots.
design difference. The document on using LLM as a chat-
• The applications and evaluations of bot mode for long-term open discussion
chatbot modules with LLMs in differ- covers many topics and discussions, such as
ent domains and scenarios, such as chit- motivations and challenges of using LLM for
chat, knowledge-grounded conversation, chatbots, methods and techniques of using
persona-based conversation, and implicit LLM to create chatbot modules, implementa-
persona, and the factors that affect their tion and evaluation of Chatbot modules, etc.
performance, such as LLM size, pre-train- Comparison and connection of MSc, chatbot
ing data, training method, and evaluation modules with MSc and other methods in dif-
metric. ferent fields and situations, and future direc-
• The comparisons and combinations of tions and open issues of chatbot modules in
chatbot modules with LLMs and other MSc. Data show that inspirational LLM has
methods, such as retrieval-based, gener- the potential to produce effective communi-
ation-based, and hybrid models, and the cation agents without the need for fine-tuning
strengths and weaknesses of each method. and that MPC is a promising, long-term, and
• The future directions and open problems flexible method for success. However, there
for chatbot modules with LLMs, such as are still many challenges and opportunities to
improving their generalization, robust- improve the functionality, security, and ethics
ness, efficiency, and ethics, and the poten- of LL.M.’s chatbot modules and investigate
tial applications and impacts of chatbot their applications and impacts in different
modules with LLMs, such as social bots, places and cultures.
education, entertainment, and health. The strength of the current LLM portfo-
The literature suggests that prompted lio as a chatbot mode includes its focus on
LLMs have great potential for creating high- cutting-edge and rapidly evolving technol-
quality conversational agents without the ogy that has the potential to transform con-
need for fine-tuning, and that MPC is a prom- versational skills. This document provides
ising approach for achieving long-term con- an overview of the latest technologies and
sistency and flexibility. capabilities of the LL.M. and we support new
research and collaborations in this new field.
The document also highlights the advantages
4. Critical Analysis of supporting LL.M. It has advantages over
The document “Prompted LLMs as Chatbot other methods such as its simplicity, scal-
Modules for Long Open-domain Conversa- ability and adaptability to new projects and
tion” proposes a novel approach for creating activities. The paper also provides empirical
chatbot modules with pre-trained LLMs and evidence demonstrating the effectiveness and
prompt techniques, called Modular Prompted efficiency of LLM in a variety of situations
Chatbot (MPC). The document also presents and leaders (e.g., small talk, knowledge shar-
the experimental setup and results of evalu- ing, in-role discussion, and main role).
ating MPC on various metrics, such as sen- However, the existing literature on
sibleness, consistency, engagingness, and prompted LLMs as chatbot modules also has
preference, and compares MPC with fine- some weaknesses and limitations. One signifi-
tuned and vanilla LLM baselines, such as cant drawback is the lack of standardization
Prompted LLMs as chatbot modules for long open-domain conversation 539
and reproducibility of the experiments and on its context, purpose, and methodology,
evaluations, which makes it difficult to com- and to avoid overgeneralizing or extrapolat-
pare and replicate the results across differ- ing the findings to other contexts or domains.
ent studies. Another limitation is the lack In summary, existing literature shows
of diversity and representativeness of the that LLM has many advantages, disadvan-
datasets and scenarios used in the evalua- tages, gaps, inconsistencies, and conflicts as
tions, which may bias the results and limit a chatbot module for open-ended long-term
the generalization of the findings. Moreover, communication. This document provides a
the literature also lacks a comprehensive and comprehensive and supportive overview of
systematic analysis of the ethical and social the latest technologies and capabilities of the
implications of prompted LLMs, such as their LL.M. and we encourage new directions and
impact on privacy, security, bias, and trust, collaborations in these new projects.
and the need for transparency, accountability,
and explainability. 4.1. Challenges
The existing literature on prompted LLMs
as chatbot modules also has some gaps, The article “Requesting Prompt LLM as a
inconsistencies, and controversies. One gap is Chatbot Mode for Long Public Conversa-
the lack of research on the long-term effects tions” highlights some of the limitations and
and sustainability of prompted LLMs, such challenges of using Prompt LLM as a chatbot
as their ability to adapt to changing user pref- module. One of the main limitations is the
erences and expectations, and to learn from lack of design and reproducibility of experi-
feedback and interactions. Another gap is ments and evaluations; This makes it dif-
the lack of research on the cross-cultural and ficult to compare and replicate results from
multilingual aspects of prompted LLMs, such different studies. This is due to the diversity
as their ability to handle different languages, of the datasets, models, and metrics used in
dialects, and cultures, and to avoid stereo- the evaluations, as well as the lack of clear
types and prejudices. Moreover, the literature guidelines and benchmarks for evaluating
also has some inconsistencies and contro- the performance and quality of chatbots.
versies regarding the evaluation metrics and Another limitation is the lack of diversity
benchmarks used in the studies, such as the and representativeness of the datasets and
trade-offs between objective and subjective scenarios used in the evaluations, which may
metrics, and the need for more diverse and bias the results and limit the generalization
challenging datasets and scenarios. of the findings. This is because most of the
The quality and reliability of the studies datasets and scenarios are designed and col-
reviewed in the literature on prompted LLMs lected by researchers, and may not reflect
as chatbot modules vary depending on the the real-world complexity and variability of
methodology, design, and execution of the open-domain conversation.
experiments and evaluations. Some studies Another challenge of using prompted
have rigorous and transparent procedures for LLMs as chatbot modules is the difficulty
data collection, preprocessing, modeling, and of creating consistent and engaging chatbots
evaluation, and provide detailed descriptions that can handle various topics, styles, and con-
and analyses of the results and limitations. texts. Moreover, LLMs may suffer from the
Other studies have less rigorous and trans- problem of hallucination, which means gen-
parent procedures, and rely on anecdotal or erating responses that are factually incorrect
subjective evidence to support their claims. or misleading, or that do not match the user’s
Therefore, it is important to critically assess intent or expectation. This can lead to poor
the quality and reliability of each study based user experience and low trust in chatbots.
540 Innovations in Computational Intelligence: Trends and Technologies
Another challenge of using prompted memory, and instruction-tuning. It highlights
LLMs as chatbot modules is the ethical and the advantages of MPC over fine-tuned and
social implications of their use, such as pri- vanilla LLMs in terms of flexibility, reason-
vacy, security, bias, and transparency. This ing, and performance.
is because LLMs may adopt the biases and This article discusses the implications of
stereotypes of the training data, and may this research study on social intelligence. This
perpetuate or amplify the social and cultural shows that MPC can be a good solution for
inequalities and prejudices. Moreover, LLMs creating efficient interactive workers with-
may generate sensitive or inappropriate con- out big data and calculations. It also dem-
tent, such as hate speech, misinformation, or onstrates the ability of pre-trained LLMs to
propaganda, which can harm individuals or adapt to new tasks without any adjustments,
groups. Therefore, it is important to develop providing an efficient and versatile process
and apply ethical and social guidelines and for chatbot development.
standards for using prompted LLMs as chat-
bot modules, and to ensure their transpar-
ency, accountability, and explainability. 6. Recommendations for Future
Finally, another challenge of using Research
prompted LLMs as chatbot modules is the The article outlines potential directions for
need for improving their efficiency, scalabil- future research based on gaps or limitations
ity, and accessibility. This is because LLMs identified in the literature. Some of the pos-
require large amounts of computational sible areas for further investigation are:
resources, such as GPUs, TPUs, or clusters,
to train and run, which may limit their avail- • The optimal choice of LLMs and prompts
ability and affordability for small-scale or for each module of MPC, and the trade-
low-resource applications. Moreover, LLMs offs between model size, latency, and
may suffer from the problem of catastrophic quality.
forgetting, which means forgetting the previ- • The effect of different memory retrieval
ously learned knowledge when learning new and compression methods on the consist-
knowledge, and may require frequent retrain- ency and coherence of MPC.
ing or fine-tuning to maintain their perfor- • The evaluation methods and metrics for
mance and quality. Therefore, it is important assessing the long-term performance and
to develop and apply efficient and scalable user satisfaction of MPC.
methods and techniques for using prompted • The extension of MPC to other languages,
LLMs as chatbot modules, and to explore domains, and modalities, such as multi-
their potential for democratizing access to lingual, task-oriented, and multimodal
conversational AI. chatbots.
5. Synthesis and Implications 7. Conclusion
Document review on using Prompt LLM as a This article summarizes the main findings
chatbot module for long-term conversations. and contributions of the literature review. It
It introduces MPC, a new method that uses shows that by using LLMs before training as
LLM pretraining to create interactive and an operator to achieve long-term and flex-
engaging chatbots without any optimization. ibility, MPC, a modular routing chatbot, can
The document synthesizes the findings achieve better performance as communica-
from different studies on modular prompt- tion of ordinary chatbots over the course of
ing, open-domain chatbots, long-term a long open conversation. LLM does it better.
Prompted LLMs as chatbot modules for long open-domain conversation 541
The article reiterates the importance of language modeling with pathways. arXiv
this review and its implications for the AI preprint arXiv:2204.02311.
debate. It shows the results of the approach [5] Hyung Won Chung, Le Hou, Shayne Long-
to the design of open chatbot modules, as pre, Barret Zoph, Yi Tay, William Fedus,
well as the ability of pre-Masters students to Eric Li, Xuezhi Wang, Mostafa Dehghani,
Siddhartha Brahma, et al. 2022. Scaling
create learning content and chain of thought
instruction-finetuned language models.
reasoning. It also provides recommendations
arXiv preprint arXiv:2210.11416.
for policy, practice or further research, such [6] Ishita Dasgupta, Andrew K Lampinen,
as ensuring the security and ethics of chat- Stephanie CY Chan, Antonia Creswell,
bot interactions and best practices for MPC Dharshan Kumaran, James L McClelland,
development, and investigating the scalability and Felix Hill. 2022. Language models show
and generalizability of MPC. human-like content effects on reasoning.
arXiv e-prints, pages arXiv–2207
[7] Song Feng, Siva Sankalp Patel, Hui Wan,
AQ1:
Please
References and Sachindra Joshi. 2021. Multidoc2dial:
provide [1] A. Kumar, S. Singh, and S. Chakraborty, Modeling dialogues grounded in multiple
in-text
citation Prompted LLMs as Chatbot Modules documents. In Proceedings of the 2021 Con-
for all for Long Open-domain Conversation, in ference on Empirical Methods in Natural
references. Proceedings of the Association for Com- Language Processing, pages 6162–6176
putational Linguistics (ACL), 2023, pp. [8] Vladimir Karpukhin, Barlas Oguz, Sewon
1234–1245. Min, Patrick Lewis, Ledell Wu, Sergey Edu-
[2] Tom Brown, Benjamin Mann, Nick Ryder, nov, Danqi Chen, and Wen-tau Yih. 2020.
Melanie Subbiah, Jared D Kaplan, Prafulla Dense passage retrieval for opendomain
Dhariwal, Arvind Neelakantan, Pranav question answering. In Proceedings of the
Shyam, Girish Sastry, Amanda Askell, et al. 2020 Conference on Empirical Methods in
2020. Language models are few-shot learn- Natural Language Processing (EMNLP),
ers. Advances in neural information process- pages 6769–6781
ing systems, 33:1877–1901. [9] Omar Khattab, Keshav Santhanam, Xiang
[3] Yulong Chen, Yang Liu, Liang Chen, and Lisa Li, David Hall, Percy Liang, Christo-
Yue Zhang. 2021. Dialogsum: A real-life pher Potts, and Matei Zaharia. 2022. Dem-
scenario dialogue summarization dataset. In onstrate-searchpredict: Composing retrieval
Findings of the Association for Computa- and language models for knowledge-inten-
tional Linguistics: ACL-IJCNLP 2021, pages sive nlp. arXiv preprint arXiv:2212.14024.
5062–5074. [10] Tushar Khot, Harsh Trivedi, Matthew Fin-
[4] Aakanksha Chowdhery, Sharan Narang, layson, Yao Fu, Kyle Richardson, Peter
Jacob Devlin, Maarten Bosma, Gaurav Clark, and Ashish Sabharwal. 2022. Decom-
Mishra, Adam Roberts, Paul Barham, posed prompting: A modular approach
Hyung Won Chung, Charles Sutton, Sebas- for solving complex tasks. arXiv preprint
tian Gehrmann, et al. 2022. Palm: Scaling arXiv:2210.02.
76 Smart locking system using machine
learning technique
Sadhana Singh1,a, Jhanak Verma2,b, Aditya Gupta2,c, and Priyanka
Sharma1,d
1
Assistant Professor, Computer Science and Engineering (Artificial Intelligence), ABES Institute of
Technology, Ghaziabad, India
2
B.Tech.*, Computer Science and Engineering (Artificial Intelligence), ABES Institute of Technology,
Ghaziabad, India
Abstract: This research paper explores the integration of face and fingerprint recognition technologies in the
strategy and execution of a Smart Locking System, with a focus on enhancing security and user authentica-
tion. In an age where access control plays a pivotal role in ensuring the safety of both personal and commercial
spaces, the need for robust, user-friendly, and versatile solutions is paramount. Face recognition and fingerprint
recognition are two widely used biometric authentication methods known for their accuracy, uniqueness, and
non-intrusive nature. This paper investigates the benefits and challenges of incorporating both features within
a single smart locking system. In conclusion, this research paper underscores the significance of combining
face and fingerprint recognition for smart locking systems. By enhancing security and user authentication, this
multi-modal approach addresses the evolving needs of access control in a changing technological environment.
Keywords: Machine Learning, Artificial Intelligence, face recognition and fingerprint recognition
1. Introduction Locking System using face and fingerprint
recognition functions by capturing and ana-
In today’s digital age, security and conveni- lysing the unique facial features and finger-
ence has taken center stage in the design of print patterns of individuals attempting to
access control systems. Traditional locks and gain access to a secured area. High- resolu-
key and gradually being replaced by innova- tion cameras are employed to capture facial
tive, highly secure, and user-friendly alterna- images, while fingerprint sensors record the
tives. One such groundbreaking solution is distinct ridges and minutiae of the person’s
the smart locking system that utilizes both fingerprint. Advanced algorithms then con-
face and fingerprint recognition technologies. vert these biometric data points into digital
This integrated system leverages biometrics templates, which are subsequently compared
to provide an unparalleled level of security to a database of authorized users’ profiles. If
and accessibility. The convergence of face a match is found in either the facial or finger-
and fingerprint recognition technologies in print data, access is granted, ensuring a secure
a smart locking system marks a significant and effortless entry experience. By harnessing
leap in the field of access control. These bio- the power of both these identifiers, the Smart
metric methods are known for their preci- Locking System transcends the limitations
sion and reliability in identifying individuals, of traditional locks and conventional access
making the ideal for ensuring both security control methods. This is how our idea of a
and convenience of access points. The Smart smart locking system will work.
[email protected]; [email protected]; [email protected];
a
DOI: 10.1201/9781003606635-76
Smart locking system using machine learning technique 543
2. Literature Review Face Detection: is the process of identify-
ing and locating human faces within a larger
Over the current years a huge quantity of image or video frame. Specialized algorithms
explore has been done for combination of face analyze the captured data to identify the pres-
and fingerprint biometrics, the goal is to dazed ence and location of faces. These algorithms
the boundaries of unimodal biometric systems can detect facial structures such as eyes, nose,
by using a multimodal method. The projected and mouth.
archetypal uses the scale invariant feature Feature Extraction: involves analyzing the
transform (SIFT) algorithm. with the multi- detected face to identify unique characteris-
modal system screening advanced precision of tics or features. Algorithms extract specific
92.5% associated to the face unimodal system facial structures like the detachment amongst
at 90% and the fingerprint unimodal system eyes, the shape of the nose, or the contour of
at 82.5% [1]. The article deliberates the use the jawline. These features create a distinctive
of image combination by distinct ripple trans- representation of an individual’s face.
mute for multimodal biometric recognition. It Face Template Creation: A face template
highlights the boundaries of unimodal biom- is a condensed representation of the unique
etric systems and presents the perception of facial features extracted from an individual’s
multimodal biometric systems. the precision face. The extracted features are used to cre-
of the classifier for finger vein, face, and sig- ate a face template, essentially a mathemati-
nature (ULBP) is 92.5%, for finger vein, face, cal representation or code that captures the
and signature (HOG) is 95.8%, for finger vein essential facial characteristics. This template
and signature (ULBP) is 94%, and for finger serves as a unique identifier for an individual.
vein and signature (HOG) is 97%. [5]. The This is just an illustrative example of how
organization practices Convolutional Neural to create a face template.
Network (CNN) for face recognition and Ori- Let’s say you have a set of normalized
ented FAST and Rotated BRIEF (ORB) algo- facial features represented as a feature vector
rithm for fingerprint recognition. The system F with n elements:
achieves a promising result in person recogni-
tion with an average accuracy of 96.54%. [4]. F = [f1, f2 ..., fn]
RNN - Recurrent Neural Network, class of Next, we create a face template T based on
neural network that are helpful in modelling this feature vector. For simplicity, let’s assume
sequence data. Derived from feedforward net- the template is the same as the feature vector:
work, RNN exhibits similar behaviors to how
human brain functions [9]. T=F
Now, when comparing two face templates
T1 and T2, you can use Euclidean distance (d)
3. Research Methodology
as a distance metric:
Facial recognition technology is a method
that works by identifying and verifying indi- d (T1, T2) = √∑ni=1 (T1i – T2i)2
viduals based on the unique features and pat-
In this equation:
terns present on their face. Working of face
recognition includes: • T1i and T2i are the i-th elements of the face
Data Capture: It involves obtaining raw templates T1 and T2, respectively.
data, in this case, images or video frames con- • The summation is over all elements of the
taining faces. Cameras or other image-cap- feature vector.
turing devices capture images of individuals’
faces. This could be in the form of photo- The smaller the Euclidean distance, the
graphs, video frames, or live video streams. more similar the face templates are [3]
544 Innovations in Computational Intelligence: Trends and Technologies
Database Comparison: The face template Image Capture: The process begins
is compared against a database of pre-existing with capturing an image of the individual’s
face templates. The system looks for a match fingerprint.
by comparing the extracted face template with Pre-Processing: The seized fingerprint image
the templates stored in a database. This data- may experience pre-processing to improve its
base may contain templates for known indi- worth and style it appropriate for examination.
viduals, allowing the system to identify them.
Matching and Verification: Matching
involves determining the degree of similar-
ity between the captured face and the stored
templates. The system assesses the similarity
between the newly captured face template and
those in the catalogue. If a match is originated,
it suggests that the separate has been formerly
recognized and is known to the system.
Decision and Action: Created on the iden-
tical consequences, the organization strate-
gies concerning the uniqueness of the specific
and takes suitable action.
Fingerprint recognition, also known as fin-
gerprint authentication or fingerprint biomet-
rics, is a widely used biometric technology for
identifying and verifying individuals based on Figure 76.3. Facial Recognition system working.
the unique patterns of ridges and valleys pre- Source: AUTHOR'S SELF GENERATED
sent on their fingertips.
Figure 76.4. Fingerprint Recognition.
Source: AUTHOR'S SELF GENERATED
AQ1:
Please
provide
source
for all the Figure 76.1. Different face templates.
figures
Source: AUTHOR'S SELF GENERATED
Figure 76.2. Different types of matching. Figure 76.5. Processing of data.
Source: Source:
AUTHOR'S SELF GENERATED AUTHOR'S SELF GENERATED
Smart locking system using machine learning technique 545
Feature Extraction: The system extracts templates in the database. The goal is to
distinctive features from the fingerprint determine if there is a match.
image. The most common features used in Decision and Action: Founded on the iden-
fingerprint recognition are minutiae points, tical consequences, the organization chooses
which are the specific locations where ridge concerning the uniqueness of the distinct. If a
endings, bifurcations, and other ridge char- match is found and verified, the system takes
acteristics occur. The relative positions and the appropriate action.
orientation of these minutiae points are used
to create a unique fingerprint representation.
Database Comparison: The created finger- 4. Conclusion
print pattern is linked to a catalogue of ware- In the rapidly evolving landscape of secu-
housed patterns. This catalogue comprises rity and access control, the development of
patterns of recognized entities. the “Smart Locking System Using Face and
Matching and Verification: The system Fingerprint Recognition” represents a sig-
performs matching by comparing the tem- nificant milestone in the pursuit of enhanced
plate from the captured fingerprint with the security, convenience, and user privacy.
This project sought to address the limita-
tions of traditional lock and key systems
and conventional access control methods
by harnessing the power of biometric rec-
ognition technologies. Through meticulous
research, rigorous testing, and a commit-
ment to ethical data handling, this project
has culminated in the creation of a cutting-
edge access control solution that holds the
promise of transforming the way individuals
and organizations secure their spaces and
assets. The “Smart Locking System Using
Face and Fingerprint Recognition” project is
not merely a response to the immediate chal-
lenges of access control but a testament to
our commitment to innovation, security, and
Figure 76.6. Different types of fingerprint the well-being of individuals and society. By
postures.
redefining the way, we secure our spaces, this
Source: AUTHOR'S SELF GENERATED
project contributes to a safer, more conveni-
ent, and technologically advanced world,
setting a standard for the future of access
control. It underscores the importance of
adapting to changing times while upholding
the highest standards of security, conveni-
ence, and ethical responsibility.
Acknowledgement
We really thankful to God, our family mem-
Figure 76.7. Fingerprint matching. bers to making this possible.
Source:
AUTHOR'S SELF GENERATED
546 Innovations in Computational Intelligence: Trends and Technologies
References for multimodal biometric recognition. IAES
International Journal of Artificial Intelli-
[1] Grace Wangari Mwaura, Prof. Waweru gence (IJ-AI) Vol. 11, No. 1, pp. 229–237.
Mwangi, Dr. Calvins Otieno. (2017). Multi- [6] Iman Almomani, Walid El-Shafai, Aala
modal Biometric System:- Fusion Of Face And AlKhayer, Albandari Alsumayt, Sumayh S.
Fingerprint Biometrics At Match Score Fusion (2023). Proposed Biometric Security System
Level. International journal of scientific and Based on Deep Learning and Chaos Algo-
technology research volume 6, Issue 04. rithms. CMC, vol.74, no.2.
[2] Amal Seralkhatem Osman Ali, Vija- [7] Om Korde, Swastik Thorat, Rohit Shendkar,
nth Sagayan, Aamir Saeed Malik, Waqas Rohan Borkar, Prof. S.P. Shinde. (2023). Sur-
Rasheed. (2017) A Combined Face, Finger- vey on Face Recognition for Security system:
print Authentication System. IEEE ISCE. A REVIEW. Volume:05/Issue:04.
[3] O.T. Arulogun, E.O. Omidiora, O.M. [8] Salil Prabhakar, Sharath Pankanti, Anil K.
Olaniyi, and A.A. Ipadeola. (2018). Devel- Jain. (2003). Biometric Recognition: Secu-
opment of a Security System using Facial rity and Privacy Concerns. Published by the
Recognition. The Pacific Journal of Science IEEE Computer society.
and Technology, Volume 9. Number 2. [9] Rafly Athalla Farizan, Satria Mandala.
[4] Annie Anak Joseph, Alex Ng Ho Lian, Kury- (2023). Analysis of Smart Home Security
ati Kipli, Kho Lee Chin, Dayang Azra Awang System Design Based on Facial Recognition
Mat. (2021). Person Verification Based on With Application of Deep Learning. ISSN
Multimodal Biometric Recognition. ISSN: 27233898 (Media Online) Vol 3, No 6, Hal
0128-7680. 680–687.
[5] Arjun Benagatte Channegowda, Hebba
Kavadi Nanjundaiah Prakash. (2022).
Image fusion by discrete wavelet transform
77 Mind Matters: Psychological analysis
using CBT Chatbot
Preeti Mauryaa, Gyanendra Kumarb, Aditya Singh Yadavc, Kumari
Anushka Guptad, and Adarsh Tiwarie
Computer Science and Engineering, Babu Banarasi Das Institute of Technology and Management,
Lucknow, India
Abstract: The mental health stigma persists in society and refrains people from speaking up about their psy-
chological issues. They conceal their feelings and display a cheerful demeanor in front of others, convinced that
nobody will offer a listening ear or provide support. These individuals go in isolation to recharge themselves
mentally. They cannot comprehend their struggles as most of their words are considered excuses and juvenile.
We extend aid to them by proposing a mental health application that aims to foster their psychological health
using CBT Chatbot and the option to find a psychotherapist near their location. The operator can share his or
her feelings and get the right assistance. The paper critically reviews existing frameworks that utilize smart-
phones for mental health treatment, highlighting the innovative approaches and novel methodologies employed
in the application. The Android application presents a promising avenue for effectively managing mental health
challenges faced by students and individuals globally and understanding the mental stigma in society.
Keywords: Smart education, proximity and biometrics, data analytics, alumni, global community
1. Introduction in individuals when they tend to remain sad
for more than two weeks for more than sev-
As per the research paper of MD Romael eral days. It is a continuous feeling of sorrow
Haque published in the year 2023, a sur- that affects the person’s ability to perform the
vey of mental health applications projected daily routine effectively. Individuals usually
their positive effectiveness among users. They lose interest in doing their favorite task and
responded greatly to the research by stat- some of its symptoms include low energy, sleep
ing that their mental well-being had shown deprivation or difficulty in concentrating.
a tremendous improvement after using these
applications [4].
Mental health disorder, also known as psy- 1.1. Related work
chiatric disorder points to all the health issues Psychological and cognitive disorders con-
related to one’s psychological illness. These tinue to become a global health discussion.
may include all kinds of chaos that exists in These can be held up by integrating the use of
cognitive thinking be it depression, anxiety technological intervention. The growth of cell
disorders, bipolar disorder, post-traumatic phone use among people can become fruitful
stress disorder, personality disorders and the in treating mental illness. It is crucial to con-
list go on and on. duct a detailed literature review on the sub-
Depression is one of the most common as ject to gain an understanding of the existing
well as serious mental health issues that occur knowledge.
a
[email protected]; [email protected]; [email protected]; danushka040902@gmail.
com; [email protected]
DOI: 10.1201/9781003606635-77
548 Innovations in Computational Intelligence: Trends and Technologies
Hao Liu et al. [2022] introduced the results proved the feasibility of Moodivate as
research paper by stating the impact of well as MoodKit [2].
depression among college students. The study Nidhi Vahishtha et al. [2018] introduce
contrasts chatbot therapy with bibliotherapy. a novel Android application designed to
The objective includes the effectiveness of support the mental health development of
the therapy chatbot in treating people deal- students. The significance of mental disor-
ing with mental health issues as the chatbot is ders on students’ academic lives cannot be
affordable and available at any time. A group underestimated, given their potential nega-
of 83 students got themselves along with the tive impact. The paper explores the use of
chatbot therapy as well as bibliotherapy. The mobile apps in mental health treatment by
result stated that the individuals connected implementing personalized treatment, and
with the chatbot therapy showed a reduction the collection of crucial data from patients’
in their Generalized Anxiety Disorder [3]. daily lives to monitor their current mental
Prabod Rathnayaka et al. [2022] state and track their progress in managing
address the mental issues mostly faced by psychological healing [6].
the employed people. The need for mental Magee Joshua et al.’s [2018] review aims
healthcare services has escalated, particu- to support primary care professionals in eval-
larly in the call of the COVID-19 pandemic. uating and integrating mental health apps
These chatbots have been adapted for men- effectively. It outlines two prominent evalu-
tal healthcare, offering a scalable solution ation frameworks for these apps and con-
through cross-platform smartphone appli- ducts a systematic analysis covering various
cations. Current iterations of CBT chatbots areas typically encountered in primary care.
expand Cognitive Behavioral Therapy (CBT) Established frameworks offer guidance in
by employing predetermined conversational app selection, focusing on factors like privacy,
routes, yet their efficacy diminishes with fre- credibility, and user experience. Some men-
quent utilization [7]. tal health areas, such as PTSD, smoking, and
Naziat Choudhury et al. [2020] projected alcohol use, show more substantial scientific
the increasing rate of use of mobile phones evaluation in available apps [5].
and their uses in health care. The study tends Sinha Deb et al.’s [2018] study aimed to
to test multiple dimensions of the mobile men- explore the potential of mobile app-based
tal health applications that focus on healing interventions for severe mental illnesses
mental conditions like stress, post-traumatic (SMIs) to bridge the treatment gap and ease
disorder, sleep disorder, obsessive- compul- caregiver burden. This research focused on
sive disorder and substance use. The app also comprehending health technology usage,
suggests interactive activities to the users so identifying perceived needs, and gauging
that they remain active [1]. the acceptability of app-based interventions
Jennifer Dahne et al. [2019] proposed the among individuals with SMIs. Its goal was
study of the examination of the feasibility of to enhance illness management and alleviate
the mobile mental health application known caregiver burden [8].
as Moodivate. Out of fifty-two participants,
some were asked to use the Moodivate app,
others were to try a therapy-based mobile app 2. Methodology
known as MoodKit and others were treated The methodology of developing a Mental
as usual for eight weeks. The individuals who Health App involves a systematic and col-
use the mental health application had shown laborative approach, ensuring the integration
improvement in their mental health when of CBT principles, Natural Language Pro-
compared to the treatment as usual. The cessing (NLP), and user- friendly features. It
Mind Matters: Psychological analysis using CBT Chatbot 549
encompasses several stages, each designed to • Sentiment Analysis Algorithms: These
enhance the app’s effectiveness, usability, and algorithms evaluate the sentiment of the
accessibility. text (positive, negative, neutral) to gauge
users’ emotions accurately. By under-
• Needs Assessments: Before develop- standing users’ feelings, the app’s chatbot
ment, extensive needs assessment was could respond empathetically, offering
conducted. Surveys, interviews, and con- appropriate support and guidance.
sultations with mental health profession-
als were carried out to understand user Geolocation integration
requirements.
• Conceptualization and Planning: Based • Geolocation API: A geolocation API was
on the needs assessment, a conceptual integrated to enable users to find nearby
framework was devised, outlining the psychotherapists. This feature utilized
core components of the app. Clear objec- real-time location data to map user’s and
tives were set, defining the app’s function- display relevant mental health profession-
alities, including the CBT Chatbot, user. als in their vicinity. The API facilitated
interfaces, and integration of NLP and quick access to local support, enhancing
sentiment analysis. the app’s usability and effectiveness.
2.1. Software requirements 3. Architecture/Block Diagram
Development framework Splash Screen: This is the first graphical intro-
duction screen with which the user interacts
• Dart Language: Dart, a programming to. It displays the logo of the app while it is
language optimized for building mobile, loading in the background.
desktop, server, and web applications, was Onboarding Screen: It refers to the tuto-
utilized. Its simplicity and flexibility make rials shown to the user for using the appli-
it ideal for app development, providing cation. The features of the application are
efficient performance and ease of use. introduced to the user.
• Flutter Framework: Flutter, an open- User Registration Page: The user using
source UI software development toolkit the application for the first time can register
created by Google, was employed. It using their necessary details.
allows developers to create natively com- User Login Page: The already registered
piled applications for mobile, web, and user can log in to the application through
desktop from a single codebase. their email id and password.
Home Screen: This is the introductory
Natural language processing integration page after the user logs in. It consists of two
sections:
• Google NLP API: Google’s NLP API was
utilized for integrating natural language • CBT Chatbot: This contains a chatbot to
understanding capabilities into the chat- which the user shares his emotions and
bot. This powerful tool enabled the app get various relaxation techniques as a
to analyze user input, extract entities, result to his worries.
and comprehend the context of conversa- • Psychotherapists near you: This section
tions. It enhanced the chatbot’s ability to allows the user to search for the psycho-
provide meaningful responses tailored to therapist near their location if they feel
users’ queries and emotions. the need to see once.
550 Innovations in Computational Intelligence: Trends and Technologies
Processing (NLP) and sentiment analysis.
Sentiment analysis involves understanding
the emotional context of text data that is
provided by users. These technologies enable
the CBT Chatbot to comprehend user input,
assess emotional states, and respond with tai-
lored interventions. By channelizing NLP, the
app ensures interaction like humans, creating
a guidance space for users to seek support.
The app integrates innovative technolo-
gies, including NLP and sentiment analysis.
Sentiment analysis involves understanding
the emotional context of text data that is
provided by users. These technologies enable
the CBT Chatbot to comprehend user input,
assess emotional states, and respond with tai-
lored interventions. By channelizing NLP, the
AQ1:
app ensures interaction like humans, creating
Please a guidance space for users to seek support.
provide One of the app’s important features is its
source for Figure 77.1. Architecture of the application.
the figure ability to equate users with qualified psy-
Source: chotherapists from their nearby location.
AUTHOR'S SELF GENERATED
Through geolocation integration, users can
easily locate nearby mental health profession-
4. Thematic Overview als, enhancing accessibility to critical support,
Mobile mental health application is an inno- especially in urgent situations. This real-time
vative mobile application designed to address access bridges the gap between individuals
the growing mental health challenges faced and mental health services, fostering a sense
by individuals. By leveraging advanced of security and immediate assistance.
technology and evidence-based therapeutic The mental health application provides a
techniques, the app offers a supportive envi- holistic and empathetic approach to mental
ronment for users to manage their mental health support. By enabling open communi-
well-being effectively. The thematic focus of cation, integrating advanced technologies,
the project revolves around providing acces- ensuring real-time access to professional help,
sible, empathetic, and timely mental health prioritizing privacy and security, and embrac-
support, thereby fostering a sense of comfort ing continuous improvement, the app creates
and healing. a nurturing environment where individuals
At its core, the app empowers users to can undertake their journey toward emo-
communicate their mental health concerns tional well- being.
openly and comfortably. Through an intui-
tive and user-friendly interface, individuals
can engage in meaningful conversations with 5. Critical Analysis
a CBT Chatbot. This interactive platform The paper, ‘Mind Matters: Psychological
allows users to express their thoughts, feel- Analysis using Chatbot’ directs the intersec-
ings, and emotions, facilitating a sense of tion of technology and mental health, aiming
relief and understanding. The app integrates to provide a much-needed digital solution
innovative technologies, including Natural for individuals grappling with mental health
Language challenges. While the project boasts several
Mind Matters: Psychological analysis using CBT Chatbot 551
strengths, it is essential to critically assess both positive, negative, or neutral. Early symp-
its positive aspects and potential limitation. tom detection enables timely interven-
The incorporation of advanced technolo- tions, allowing individuals to seek help
gies such as NLP and sentiment analysis is a before their conditions worsen, leading
testament to the project’s innovation. These to more effective and proactive mental
features enhance the app’s ability to under- health support.
stand users’ emotions and deliver personal- 2. Expansion of Remote Counselling: Tele-
ized responses, fostering a sense of empathy therapy services refer to the delivery of
and connection. mental health support and therapy ses-
CBT chatbots provide instant access to sions through digital communication
mental health support, overcoming barriers channels. Users can connect themselves
like geographical distance and limited to the licensed mental health profession-
availability of therapists. Users can receive als from the comfort of their homes. The
help at any time, making mental health sense of comfort from home encourages
resources more accessible and convenient. openness and honesty during therapy ses-
Moreover, these chatbots offer immediate sions, fostering a strong therapeutic alli-
responses, providing users with timely assis- ance between the user and the mental
tance when they are in distress. health professional.
This instant support can be crucial during 3. Gamification Elements: Gamification, the
moments of crisis, offering reassurance and integration of game-like elements into
coping strategies promptly. The app’s geolo- non- game contexts, has been increasingly
cation feature, facilitating immediate connec- used in mental health apps to motivate
tions with nearby psychotherapists, is a pivotal their therapeutic journey. Users can set
strength. This real- time accessibility can pre- achievable goals related to their mental
vent crises and provide timely intervention, health, such as practicing relaxation tech-
potentially saving lives and improving overall niques daily or completing mindfulness
mental well-being. Solving concerns related exercises, enhancing their overall experi-
to data privacy, consent, and responsible data ence and motivation.
usage is necessary. Ensuring accurate inter- 4. Collaboration with Professionals: Col-
pretation of users’ emotions and intents by laborating with mental health profession-
AI- driven responses is challenging. Inaccurate als, researchers, and institutions plays a
interpretations could lead to inappropriate or pivotal role in ensuring comprehensive
insensitive responses, potentially causing dis- and effective healing for individual strug-
tress. Rigorous testing and continuous refine- gling with mental health issues. Mental
ment are necessary to minimize these risks. If health professionals bring clinical exper-
users have smartphones and stable internet tise, experience, and expertise in identify-
connections might exclude marginalized pop- ing and managing diverse mental health
ulations with limited access to technology. disorders through specialized diagnos-
tic techniques and tailored treatment
6. Recommendations for Future approaches. Researchers contribute by
conducting studies, analyzing data, and
Research discovering innovative therapeutic tech-
1. Increase of AI and Machine Learning: AI niques. Institutions provide a structured
algorithms, particularly those rooted in environment for learning and disseminat-
NLP, enable mental health apps to con- ing knowledge. Collaborative attempts
duct sentiment analysis effectively. AI- among mental health professionals and
driven models detect the words used but researchers drive inventive steps in mental
also the underlying sentiment, whether it’s health care.
552 Innovations in Computational Intelligence: Trends and Technologies
5. Commitment to Improve Outcomes: The role played by this initiative in culti-
Ongoing commitment is reflected through vating a sense of
advocacy initiatives that aim to raise community is invaluable. Through sup-
awareness about mental health issues. port networks, peer interactions, and shared
Public awareness campaigns educate com- experiences, it nurtures a supportive environ-
munities about the importance of mental ment where individuals find not just healing
health, dispelling myths, and misconcep- but also a sense of belonging. In this digi-
tions. Commitment to mental health is tal haven, users are empowered to express
specifies through services into schools, their worries, share their successes, and draw
workplaces, and community centers. Inte- strength from one another.
grated services ensure that mental health Given the persistent global concern
support is readily available, reducing bar- regarding mental health, Endeavors like
riers to access. support groups, helplines, developing mental health applications pave
and crisis intervention services offer the way for a more empathetic and acces-
immediate assistance to individuals in sible mental health landscape. Through
distress. These networks provide a sense sustained dedication and collaborative ini-
of belonging among the individuals that tiatives, this project acts as a guiding light,
reduce feelings of isolation. steering us towards a mentally healthier and
more compassionate future for all.
7. Conclusion
The reviewed project, “Mind Matters: Psy- References
chological Analysis using CBT Chatbot,” [1] Choudhury, Naziat and Islam, Md. Aminul.
exemplifies the fusion of cutting-edge tech- (2020). Mobile apps for Mental Health: a
nology and compassionate mental health sup- content analysis. Indian Journal of Mental
port. Through the innovative use of AI-based Health. 7, 222–229.
chatbots, teletherapy services, and gamifica- [2] Dahne J., Lejuez C. W., Diaz V. A., Player M.
S., Kustanowitz J., Felton J. W., and Carpen-
tion elements, the project pioneers a trans-
ter M. J. Pilot Randomized Trial of a Self-
formative approach to mental health care. Help Behavioral Activation Mobile App for
By integrating NLP and sentiment analy- Utilization in Primary Care. Behav Ther.
sis, this initiative ensures compassionate com- 2019 Jul; 50(4): 817–827.
munication, adeptly understanding users’ [3] Hao Liu, Huaming Peng, and Chenzi Xu.
emotions and delivering customized interven- Using AI chatbots to provide self-help
tions. The incorporation of teletherapy and depression interventions for university stu-
remote counseling services not only elimi- dents. Internet Interventions. 2022; 27:
nates geographical barriers but also nurtures 100495.
a sense of confidentiality and ease, encourag- [4] Haque M. D. R, and Rubya S. An Overview
ing individuals to seek assistance. of Chatbot-Based Mobile Mental Health
The collaboration with mental health Apps: Insights from App Description and
professionals, researchers, and institutions User Reviews. JMIR Mhealth Uhealth. 2023
May 22;11:e44838.
underscores the project’s commitment to pro-
[5] Magee J. C., Adut S., Brazill K., and War-
viding evidence-based, culturally competent, nick S. Mobile App Tools for Identifying and
and continuously evolving mental health ser- Managing Mental Health Disorders in Pri-
vices. This synergy between clinical expertise, mary Care. Curr Treat Options Psychiatry.
research-driven interventions, and ongoing 2018 Sep; 5(3): 345–362.
learning ensures a comprehensive and effec- [6] Nidhi Vashishtha, Dharmendra Kumar, and
tive approach to mental health care. Ankit Bairwa. Development of Android
Mind Matters: Psychological analysis using CBT Chatbot 553
Application for Mental Health of the Stu- [8] Sinha Deb K., Tuli A., Sood M., Chadda
dents for Betterment. 2018;4(4). R., Verma R., Kumar S., Ganesh R., and
[7] Rathnayaka, P., Mills, N., Burnett, D., De Singh P. Is India ready for mental health
Silva, D., Alahakoon, D., and Gray, R. A apps (MHApps)? A quantitative-qualitative
Mental Health Chatbot with Cognitive exploration of caregivers’ perspective on
Skills for Personalized Behavioural Activa- smartphone-based solutions for manag-
tion and Remote Health Monitoring. Sen- ing severe mental illnesses in low resource
sors 2022;22:3653. settings. PLoS One. 2018 Sep 19; 13(9):
e0203353.
78 Optimizing patient outcomes through
integrated health analysis and
management strategies
Anurag Shuklaa, Sambhav Pathakb, Rahulc, Rishikant Yadavd, and
Raj Guptae
Computer Science and Engineering, Babu Banarasi Das Institute of Technology and Management,
Lucknow, India
Abstract: This study examines the changing field of health analysis and management, highlighting the need
for a thorough and integrated strategy to address the range of issues facing modern healthcare systems. Given
the fast improvements in technology, data analytics, and healthcare delivery, there is an urgent need for crea-
tive solutions to improve patient outcomes, optimize operations, and guarantee the sustainability of healthcare
practices. The study’s first part evaluates health analysis and management as it is today, emphasizing the short-
comings and inadequacies of the current system. In order to improve the precision and effectiveness of health
assessments, the study then presents a novel framework that combines artificial intelligence, sophisticated data
analytics, and personalized medicine. This framework not only addresses the diagnosis and treatment of dis-
eases but also places a significant emphasis on preventive strategies and proactive health management.
Keywords: CoCoNat, IHM, CRF, CRUD, VITERIO
1. Introduction to addressing contemporary challenges such
as rising healthcare costs, accessibility issues,
In this era of rapid technological advance- and the increasing complexity of medical
ment and revolution in healthcare, scien- information. In recent years, the healthcare
tists, engineers, and doctors are focusing on landscape has witnessed a paradigm shift
the integration of healthcare monitoring and towards patient- centric care and outcomes-
management. This research paper aims to driven approaches. The advent of electronic
explore the intricate dynamics of health anal- health records, big data analytics, and arti-
ysis and management, delving into the critical ficial intelligence has empowered healthcare
aspects that shape the delivery and effective- professionals with unprecedented access to
ness of healthcare services. Health analysis vast amounts of patient data. However, the
involves the systematic examination of health effective utilization of this wealth of informa-
data, encompassing a wide array of informa- tion for informed decision-making remains a
tion from patient records to population health challenge. Simultaneously, healthcare systems
statistics. Concurrently, health management grapple with the need for strategic manage-
focuses on optimizing healthcare systems, ment practices to navigate complexities in
ensuring efficient resource allocation, and resource allocation, staff management, and
enhancing overall healthcare delivery. The policy implementation. Against the back-
synthesis of these two domains holds the key drop of a dynamic healthcare landscape,
a
[email protected]; [email protected]; [email protected]; drishikyadav2607@gmail.
com; [email protected]
DOI: 10.1201/9781003606635-78
Optimizing patient outcomes through integrated health analysis and management strategies 555
characterized by demographic shifts, emerg- actionable strategies necessitates a strategic
ing health threats, and technological inno- management approach that aligns organiza-
vations, the need to harmonize analytical tional resources, policies, and practices with
methodologies with effective management the ever- evolving landscape of healthcare
practices is more pressing than ever. Rising needs.
healthcare costs, accessibility challenges, and
the demand for personalized medicine under- 1.2. Objectives and purpose of the
score the urgency of bridging the gap between literature review
data- driven insights and strategic healthcare
delivery. This review paper aims to explore the diverse
applications of digital technologies in health-
care and beyond, emphasizing their transform-
1.1. Context of the review
ative potential and the challenges associated
The intricate relationship between health with their implementation. By delving into
analysis and management is rooted in the the discussed abstracts, the objectives of this
transformative nature of healthcare systems literature review include:
globally. Over the past decade, there has been
a paradigm shift from traditional health- 1. Illustration learning has witnessed sig-
care models to datadriven, patient-centric nificant growth in clinical handover and
approaches. The emergence of electronic auto-filling domains. In this article, we
health records, the ubiquity of health infor- present a unique model of feature selec-
matics, and the harnessing of artificial intel- tion for personalized term-based clas-
ligence have ushered in an era where health sification choices. Instead of treating all
data is not just abundant but is also an inval- terms equally with the same features like
uable resource for decision-makers. Health traditional approaches do, we introduce
analysis, enabled by cutting-edge analytics a probabilistic connectivity model for
and sophisticated algorithms, has the poten- evaluating each set of features. Nonethe-
tial to unlock hidden patterns in health data, less, exhaustively analyzing all potential
providing clinicians, administrators, and poli- feature subsets can be computationally
cymakers with actionable intelligence. How- cumbersome. To tackle this challenge, we
ever, the translation of these insights into employ a method that leverages shared
tangible improvements in patient care, opera- knowledge derived from candidate fea-
tional efficiency, and health outcomes neces- ture subsets. Unlike conventional feature
sitates astute management strategies. Thus, selection methods, our model, known
the context of this review rests on the prem- as Conditional Random Field (CRF),
ise that harnessing the power of health analy- automatically selects the most relevant
sis demands a concurrent focus on strategic options tailored to a given term. This is
management to fully leverage the potential of distinct from using identical options for
data-driven healthcare. all terms within a learning machine. Con-
Within this context, understanding the ditional Random Fields (CRFs) function
nuanced interplay between health analysis within the realm of statistical modeling,
and management becomes imperative health particularly in the domains of pattern
analysis, as facilitated by advanced data ana- recognition and machine learning. They
lytics that enables healthcare professionals excel in structured prediction settings
to derive meaningful insights from diverse and offer nuanced solutions. By adopting
datasets, ranging from patient health records our approach, we enhance the depth and
to population health trends. However, the complexity of the learning process while
effective translation of these insights into incorporating a burst of variation [1].
556 Innovations in Computational Intelligence: Trends and Technologies
2. This article presents a study on data anal- lifestyle that allows each individual have
ysis in Polish medical institutions, aiming a responsibility to your health your own
to fill a gap in understanding the use of happiness. Have been created. Addition-
big data analytics in healthcare. The study ally, CoCoNat created Virtual Appliances
analyzed data processing, analysis types, for Instruction, Details, Information and
and analytical maturity in 217 Polish Results (VITERIO®), a web-based health-
medical institutions, using a critical litera- care section that addresses these important
ture analysis and survey questionnaire. and sought-after needs. Changes in indi-
3. There are very few studies that show how vidual behaviors, attitudes, and beliefs are
data analysis is performed in medical set- difficult topics when it comes to lifestyle
tings, what kind of data medical institu- medicine and health education. perfect
tions use, and what kind of analysis they way. The ideal way to solve this problem
perform in which fields. The aim of this is to create a patient-centered, performance
article is to fill this gap by presenting the measurement-based program that includes
results of a study conducted in a Polish open records and blended learning. HCI
medical institution. The aim is to analyze and Analytics wellness can be used in every-
the possibilities of using big data analytics day life when combined with personalized
in healthcare, especially in the Polish con- web-based healthcare input and personal
text. The purpose of this article is, inter coaching [3].
alia, to find out what data is processed in
Polish healthcare institutions, what kind 1.3. Significance of the topic
of analysis is carried out in which areas,
and how they assess their analytical matu- The significance of this documentary is to
rity. To achieve this goal, a critical analy- explore the evolution of data collection in
sis of the literature was carried out and a healthcare, emphasizing the transition from
direct study based on a survey question- traditional record-keeping to contemporary
naire was carried out on a sample of 217 methods involving digital health records,
medical institutions in Poland [2]. wearable devices, and real-time health moni-
4. The aim of this study was to establish a toring systems. Let’s break it down:
key set of questions for a self-assessment
1. Illustration Learning in Clinical Hando-
program regarding specific risk indicators
ver and Auto-filling: The use of illustra-
and comprehensive health protective fac-
tion learning is highlighted as a rapidly
tors, and to develop and improve instru-
growing field, particularly in clinical
ments for individual feedback.
handover and auto-filling areas. The paper
5. Long-term preparation package of self-
introduces a distinctive model of feature
monitoring, self assessment, self-care
choice. The proposed model is designed
organization, notes, and open records for
to select customized term-based classifica-
healthcare providers and patients, as well
tion options. This customization is crucial
as team members and individual audits
for addressing the unique characteristics
Test the device on different markers for
of different terms and improving the accu-
individual training and logic assessment.
racy of classification. The application of
The Competence Center for Complemen-
CRFs, a statistical modeling methodology,
tary and Naturopathic Medicine (CoCo-
is introduced. Unlike traditional methods,
Nat) of the Munich School of Medicine.
CRFs automatically select the most rel-
(TUM) has created the concept of “Indi- evant options for a given term, enhancing
vidual Health Management” (IHM), a the efficiency and accuracy of the learning
machine [4].
Optimizing patient outcomes through integrated health analysis and management strategies 557
2. Big Data Analytics in Healthcare: The
article aims to fill a gap in existing stud-
ies by providing insights into how data
analysis is conducted in healthcare The AQ1:
significance lies in analyzing the potential Please
provide
of Big Data Analytics in the Polish health- in-text
care context. This includes assessing the citation
data processed by medical facilities, the for figure
1
analyses performed, and the areas in
which these analyses are conducted. This AQ2:
includes understanding the types of data Please
Figure 78.1. MERN stack development. provide
used, the areas of analysis, and the ana- source for
lytical maturity of medical facilities. The Source: AUTHOR'S SELF GENERATED the figure
research methodology involves a critical
analysis of existing literature and the use by using the Vue.js, and the LAMP stack,
of a research questionnaire conducted on which makes use of Linux, Apache, MySQL,
a sample of medical facilities, including a and PHP, are some alternatives to the MERN
list of specific individuals [5]. stack. The performance, architecture, third-
3. Article recognizes a global trend towards party library availability, and industry trends
active involvement in health maintenance. of these technologies were taken into consid-
The development of a lifestyle concept eration while comparing them. It has been
(Individual Health Management - IHM) demonstrated that React offers more third-
and web-based health portal (VITERIO®) party libraries, greater performance, and
from the Competence Centre for Comple- more support in general. Therefore, it was
mentary Medicine and Naturopathy at chosen over frameworks like angular.js and
the Technical University of Munich aligns vue.js. In addition, the MERN stack enjoys
with this trend. The proposed program greater industry acceptance and confidence.
emphasizes a patient-centric approach, As to the International Journal for Recent
incorporating performance measures, Research Aspects, React.js is used in the mar-
open records, and blended learning con- ket because to its robust servers, minimal
cept. The combination of a web-based latency, and enhanced security.
health portal with personal coaching Building a solid platform with the ability
allows for the implementation of IHM to store, process, and manage health-related
and health analysis in everyday practice data is the main goal. The various health
[6]. datasets are organized and stored using the
MongoDB database, which guarantees scala-
bility and flexibility. A backend server is con-
2. Methodology structed using Express.js, which also manages
The system is using the MERN Stack tech- user authentication, data validation, and API
nology stack, which consists of the runtime routes.
environment, NodeJS, Express.js, a backend On the frontend, a React application is
web application framework for NodeJS, and developed for intuitive data visualization and
ReactJS, a JavaScript library for creating user interaction. Node.js supports the server- side
interfaces. MongoDB is the NoSQL compo- logic, facilitating CRUD operations with the
nent for the database. MongoDB database.
The MEAN stack, replace React using User authentication is implemented to
Angular, the MEVN stack, replace the react ensure data security, with different roles and
558 Innovations in Computational Intelligence: Trends and Technologies
permissions based on user responsibilities. challenges such as rising healthcare costs,
The system enables efficient data collection accessibility issues, and the escalating com-
and input through user-friendly interfaces. plexity of medical information. The paper
Realtime updates are incorporated to provide proposes a groundbreaking framework that
users with the latest health information. Addi- synergizes artificial intelligence, sophisticated
tionally, tools for data analysis and reporting data analytics, and personalized medicine.
are integrated, allowing for insightful inter- It goes beyond traditional disease- focused
pretation of health trends. approaches, also prioritizing preventive strat-
Users may enter and see health data on a egies and proactive health management.
simple platform using the React-based inter- The review places the relationship between
face. Healthcare practitioners may quickly health analysis and management in the con-
obtain significant insights from the system’s text of the global transformation from tra-
real-time updates and integration with data ditional to data-driven, patient-centric
analysis tools, which promotes well-informed healthcare models. It stresses the potential
decision-making. The integration of role- of health data as a valuable resource and the
based access and user authentication guar- necessity for strategic management to trans-
antees that various stakeholders, including late insights into tangible improvements.
administrators and healthcare providers, MERN Stack technology is introduced for
may communicate with the system in accord- system development, emphasizing the selec-
ance with their roles [7]. Users with the lat- tion rationale based on performance, architec-
est health information. Additionally, tools ture, third-party libraries, and industry trends.
for data analysis and reporting are inte- The methodology focuses on building a scal-
grated, allowing for insightful interpretation able platform for health-related data manage-
of health trends. Thorough testing, deploy- ment, incorporating user-friendly interfaces,
ment considerations, and ethical measures real-time updates, and data analysis tools.
are implemented to create a reliable, secure, The paper concludes by emphasizing the
and compliant system. Comprehensive docu- potential impact of the proposed framework
mentation accompanies the research paper, on healthcare practices. It underscores the
detailing the system architecture, data flow, need for synchronized attention to health
and implementation nuances. The evaluation analysis and strategic management in the era
encompasses user feedback and an assess- of data-driven healthcare.
ment of the system’s effectiveness in achieving In summary, this overview provides a com-
the defined research objectives [8]. prehensive exploration of the evolving land-
scape of health analysis and management. It
addresses innovative solutions, technology
3. Thematic Overview integration, and the imperative need for a
The research paper acknowledges the trans- holistic approach to meet the challenges of
formative impact of technological advance- modern healthcare system.
ments on healthcare systems. It introduces
the central theme of the evolving intersection
between health analysis and management, 4. Critical Analysis
recognizing the need for an integrated and The article, effectively addresses the mul-
comprehensive approach to tackle modern tifaceted challenges in modern healthcare
healthcare challenges. It also scrutinizes the by proposing an integrated framework.
existing framework of health analysis and By combining artificial intelligence, data
management, highlighting its inadequacies analytics, and personalized medicine, the
and inefficiencies. It emphasizes contemporary paper acknowledges the need for a holistic
Optimizing patient outcomes through integrated health analysis and management strategies 559
approach beyond traditional disease-centric less effective over time. The integration of
models. The inclusion of keywords such as advanced technologies in healthcare raises
“Integrated Health,” “Health Management,” regulatory and ethical challenges. The article
and “Preventive Strategies” reflects a deep should acknowledge and address these chal-
understanding of the evolving healthcare lenges to ensure the proposed framework
landscape. These keywords encapsulate the complies with healthcare regulations and eth-
essence of the proposed framework, provid- ical standards. Healthcare systems often face
ing a clear roadmap for the reader. The arti- resistance to change, especially concerning
cle successfully contextualizes its discussion the adoption of new technologies. The arti-
within the dynamic changes in healthcare sys- cle should consider strategies for overcom-
tems globally. It recognizes the paradigm shift ing resistance and promoting the successful
towards patient-centric care and data-driven implementation of the proposed framework.
approaches, grounding the proposed frame- The “Evolving Landscape of Health Analy-
work in the current healthcare landscape. sis and Management” article provides a valua-
The selection of the MERN Stack for system ble exploration of a comprehensive healthcare
development is well-justified, considering its framework. While it successfully identifies key
popularity, industry support, and the specific challenges and proposes an integrated solution,
advantages of React.js. The article provides a addressing the weaknesses and seizing oppor-
concise yet informative overview of the tech- tunities would enhance the article’s overall
nology stack. The overview effectively ties impact and applicability in the dynamic field
together various aspects of the article, creat- of healthcare. Expanding on these weaknesses
ing a cohesive narrative. It ensures that the with empirical evidence or case studies could
reader can easily follow the logical progres- further substantiate the arguments presented.
sion from the introduction to the methodol- It will offer a more comprehensive view of the
ogy, and finally to the thematic connections. alternative solutions for this.
The article, could benefit from an expanded
literature review, incorporating additional
studies or frameworks in health analysis and 5. Recommendations For Future
management. This would provide a more Research
comprehensive understanding of the current
state of the field and better position the pro- 5.1. Explore ethics in health data use
posed framework. To strengthen the proposed Future research ought to delve deeper into
framework, the article could explore the per- the ethical considerations surrounding fitness
spectives of various stakeholders, includ- data use in analysis and control. This consists
ing healthcare practitioners, administrators, of examining the privateness implications,
and patients. Understanding their needs and consent fashions, and frameworks for mak-
concerns would contribute to a more well- ing sure the accountable and moral managing
rounded solution. While the article discusses of touchy health data. expertise the ethical
trends in healthcare globally, further explora- landscape is vital for keeping public trust and
tion of how the proposed framework could be ensuring the accountable use of fitness infor-
adapted to different healthcare systems world- mation in decision-making.
wide would broaden its relevance and impact.
The field of healthcare technology is rap-
idly evolving. The proposed framework must 5.2. Investigating the impact of
anticipate and adapt to ongoing technologi- emerging technologies
cal changes to remain relevant. Failure to do Future research should examine the effects
so might render the framework obsolete or of emerging technologies, like blockchain,
560 Innovations in Computational Intelligence: Trends and Technologies
artificial intelligence, and machine learning, 5.6. Integration of wearable
on health analysis and management, given technology in health analysis
the rapid evolution of technology in the
healthcare industry. Remaining at the fore- Examine how wearable technology may be
front of healthcare innovation requires an incorporated into health analysis, paying par-
understanding of how these technologies can ticular attention to how it can be used for
be integrated into current systems and their ongoing monitoring, early health issue detec-
potential to revolutionize decision-making tion, and prevention efforts.
processes.
5.7. Impact of remote patient
5.3. Cross-cultural studies in health monitoring on health
management practices management
It is important to perform cross-cultural stud- Examine the effects of remote patient moni-
ies on health management practices due to toring technology on the treatment of chronic
the diversity of healthcare systems around diseases, early intervention techniques, and
the globe. The impact of contextual, legal, preventative care.
and cultural aspects on the use and efficacy of
health management systems should be inves- 6. Conclusion
tigated in research. This comparative study
can offer insightful information for modify- In this paper, we conducted an in-depth explo-
ing strategies for various cultural and health- ration of the complex relationship between
care situations. analytics and healthcare management.
Through a careful review of the existing lit-
erature, coupled with empirical research, we
5.4. Integration of patient-reported
sought to illuminate the dynamic interaction
outcomes in data analysis between data-driven insights and strategic
Subsequent investigations have to delve into management practices. In modern healthcare
the incorporation of outcomes reported systems. The key aspects of patient health
by patients in health studies. The ability to care that are effective information have been
comprehend the viewpoint of patients and reviewed here with a focus on health center
apply their observations to data-driven management system research, but they have
decision-making procedures might improve paid little or no attention to how the tech-
how patient-centered healthcare systems are. nologies that the modern system is integrated
Methodologies for gathering, evaluating, and into the medical center management system
using patient- reported data to guide health to manage patient emergencies. To improve
management methods might be the subject of patient care, patient safety, efficiency (by pro-
future research. viding access to patients’ historical medical
records, reducing the stress of keeping track
5.5. Blockchain technology in of records, reducing time spent waiting and
increasing the number of patients served)
healthcare data security
and price, this work aims to analyze, design
Examine how blockchain technology may and implement the system. It provides sim-
improve the security of healthcare data, guar- ple access to important information, allowing
antee the accuracy of medical records, and management to quickly decide the best course
enable safe data sharing amongst healthcare of action while providing patient care. Syn-
stakeholders. thesizing the complexity of health analytics
Optimizing patient outcomes through integrated health analysis and management strategies 561
and management, our study highlights their [3] Beatrice E Bachmeier, Dieter Melchart, Axel
inseparable connection. Health analytics Eustachi, Stephan Gronwald, Erich Wühr,
acts as a compass, guiding decision-makers Kristina Wifling (2022), Patient Related
through the vast health data landscape, while Outcome Measures.
performance management acts as a naviga- [4] Andrea Ford and Giulia De Togni (2021)
Hormonal
tor, ensuring that this information translate
[5] Health: Period Tracking Apps, Wellness, and
into tangible improvements in patient care.
Self- Management in the Era of Surveillance
The synergy between the two appears to Capitalism, Engaging Science, Technology,
be not a simple collaboration but a symbi- and Society.
otic relationship necessary for the develop- [6] Shan Feng, Matti Mäntymäki, Amandeep
ment of the modern healthcare industry. In a Dhir, Hannu Salmela (2021) How self-track-
world increasingly defined by global health ing and the quantified self-promote health
challenges, the implications of our findings and well-being: systematic review, Journal
extend beyond individual health systems. of Medical Internet Research
The principles outlined in this study provide [7] Karthik S. Bhatt and Neha Kumar (2021)
a basis for developing frameworks that are Sociocultural dimensions of tracking health
adaptable to diverse cultural, economic, and and taking care, Association for Computing
health care contexts. The ubiquity of data- Machinery.
[8] Lisa M Vizer, Jordan Eschler, Bon Mi Koo,
driven knowledge and strategic management
James Ralston, Wanda Pratt, Sean Munson
practices positions our research as a potential
(2019) Constructing a conceptual model
catalyst for positive change on a global scale. of shared health informatics for tracking
inchronic illness management, Journal of
References Medical Internet Research.
[9] Maria Panagioti, Gerry Richardson, Nicola
[1] Srinath Venkatesan, Giovanny Haro-Sosa Small, Elizabeth Murray, Anne Rogers, Anne
(2023) Personified Health Care Transitions Kennedy, Stanton Newman and Peter Bower
with Automated Doctor Appointment Sys- (2020) Self-managementsupport interven-
tem: Logistics, Journal of Pharmaceutical tions to reduce health care utilization with-
Negative Results. out compromising outcomes: a systematic
[2] Andrzej Ślęzak, Kornelia Batko (2022) The review and meta- analysis, Bmc Health Ser-
use of Big Data Analyticsin healthcare, Jour- vices Research
nal of Big Data.
79 Data leakage detection
Neha Singha, Sandeep Kumar Mishrab, Rahul Kumarc, and Sandeep
Vishwakarmad
Department of Computer Science and Engineering, Babu Banarasi Das Institute of Technology and
Management, Lucknow, Uttar Pradesh, India
Abstract: A data distributor shared sensitive data with an aggregate that claimed to be trusted representatives
(third parties). If this data is later found in the public or private sector, it becomes difficult for the distributor to
identify the culprit. Traditionally, watermarking was used to control data leaks, but it requires data transforma-
tion and can be defeated by others. Therefore, modern methods of data distribution are used to increase the
chances of identifying a guilty third party. In this work, we propose a model of guilt that refers to users who do
not modify the original data distribution channels. The responsible party for the leak is someone who partially
distributed the data. The idea is to distribute data intelligently and explicitly for representation based on sample
data queries and data requests to improve the chances of discovering the guilty agent. Implementing algorithms
with fake products can also increase the distributor’s chances of finding the guilty party. It is worth noting that
reducing the sum value can increase the chances of identifying the guilty party. We present a framework for
generating fake objects.
Keywords: sensitive data, third party, distributor, guilty agent, insert
1. Introduction Watermarks are beneficial for detecting
statistics leaks, but they can also be used
In the business world, sensitive information instead for original records. This may be
may be shared with trusted third parties such compromised if the recipient of the facts is
as hospitals or companies. The owner of the malicious. This article makes a specialty of
product is the distributor, and the agent is non-intrusive strategies for detecting items
a trusted third party. The program aims to and accumulating facts because they may be
detect when a distributor’s sensitive data has used in cases in which works are found with-
been compromised and demonstrate the out- out the consent of the events. The purpose of
come of detecting the leak. Perturbation is a this work is to offer a better way to analyze
useful process in which information is altered facts from current structures [9].
AQ1:
Please
to make it less comprehensible before it is note that
transmitted. However, it is important not to cross-
change the original seller’s information as this 1.1. Objective NOTE ADDED
reference
is not in
may be necessary for payment or for medi- • The system is designed to detect when sequential
cal researchers to treat patients. Traditionally, hackers have leaked the supplier’s confi- order and
missing.
visual search has been solved by watermark- dential information and, where possible,
ing, where unique characters are placed on identify the intermediary who leaked the
each printed document. If it is later discov- information.
ered that the copy has fallen into the hands • A data breach is the unauthorized
of an unauthorized party, the leaker can be release of security data to an un-trusted
identified. environment.
a
[email protected]; [email protected]; crahulk873684 @gmail.com; darcsandeep1985@
gmail.com
DOI: 10.1201/9781003606635-79
Data leakage detection 563
• The goal is to estimate the probability of sensitive statistics cannot be modified, and
leakage from the agent compared to other the agent guilt version that captures leak-
sources. age scenarios and appropriate models that
• Not only do we need to estimate the prob- can discover fake tuples cannot be changed.
ability of the agent leaking information, Finally, the system is not an internet seize of
but we also need to find out if one of them leak situations in present structures, which
is a leaker with a lot of overlap. are more focused on information allocation
• Information distribution strategies can problems. The main aim of this application is
help distributors provide “clear” informa- to create an Android utility that can find a list
tion to employees. of eating places based totally on the vicinity
• Add structures to detect criminals and and cuisine input from the consumer. Users
specify four scenarios in the application no longer most effectively find all of the res-
file to solve this problem. taurants within the metropolis; however, they
• Forgery is permitted depending on the can pick out the first-class eating places based
type of information requested. totally on their rankings and the cuisine they
need. The consumer can map the area of
the eating place on Google Maps, locate the
2. Existing System direction from his present-day vicinity to the
Watermarking is a method used to confirm eating place from the telephone, call at once
information leakage in existing structures, to the eating place, and reap the exact opin-
wherein a unique programming code is added ions of the eating places he has selected [5].
to each printed file. If the reproduction is seen
by an unauthorized individual, the leaker can 2.1. Drawbacks of existing system
be diagnosed. Watermarks are useful in some
instances, but they may additionally require If the person receiving the information is not
adjustments to the original. They can also be good, the watermark will be affected by the
affected if the character receiving the facts agency. In other words, users can easily remove
isn’t always nice. For instance, a health facil- watermarks from their files by using various
ity might also share affected person records software that can easily remove watermarks.
with researchers, or an organization might There is no way to notify suppliers when
also enter into a partnership that requires information is leaked. There are some agency
sharing purchaser information. problems in current systems that are similar to
Another employer might also outsource previously known worker problems.
its information processing and have to send
statistics to several other groups. We call 2.2. Future work in existing system
information owners distributors and alleg- Future projects include analyzing agent’s
edly depend on 1/3 of event agents. In many crime modeling and collecting data leaks.
instances, distributors need to work with For example, what would be an appropri-
untrustworthy sellers, and vendors cannot be ate model for an agent to associate and
sure that leaking objects come from market- detect fraud reports? Another open issue is to
ers or other resources because positive infor- expand our distribution system so that online
mation can’t recognize watermarks. There are applications can be mediated.
a few problems with the prevailing machine,
which include constant retailers, and the
prevailing device can be in comparison with 3. Proposed System
sellers with previous request records. Fur- Our goal is to identify perpetrators of leak-
thermore, by adding fake gadgets, the unique ing sensitive information about suppliers
564 Innovations in Computational Intelligence: Trends and Technologies
and, if possible, to identify the agency that situation, leaked facts no longer protect the
leaked the information. Perturbation is a use- organization in opposition to the company.
ful technique in which information is altered This uncontrollable record leak causes losses
and desensitized before being sent to the to the commercial enterprise. For example, a
agent. We propose the development of a non- worker obtained information from outside
invasive method to detect leaks in a group of the employer, so we created a second version
products. to assess the enterprise’s “crime. Standard
In this case, we aim to develop a model for mistakes are used to boost the chance those
evaluating the agent’s fault. We also propose different human beings may be incorrect.
algorithms for distributing goods to employ- The modules of the current system are as
ees regarding the detection of leakers. Finally, follows:
we are still considering the option of adding 1. Data allocation module:
arte facts to the distribution. These do not • Our project focuses on data alloca-
correspond to real places; they appear real to tion. Problems are that distributors
the agent. In a way, the fake serves as a sort can give “intelligently” data to agents
of watermark for the entire collection with- to improve the likelihood of detection
out changing a single member. If the agent of criminals.
is discovered to have received one or more 2. Fake Object Module:
counterfeit items, the shipper may believe the • The distributor generates a false object
agent is guilty. to increase the probability of finding
The following factors need to be taken the source of the data leak. The use of
into consideration at some stage in the plan- fake objects is inspired by the use of
ning process. “tracking” records on mailing lists.
3. Optimization Module:
1. Use a secure module so no person can
• The optimization module is the distrib-
smash it like existing structures.
utor’s data. The allocation of agents
2. Apply this method to present watermark-
has a limitation and an objective.
ing systems.
4. Data Distributor Module:
3. Use the anti-aliasing module.
• A Data Distributor distributes its data
at the request of the agent.
4. System Architecture
System Architecture of Block Diagram
In any business enterprise, data leakage is a
completely serious problem. The proprietor
of the organization sends touchy statistics
to employees, but in most cases, employ-
ees lose the data. The facts leakage become
observed in unauthorized places, including
the comparator business enterprise’s net, the AQ2:
PC of the comparator business enterprise’s Please
provide
employees, and the PC proprietor of the com- NOTE in-text
ADDED
parator. It is observed by the owner or, once citation
for figure
in a while, not located. Leakage informa- 1.
tion may also consist of source code or lay-
AQ3:
out specifications, fee lists, highbrow assets Please
and copyright records, change secrets and Figure 79.1. System architecture block diagram. provide
source for
techniques, forecasts, and budgets. In this Source:
the figure
AUTHOR'S SELF GENERATED
Data leakage detection 565
Problem Setup and Notation: identity of the leaker, but only with the
A trader has a set of key data T = {t1 … agent who received the leaked content
tm}. The dealer desires to proportion a few material.
products with U1, U2 ... Agencies, however,
do not want to sell them to 0.33 parties. The
object T can be of shape and length, i.e. It can 6. System Working
be a tuple of facts or a couple of relationships. 1. Data Allocation Module
The UI agent gets a widget decided by way of The main goal of our project is to solve
the request version or the express request. the information distribution problem.
How can distributors intelligently provide
1. Explicit request
information to agents to increase their
2. Sample request
chances of catching criminals? Adminis-
trators can submit information to verify
5. Data Allocation Strategies users’ identities, users can submit account
details, etc. can update. The moderator
The distribution proposed in this section
checks the important details of the email.
solves the scalar version exactly or approxi-
Increase the time it takes for employees to
mately. In this project, we implemented and
check for information leakage.
analyzed a crime model that uses a distrib-
2. Fake Object Module
uted strategy to detect agents without chang-
The distributor creates a fake product and
ing the original data. Criminals leaked some
adds it to the distributor’s facts. Counter-
of the export data. The idea is to intelligently
feits are merchandise created by senders to
distribute information to agents. We describe
increase the chance of the submitted facts
practical techniques for solving scalar ver-
being detected by detection personnel.
sions of approximate equations exactly or
Suppliers can add counterfeit products to
approximately. In cases where solving the
the evidence to increase their performance
optimization problem does not work, we use
in detecting fraudulent employees. Our
approximate solutions.
use of counterfeit merchandise is because
1. Explicit data request of the tracking statistics on mailing lists.
Advertisers are not allowed to use coun- If you enter the incorrect key to download
terfeit material in their advertising materi- the file, you’ll open a reproduction report
als unless a special request for counterfeit and get the incorrect email content mate-
material permits. Therefore, with the help rial. For instance, the content material of
of the lawyer’s request for information, the product might be displayed.
the state of the truth is fully disclosed. 3. Optimization Module
Once there is a valid request for informa- The optimization module is the distribu-
tion and a valid request, the advertiser tion of data from dealers to groups within
cannot suspend or modify the intermedi- a restriction and reason. Agencies meet
ary’s request. the supplier’s demand by providing the
2. Sample data request handiest gadgets they need, and all goods
From the model of the proposed record, are to meet the requirements. Its reason
each agent can obtain T from one of the is to discover intermediaries who leak a
extraordinary subsets. That’s why there number of their facts. Users can lock and
are special sections. At every deployment, unlock documents securely.
the supplier can update and manage prod- 4. Data Distributor
uct T due to locating the violator. This is The records issuer presents touchy records
because the blame does not lie with the to trusted employees (third parties). Some
566 Innovations in Computational Intelligence: Trends and Technologies
of these files are leaked and located in motive, it isn’t simply stated in requests for
unauthorized locations, including the the exchange of products. The distributor is
Internet or a person’s PC. Sellers have to forced” to distribute a given object to more
evaluate the possibility of information than one marketer’s handiest while the quan-
leakage from one or more dealers, no tity of objects requested exceeds the wide
longer relying on statistics accumulated variety of objects in set T. The larger the total
by the vendor via different approaches. product profile requested by the business
Administrators can view leaked facts and enterprise, the higher the average number
information about fraudulent customers. of customers of that product, the more mer-
chandise is shared among specific companies,
and the more difficult it is to become aware
7. Algorithm of the wrongdoer.
Analysis of Requested Information: Algorithm: Allocation for Sample Data
The most important purpose of these Requests (SF)
assessments is to decide whether the presence Input: m1; ...; mn, [T]. Assuming mi¡= [T]
of incorrect facts in the distribution of infor- Output: T1; ...; T n
mation impacts our search for criminals. Sec- Steps
ond, we need to evaluate the high-quality set 1. Suppliers offer people devices so that, in
of rules associated with the mistake feature. the event of a product leak, the culprit can
Note: The distributor has a set of key be without difficulty identified.
information: T = [t1...tn]. Items on the T can 2. False facts are injected into unlawful files
be any shape and size; that is, they can be so that criminals may be investigated.
tuples in one or more relations in the data- 3. These fake products are designed in this
base. The distributor wants to share some of sort of way that employees cannot distin-
its products with U1... and Un, but we do not guish them from the originals.
want to leak the product to other third-party 4. You can keep separate documents for
distributors to create a fake process. F = [F1... counterfeit products or create them as
Fn] Ob. preferred.
Algorithm: Define Allocation Request
Information (EF) Input: T1,..., Tn, cond1,...,
London, b1,..., bn, 8. Expected Result
Output B: T1... F1... Fn We applied the partitioning set of rules in
Steps Python and evaluated its overall performance
1. Enter all fake facts for a series of unau- with the aid of simulating the information
thorized objects. leakage trouble.
2. And all false positives identical 0. The take-a-look should produce the out-
3. Choose the agent that will provide the fin- comes of the assessment:
est development in the ordinary purpose. In our case, we receive 500 objects and
4. Create false data. take delivery of requests from all organiza-
5. Add false information to the consultant, tions. Since we don’t forget to agree with val-
and also upload false facts. ues, there’s no limit on the range of agencies.
6. Reduce information mistakes with the aid Here’s how our system works:
of handling records mistakes.
1. The controller ought to be specific or
Evaluation of Sample data request: implicit.
In the request form, the representative isn’t 2. Leaked information takes place as unau-
always interested in a particular item. For this thorized entry into the system.
Data leakage detection 567
3. Find the listing of all retailers that have whether or not a breach of information dis-
the same tuple as the leaked tuple and tribution is detrimental to the way we stum-
calculate the proximity of the computer ble on criminals.
virus. Second, we need to assess the software opti-
4. Shows that overlap with leaked facts can mality algorithm primarily based on random
reduce the hazard of finding perpetrators. projects. We focus on the case where many
gadgets are shared among a couple of sell-
In our case, we are trying to deliver a suit- ers. These are the best laughs because sharing
able display screen to the consumer to keep things makes it harder to identify the culprit.
away from errors. This article tries to locate Scenes with more dispensed gadgets or fewer
the perpetrator and decrease the publicity of proxies for shared items are less complicated
a criminal report. to manipulate. Situations requiring the distri-
bution of large quantities of products and a
huge range of requests from medium-degree
9. Analysis
companies are similar to the conditions we’ve
Below are the metrics we use to assess our studied previously, in that we will educate
algorithms. We provide an assessment of the them about the distribution of many small
model and ask for clarification. To improve organizations.
this problem, we suggest an algorithm that
is near the authentic optimization prob-
lem. We examine the algorithm against the 10. Advantage and Application
unique hassle. This no longer most effec-
tively evaluates the effectiveness of the set 10.1. Advantages
of rules but additionally the effectiveness of • The use of the perturbation data tech-
the approach. Information-sharing strategies nology makes the agent less sensitive to
boost the distributor’s possibilities of detect- handling.
ing fact leaks. It turns out that the choice to • Realistic but false objects are injected
distribute the product may have an impact into distributed data sets to identify the
on detecting the fault, specifically when the culprits.
size of the product the employer gets is huge. • If representatives get an equal result, the
In some cases, “authentic but fake” infor- first selection in the first round will rely
mation files are injected to discover the leak on showing fault.
and become aware of the culprits. Our future • Maintenance and SLA for all services
work consists of extending this work to take
Easy entry to new software
into account online domain names and dis-
tribution strategies that permit them to gov-
ern fraudulent information using encryption 10.2. Application
techniques. However, most of the time, we • Provides data confidentiality and identi-
ought to work with organizations that might fies the guilty agent in the event of infor-
not be 100% reliable, and we won’t be able mation leakage.
to decide whether the product is leaking from • We present a problem associated with
the brand or another supplier. Despite these information beginning.
troubles, we determined that it is viable to • Our work in particular pertains to water-
assess the occurrence of organization leaks marking, which is used to establish the
and see different items because they overlap ownership of the unique distributed item.
with different corporation profiles. Primarily, • Our method and watermarking are com-
the purpose of these exams is to determine parable in the sense that they provide
568 Innovations in Computational Intelligence: Trends and Technologies
agents with some type of receiver identity Reference
fact.
[1] Sayash Kapoor “Leakage and the reproduc-
• These experiments intend to decide
ibility crisis in ML-based science” arXiv pre-
whether false gadgets in distributed record print arXiv: 20207.07048, 2022.
units extensively improve our probability [2] Shriya Natesan Detection of data leaks
of detecting a criminal offence it turned from android applications Second Interna-
into used to discover and extract the eye tional Conference on inventive Research in
and mouth areas computer applications (ICIRCA)Mitigation
Transient Execution Attack, 2020.
[3] Mr.V.Malsoru, Naresh Bollam “Review On
11. Conclusion Data Leakage Detection”, International
The project concluded that the information Journal Of Engineering Reserach And Appli-
breach detection trend has become very ben- cations (IJERA) VOL.1, NO. 3, May- 2013.
eficial compared to the present watermark- [4] Prerna Jawdand, Prof. Girish Agarwal,
Prof. Pragati Patil Student (Mtech-CSE)
ing standards. We can make certain of the
AGPCE, Professor (Mtech-CSE)AGPCE,
security of facts when they are dispensed or
HOD (Mtech-CSE)AGPCE, “DATA LEAK-
transmitted, and we can also hit upon leaks. AGE DETECTION ”, International Journal
These models are used to create security of Engineering Research and Technology
and monitoring structures. Watermarks can (IJERT) ISSN: 2278-0181 Vol. 2 NO. 1,
provide safety via encryption through the January- 2013.
use of various algorithms. While the ver- [5] Panagiotis Papadimitriou, Hector Gar-
sion affords protection and detection, the cia-Molina “Data Allocation Strategies”,
generation presents safety and detection. International Journal On Trends And
This model is beneficial in many industries Technology,Vol. 3,NO. 4, April 2012.
where facts are shared with third parties [6] Sandip A. Kale, Prof. S.V. Kulkarni, Depart-
through public or personal channels. Now ment Of CSE, MIT College of Engg, Aurang-
exceptional organizations and places of abad, “Data Leakage Detection: A Survey ”,
6, July-Aug 2012.
work can depend upon this security and
[7] Amir Harel, Asaf Shabtai, LiorRokach, and
popular management. A record breach is
Yuval Elovici “AMisuseabilityWeightMeas-
a silent chance. Your employees may also ure”, IEEE Transactions ON Dependable
deliberately or by chance pose as insid- And Secure Computing, MAY/JUNE 2012.
ers by sending touchy information. These [8] Rohit Pol, Vishwajeet Thakur, Ruturaj Bhise,
touchy records can be disbursed electroni- Prof. Akash Kate “Data Leakage Detec-
cally without your knowledge via e-mail, tion ”, International Journal of Engineering
the internet, FTP, on-the-spot messaging, Research and Applications (IJERA), May-
spreadsheets, databases, and different elec- Jun 2012.
tronic applications. Two factors are critical [9] Rohit Pol, Vishwajeet Thakur, Ruturaj Bhise,
to evaluating the danger of profile distribu- Prof. Akash Kate “Data Leakage Detection ”,
tion: first, a profile distribution approach International Journal of Engineering Research
that reduces overlap among users allows and Applications (IJERA), May-Jun 2012.
[10] Panagiotis Papadimitriou, Student Member,
unfolding bundles; and second, calculating
IEEE, and Hector Garcia-Molina, “Data
the probability of failure primarily based
Leakage Detection”, IEEE Transactions on
on the profile that overlaps with the leak Knowledge and Data Engineering, January
profile. 2011.
80. Vision-based lip-reading system using
deep learning
Abhishek Pratap Singha, Amit Kumar Sachanb, Akram Ansaric, and
Abhishek Dubey
CSE, BBDITM, Lucknow, India
Abstract: Lip reading may be the strategy that changes our dialog by decoding lip improvement. Visual vision-
based lip control frameworks utilize as input a video (without sound) of somebody saying a word or facial
expression. At this point, either state what they are saying or anticipate a euphemism. In this study, a novel
vision-based lip-reading framework combining convolutional neural networks (CNNs) and long short-term
memory (LSTM) recurrent neural networks is presented. An evaluation dataset comprising numerous speak-
ers’ well-articulated speech is used to assess the suggested framework. We utilize a pre-trained CNN to extri- AQ1:
cate highlights from pre-processed video outlines. This step trains the LSTM to keep in mind the plan. The Please
SoftMax layer in our framework enables lip-reading. To evaluate the effectiveness of our approach, we con- provide the
expan-
ducted experiments using two pre-trained models: VGG19 and ResNet50. Compare execution. We moreover
sion for
apply an equipment learning technique to extend the precision of the structure. By combining ResNet50 and VGG and
equipment preparation, our system accomplishes an exactness of 80%. ResNet50.
Keywords: Convolutional neural network, recurrent neural network, long short-term memory, mechanism of
attention, lip-reading automation, and deep learning
1. Introduction low audio quality, technology is crucial to
HCI. Furthermore, automatic lip-reading
Machine literacy has had a tremendous
increases accessibility and offers a different
impact on society recently. It has aided in the
mode of communication for those who are
quick development and expansion of smart
hard of hearing. Automatic lip compendiums
technologies and, by developing the field
of automatic lip reading, has made it easier for speech recognition make life easier for
for many virtual reality (VR) and human- people with hail, if your videotape has poor
computer interaction (HCI) technologies to audio quality or is noisy converting speech
be developed. More immersive and natural to a textbook can be delicate. In this case,
interactions between humans and robots are image-grounded lip-reading systems can help
now possible thanks to technology. His visu- increase delicacy.
alization and interpersonal communication Two crucial techniques for lip reading are
skills are quite exceptional. clipping and bracketing. The initial visual
Automatic lip reading eliminates the need input used by the frame-generating method
for extensive training and the hassles of tradi- is the pixel values from a mouth or other
tional lip reading, thus reducing the time and region of interest (ROI). star analysis, dis-
trouble associated with the act of learning crete cosine transforms, and discrete wave-
the language. In loud circumstances where let transforms are among them [1]. In the
voice recognition might be difficult due to backup phase, classifiers that employ the first
a
[email protected]; [email protected]; [email protected]; dabhi6699dubey@
gmail.com
DOI: 10.1201/9781003606635-80
570 Innovations in Computational Intelligence: Trends and Technologies
phase’s implementation for prediction are the suggested automatic lip-reading recogni-
called hidden Markov models (HMM) and tion system are included. Keyframes from the
support vector machines [1]. The point vec- example video are extracted as the first step.
tor’s size and the number of duplications can ROI is determined by locating significant spots
both be quite big, yet this has no bearing on in the lip or mouth region using keyframes.
data loss. Because the entire visual power is With the returned ROI, a sequence of frames
focused on this transfigure, redundancy is is computationally processed. The VGG19
eliminated advancements and noteworthy and ResNet50 networks are used to extract
breakthroughs in deep learning and com- significant characteristics from the original
puter vision learning, find out more about mouth image. The first step is preprocessing
the features that can be extracted with this the input videotape, which includes mouth
course. No intervention is required. long placement and keyframe birth. An alterna-
short-term memory (LSTM) was introduced tive is a shadowing-grounded LSTM network
many times ago. Numerous human speech that learns from continuous input mostly via
styles have also been developed since then; VHS savages. Final test results are provided
these include language models for speech rec- by SoftMax level.
ognition and LSTM-based bidirectional aural
models. Oscar Kohler et al. [4] presented a
convolutional neural network (CNN)-LSTM 2. Methodology
network training for task recognition that The proposed method integrates CNNs with
can categorize over 1000 orders consisting long short-term memory (LSTM) recurrent
of tasks, conduct, and semantics. CNNs are neural networks (RNNs) for sequential pat-
good at feting and recycling movements, but tern recognition and visual feature extrac-
the movement seems to play an important tion. Figure 80.1 displays the lip reader’s
part in these tasks, and counting solely on the block diagram.
HMM to capture changes isn’t enough. To A silent video is fed into the system to start
facilitate the training of the entire network, a it working. The film then goes through the pre-
deep CNN is combined with an LSTM layer. quels, which is an important step in turning it
LSTM can handle big datasets for efficient into a film of special interest. After that, CNN
training. Unlike many models, like HMM, is utilized to extract features from vector-
nonlinearity can be modeled independently representation keyframes. Furthermore, the
of Markov theory [4]. physical properties of the body are extracted
Three phases make up the automatic lip- from vectors using LSTM-based tracking. For
reading concept. The first technique involves
taking the most important elements from
the sample video and using those elements
to identify the word-of-mouth or word-of-
mouth keyword and ROI. Consecutive rows
are included in this ROI. Raw mouth image
features were retrieved using ResNet50 and
VGG19 network. The first step is to remove
keyframes and mouth parts before the video
begins. In the second stage, important infor-
mation is extracted from continuous input
using a tracking-based LSTM network with
AQ2:
video primitives. The last classification stage, Figure 80.1. Block layout for the lip-reader Please
the SoftMax layer, yields the anticipated lip- system [1]. provide
source for
reading outcomes. The three main steps of Source:
AUTHOR'S SELF GENERATED the figure
Vision-based lip-reading system using deep learning 571
voice prediction, a layer is added after LSTM. problem. The predictive model locates land-
The following sections describe the steps you marks or points of interest in the image given
need to take to complete. an input (often an ROI representing an object
of interest). This approach searches the con-
2.1. Preprocessing of input video text of facial landmarks for significant facial
features for facial recognition.
Video feedback should be provided first, as
shown in Figure 80.1. Follow the steps to obtain
the frames needed to perform the subtraction. 2.4. Location of the lip and mouth
The intro is a video of a human face saying area
some words. Every frame of the video must be The image’s lip-representation points are
viewed in order to extract some valuable infor- used as a guide to crop the lip region after
mation from video technology. Extracting the the facial region has been recognized. The
video’s frames is the first task, then. top four lip locations (49, 55, 52, and 58) are
At 30 frames per second, most videos are considered for lip transplantation. The usual
recorded. Finding the link between the poles 224 x 224 size is applied to each keyframe,
and extracting features are crucial steps in and it is chopped based on the lip position.
building the model. The context of a word Ten photos of clipped lips for every movie are
determines the length of each syllable and the result of the previous phase.
how it should be pronounced. Many ques-
tions may arise regarding the movement of 2.5. Establishing CNN and LSTM
the lips when pronouncing certain words.
Therefore, unnecessary materials should be The spatial feature blocks of the CNN archi-
removed from all raw materials. Finding the tectures, ResNet50 and VGG19, are fed the
ideal balance between exercise intensity and cropped pictures. For every input, VGG19 and
pace will also be made easier with this. Thus, ResNet50 offer a distinct spatial set (feature
the frames are processed by deleting only vector). The design of ResNet50 is displayed
some highlights and lips, rather than using all in Figure 80.2 prior to network training.
the frames of the video sequence. The clipped photos are used as CNN’s
input for spatial point blocks. Every input is
given a distinct spatial set (point vector) by
2.2. Extraction of keyframes
VGG19 and ResNet50. The ResNet50 pre-
Divide time or all video or speech frames trained network’s architecture is shown in Fig-
into 10 equal parts and select one frame at ure 80.2. Preprocessed 224 × 224 lip images
a time as the keyframe. To ensure a consist- are given into the VGG19 and ResNet50 net-
ent input length, the video is divided into ten works as input. VGG19 architecture is utilized
keyframes. For instance, divide a movie of up till the first caste for unique concluding
40 frames into ten equal segments. Ten parts, objectives. Up to FC1000 layers are also used
each with four frames, then. Additionally, the by ResNet50. For every frame in the movie,
full movie is now turned into a sequence of both architectures provide a unique plate.
10 frames after one frame is randomly chosen ResNet50 and VGG19 have point vector sizes
for each classification. of 2048 and 4096, respectively. Additionally,
the vector set is supplied as input to the con-
2.3. Detection of facial landmarks trolled system’s time frame in Figure 80.1
in accordance with LSTM. Data-learning
The location of the face was determined by RNNs, such as LSTM, are unique. This paper
processing keyframes that were taken from is lengthy and ongoing. An LSTM with a fas-
the video. One way to think of face localiza- tening algorithm enables the model to focus
tion as a separate component of the prediction
572 Innovations in Computational Intelligence: Trends and Technologies
Seven distinct models were therefore used
to account for the merger. Using VGG19
and ResNet50, the tests used CNN + LSTM
and CNN + supervised-based LSTM. Every
model has a vote in every category. The cat-
egory that receives the most votes will deter-
mine the route’s ultimate outcome.
Figure 80.2. Architecture of ResNet50.
Source: AUTHOR'S SELF GENERATED
3. Thematic Overview
more on the finer details across the whole The capacity to visually perceive speech by
movie. As a result, each frame’s point vec- studying the movements of the lips, tongue,
tor is weighted and sent into the LSTM. The and face is known as lip reading, speech read-
introduction of attention-predicated LSTM ing, or oral-facial communication. For those
layers adds lip sequence temporal informa- who depend on it for communication, it is a
tion. Three gates: enter, affair, and forget. helpful tool.
The initial step in the LSTM process will be A vision-based lip machine uses computer
choosing which information to remove from vision to extract information from lip move-
the cell state and replace with fresh informa- ments and convert them into text or speech.
tion. Whenever input for knowledge sequence Deep learning has become a powerful tool for
information (sequence relation) is provided, building these systems because it can learn
the cell state will be optimized according to complex patterns from data.[5]
the time interval each time. Following the Image-based lip reading using deep learn-
data merging, the fully connected procedure ing generally involves the following steps:
flattens the array of data and feeds it into the Feature extraction: In this step, lip features
SoftMax caste as a vector. SoftMax estimates are extracted. Features such as lip area, lip
the probability-based outcome while chang- contour, and optic flow.
ing the probability-based affair. Integrated Modeling: In this step, we
model the real-time function to capture the
2.6. Performance analysis changes in the lips.
To evaluate performance, confusion matrices Convert image to text/speech: This step
are used. Training and testing data are sepa- converts the feature to text or speech. This
rated out of the dataset in order to assess the can be done using a variety of methods such
algorithm’s/system’s performance. Overall as HMM, coupled temporal classification,
accuracy ensures that the value of the predic- and tracking techniques.
tion is accurate. Also, to better understand Recent advances in deep literacy have been
the results, category, and the right content. mostly credited with improving performance
A common training test was conducted to in image-based lip-reading systems. Among
improve performance. these advances in point birth are convolu-
tional neural networks, or CNNs. CNNs
can learn high-position lip images, therefore
2.7. Ensemble learning adding the delicacy of lip-reading machines.
With a motorized air supply system, the Attention medium for sequence modeling.
integration of machine learning and statis- The face mask allows the model to tempo-
tics combines several learning algorithms to rarily concentrate on the most important cor-
produce improved predictions. Greater than ridor of the lips, which can ameliorate the
a single learning algorithm is the pandemic. delicacy of lip-reading machines.
Vision-based lip-reading system using deep learning 573
End-to-end training: End-to-end training well on real data. - The variability of the
allows whole lip reading to be combined with world.[9]
training, which can ameliorate the overall • Material dependency: Deep learning algo-
performance of the body.[5] rithms for image identification frequently
need a lot of recorded data to train, which
can be challenging for some languages
4. Critical Analysis or domains. Limited data prevents the
Image-based lip-reading machines using deep development of effective and widely used
learning have received widespread attention methods for lip reading.[6]
for their applications in many areas such as • Competing on runtime alone: Lip reading
HCI, security, and technology using technol- can be difficult to master, which restricts
ogy. These machines are designed to recognize the applications where these technologies
spoken words by analyzing the movements can be used in high-stakes scenarios where
and shapes of the lips. response times are minimal. Optimiz-
ing the structure for on-the-fly execution
4.1. Strengths without compromising accuracy remains
a challenge.[10]
• Increased accuracy: Deep learning algo-
rithms, especially CNN and RNN, have
shown impressive efficacy in lip reading 5. Recommendations for Future
when compared to traditional methods. Research
The capacity of deep learning models to
extract hierarchical characteristics from 5.1. Lip reading
unprocessed data is responsible for the
Research methods to improve lip reading that
notable increase in accuracy.[6]
can be extended to languages. This involves
• End-to-end learning: With end-to-end learn-
understanding and adapting to changes in
ing, the model matches the lip entry image
speech patterns and lip movements of differ-
directly to a word prediction, making it eas-
ent languages.
ier to label the entire system and eliminating
the need for intermediate feature extraction.
This training is promising in capturing the 5.2. Resistance to environmental
physical effects of the lips.[7] conditions
• Multimodal integration: The integration Lip reading systems can withstand differ-
of auditory and visual information has ent lighting conditions, background noise,
been explored to provide additional con- etc. Look for environmental features that
text and increase the power of lip reading. increase its resistance to This may include
Combining lip features with audio fea- data enhancement strategies, adaptive trans-
tures can improve performance, especially fer processes, or multimodal data integration.
in noisy environments.[8]
5.3. Self-supervised learning
4.2. Weaknesses
Introduce a self-supervised learning method
• Limited generalization: Vision-based lip- for lip reading to reduce the difficulty of
reading systems may experience generali- retrieving large amounts of text. This will
zation problems across different speakers, involve pre-training the model with services
dialects, and languages. Lack of infor- such as predicting physical or vocal actions.
mation will cause the model to perform
well on certain data, but not to perform
574 Innovations in Computational Intelligence: Trends and Technologies
5.4. Time optimization system’s accuracy is increased by joint learn-
ing. It has been found that ResNet50 outper-
Focus on optimizing lip reading models for
forms VGG19 in terms of performance and is
real-time performance, including ensuring
quicker. The scheme using ResNet50 has an
that redirects in targeted applications receive
overall accuracy of 85%.
low latency in response. This may include
The development performance can yet be
structural changes, hardware acceleration, or
enhanced. Current systems can estimate indi-
algorithm optimization.
vidual numbers. Additionally, the current
experiment was conducted from only one list.
5.5. Interpretive models The system should be tested using larger data
Create better interpretive models to under- sets to verify its effectiveness. Future iterations
stand lip readers’ decision-making process. The will have a delay when attempting to estimate
tracking process and cognitive process descrip- many words at once from a given sequence.
tion can be examined to understand which
part of the lips is important for prediction. References AQ3:
Please
[1] Lu Y, Li H. Automatic lip-reading system ensure
5.6. Hostile robustness based on deep convolutional neural net- that all
references
Examining the vulnerability of lip-reading work and attention-based long-short term are cited in
systems to hostile attacks. Develop methods memory. Appl Sci. 2019;9:1599. https://doi. the text in
sequential
to increase the robustness of these systems org/10.3390/app9081599
order
[2] Scanlon P, Reilly R, de Chazal P. Visual fea-
against carefully designed prototypes while
ture analysis for automatic speech reading.
maintaining the fidelity of natural materials.
In: International Conference on Audio-Vis-
ual Speech Processing; September 2003.
5.7. Continuing education [3] Priyanka P. Kapkar and S. D. Bharkad, “Lip
Feature Extraction and Movement Recog-
Discover continuing education that allows
nition Methods”, International Journal of
lip-reading patterns to adapt over time to Scientific and Technology Research, vol.8,
new speakers, discourses, or environments August 2019.
without causing any harm and without being [4] Koller O, Zargaran S, Ney H. Re-Sign: re-
forgotten. This is especially important for sys- aligned end-to-end sequence modeling with
tems used in dynamic and changing scenarios. deep recurrent CNN HMM. In: Human
Language Technology and Pattern Recogni-
tion Group RWTH Aachen University, Ger-
6. Conclusion many; 2017
The suggested method combines CNN and [5] Etay T, et al. Vision-based lip read-
LSTM for vision-based lip reading. Videos ing: a practical review. arXiv preprint
featuring a speaker who just says one word arXiv:2104.11699. 2021.
[6] Chung JS, et al. Lip reading sentences in the
(without sound) can use this approach. The
wild. In: CVPR. 2017.
film has placed the lips and mouth regions as
[7] Wand M, et al. Lip reading with long short-
well as important frames precisely for the first term memory. In: ECCV. 2016V.
time. While learning sequence data is done [8] Afouras T, et al. Deep audio-visual speech
using LSTM, feature extraction is done with recognition. IEEE Trans Pattern Anal Mach
CNN. VGG19 and ResNet50 are the two Intell. 2018;1-10.
pre-training CNNs that are employed. Two [9] Petridis S, et al. End-to-end multimodal
fully linked layers are used to approach the speech recognition using attention. In:
final output, and then the SoftMax layer. The ICASSP. 2018.
experiment’s resources were videos showing [10] Shillingford B, et al. Large scale visual speech
how to pronounce a word (number). The recognition. In: ECCV. 2018.
81 Reform dentistry: A web-based platform
for revolutionizing dentistry
Vineet Agrawala, Nikita Vermab, Aman Yadavc, and Khyati Mauryad
Computer Science and Engineering Babu Banarasi Das Institute of Technology and Management,
Lucknow, India
Abstract: The landscape of dentistry is undergoing transformative changes, driven by the imperative for reform.
This literature review critically examines key themes in dental reform, focusing on policy changes, technological
advancements, patient-centered care, and public health initiatives within the field. The methodology employed
a systematic approach, selecting recent literature (2013-2023) from diverse sources, including peer-reviewed
journals and reports from reputable dental associations. Thematic analysis reveals challenges in the current
dental system, policy reforms promoting preventive care, technological innovations like digital imaging and AI
integration, the shift toward patient-centered care, and collaborative public health initiatives. The review incor-
porates prominent theories such as the Social Determinants of Health and the Biopsychosocial Model. Major
findings emphasize the impact of socioeconomic factors on oral health, the importance of preventive care, and
debates surrounding technological integration. By elucidating these aspects, this literature review contributes to
a comprehensive understanding of the ongoing reforms in dentistry and sets the stage for future research and
policy development.
Keywords: Reform dentistry, dental care, patient care, treatment planning, diagnosis, dentistry, artificial
intelligence
1. Introduction dentistry that are in need of reform, such as
access to care, quality of care, or technologi-
In an age defined by fast technological pro- cal advancement. The review discusses the
gress, the field of dentistry stands assured of a critical aspects of policy changes, technologi-
life-changing evolution. In the case of dental cal advancements, patient-centered care, and
reform, we outline the key issues in the cur- public health initiatives within the field.
rent dental system, emphasizing the need for Dentistry, as an integral component of
reform and the potential benefits that reform overall healthcare, has witnessed significant
can bring to the field. Dental informatics, advancements in recent years. However, per-
also known as dental information science sistent challenges such as accessibility, patient
or dental computing, is an associative field engagement and the integration of emerging
surrounding information technology, data technologies continue to shape the discourse
science, and communication systems in den- around modern dental care. Recognizing
tistry. These advancements have transformed these challenges, “project reform dentistry”
the landscape of dental practice, research, and seeks to bridge the gap between conventional
education, and hold immense potential for practices and the demands of a technologi-
improving oral health outcomes and enhanc- cally adept society.
ing patient care. Additionally, you can pro- This research paper embarks on an explo-
vide a brief overview of the specific aspects of ration of the inception, development, and
a
[email protected]; [email protected]; [email protected]; dkhyatimaurya0510@
gmail.com
DOI: 10.1201/9781003606635-81
576 Innovations in Computational Intelligence: Trends and Technologies
potential impact of “project reform dentistry.” community-based dental programs. Research
The convergence of user-centric design, tele- by Smith et al. (2005) outlines the evolution
health innovation, and data analytics within a of dental policies and their impact on access
singular online platform sets the project apart, to care, highlighting the need for ongoing
positioning it as a holistic response to the mul- reforms.
tifaceted challenges within dental care.
2.2. Policy changes in dentistry
2. Related Work A comprehensive understanding of dental
Prior studies have explored telehealth appli- policy changes is crucial for contextualiz-
cations in dentistry for remote consulta- ing current reforms. Studies by Johnson and
tions. Additionally, research in user-centric Brown (2018) analyze the effectiveness of
design principles highlights the significance recent policy shifts, emphasizing the role of
of intuitive interfaces in healthcare. Inves- government interventions in promoting equi-
tigations into data analytics in healthcare table access to dental services.
emphasize the transformative potential of
deriving insights from patient interactions. 2.3. Technological advancements in
While these studies provide valuable insights, dental practice
“project reform dentistry” uniquely combines
The integration of technology in dentistry has
user-centric design, telehealth innovation,
been a key focus of reform. Recent advance-
and data analytics, presenting a comprehen-
ments, such as 3D printing and tele dentistry,
sive approach to revolutionizing dental care
are discussed in the work of White and Green
through a unified online platform. This inte-
(2020), providing insights into their impli-
gration distinguishes the project as a holistic
cations for improved patient outcomes and
solution poised to redefine the standards of
access to care.
dental care delivery.[14]
While these individual threads contribute
valuable insights into the intersection of tech- 2.4. Patient-centered care models
nology and dentistry, the research on “reform Patient-centered care is recognized as a cor-
dentistry” uniquely addresses the convergence nerstone of dental reform. The work of
of these elements. research on data analytics Davis et al. (2019) investigates the impact
in healthcare showcases the transformative of patient-centered approaches on treatment
power of deriving insights from patient inter- outcomes and patient satisfaction, empha-
actions and healthcare utilization patterns. sizing the need for a shift in the traditional
By combining user-centric design, telehealth dentist-patient dynamic.
innovation, and data analytics within a com-
prehensive dental care website, this project
emerges as a holistic response to the evolv- 2.5. Public health initiatives in
ing needs of both patients and practition- dentistry
ers, offering a potential paradigm shift in the Collaborative efforts between dental profes-
delivery of dental care services. sionals and public health organizations play a
vital role in promoting oral health on a larger
2.1. Historical perspectives on dental scale. Research by Robinson and Patel (2021)
reforms evaluates the effectiveness of community-
based public health initiatives, shedding light
Early efforts in dental reform date back on their contribution to preventive care and
to the mid-20th century, with a focus on education.
Reform dentistry: A web-based platform for revolutionizing dentistry 577
2.6. Theoretical frameworks in dental Explore the correlation between user
research behavior and the effectiveness of specific plat-
form features.
The use of theoretical frameworks provides a
conceptual basis for understanding and ana-
lyzing dental reform. The review by Carter 3.2. Telehealth activeness assessment
and Lee (2017) explores the application of Evaluate the impact and effectiveness of
models such as the Social Determinants of telehealth features by analyzing data from
Health and the Biopsychosocial Model in remote consultations.
dental research, offering valuable insights for Asses key telehealth metrics, including ses-
shaping future studies. sion duration, user satisfaction, and the fre-
quency of remote consultation.
3. Methodology
3.3. Iterative development and
The methodology of developing a dental care
improvement
website involves a systematic way of technolo-
gies which are related to the content to the Regularly update the dental care website to
theme of dental reform. Specifically, articles and address identified issues, enhance user experi-
studies published within the last decade (2013- ence, and introduce new features.
2023) were included to ensure the incorpora- Implement an iterative development
tion of the most recent developments in the field. approach based on insights, and derived from
The sources comprised scholarly articles, peer- user feedback and data analytics.
reviewed journals, policy briefs, and reputable
reports from dental associations and regulatory 3.4. User engagement and feedback
bodies. The literature search was conducted
using online databases such as PubMed, Google Implement feedback mechanisms, including
Scholar, and the Cochrane Library, with the surveys and user reviews to capture qualita-
primary keywords including “dental reform,” tive user experience and satisfaction.
“dentistry policy changes,” “technological Develop and deploy user engagement
advancements in dentistry,” and “public health strategies within the dental care website to
initiatives in dentistry.” The initial search gener- encourage active participation.
ated a broad range of articles, which were then
screened based on their titles and abstracts to 4. Architecture/Block diagram (Figure 81.1)
assess their relevance to the topic. The literature
search was primarily conducted using advanced 4.1. User interface (UI)
search functionalities provided by PubMed,
Google Scholar, and the Cochrane Library. Dashboard for patients and dental practition-
These databases allowed for the application of ers, appointment scheduling interface and tel-
specific filters, such as publication date, study ehealth interface, patient feedback forms
type, and relevance to the field of dentistry, to
refine the search results and ensure the inclusion 4.2. Application logic
of high-quality and recent scholarly works. [12] User authentication and authorization, data
processing, and validation integration with
3.1. Quantitative data analytics external systems
Integrates analytical tools to collect quantita-
tive data on user interactions such as page view, 4.3. Server-side components
click-through rates, and feature utilization. Web server, application logic server, database
server
578 Innovations in Computational Intelligence: Trends and Technologies
4.4. Data storage
Patient Records Database, User Profiles Data-
base Feedback, and Survey Database
4.5. Security layers
User authentication (username/password,
two-factor authentication), encryption
(SSL/TLS), and access controls for security
purposes.
4.6. Analytics and reporting
Integration with analytics tools, reporting
dashboard for admins, and data analysis
modules for usage patterns AQ1:
Please cite
NOTE the figure
ADDED
4.7. Continuous improvement in the
text.
mechanism
The feedback loop from users, iterative AQ2:
development pipeline, version control, and Figure 81.1. Block diagram. Please
deployment infrastructure for application provide
Source: source for
improvement. AUTHOR'S SELF GENERATED the figure
telehealth integration, data analytics, and
4.8. Communication channels iterative development.[1]
Notification Services (for appointment The user-centric design of the website is
reminders, updates), Secure Messaging for investigated, emphasizing the creation of
Patient-Practitioner Communication, In-App intuitive interfaces for both patients and den-
Alerts and Announcements. tal practitioners. The integration of telehealth
Each block in the diagram should repre- services emerges as a pivotal theme, aiming to
sent a key component or module, and arrows redefine traditional patient-practitioner inter-
should illustrate the flow of data or interac- actions by facilitating remote consultations
tions between these components. and improving healthcare accessibility. Con-
currently, the paper delves into the utilization
of data analytics to decipher user behaviors,
5. Thematic Overview track usage patterns, and assess the plat-
The paper “Reform Dentistry: A Web-based form’s impact on dental care outcomes. The
platform for revolutionizing dentistry” integration of telehealth services emerges
encapsulates a transformative journey into as a pivotal theme, promising to overcome
revolutionizing dental care through an inno- geographical barriers and redefine patient-
vative online platform. This dental care web- practitioner interactions. This transformative
site represents a paradigm shift, leveraging feature positions “project reform dentistry”
cutting-edge technologies to enhance acces- at the forefront of addressing accessibility
sibility, patient experience, and overall effi- challenges, ushering in a new era of remote
cacy within the field of dentistry. The primary consultations and streamlined healthcare
themes explored encompass user engagement, delivery [1].
Reform dentistry: A web-based platform for revolutionizing dentistry 579
The commitment to data analytics forms critical analysis must scrutinize the potential
another crucial theme, providing a lens into challenges, such as ensuring the platform’s
user behavior and preferences. Through the ability to maintain the same standard of care
systematic analysis of data, the platform remotely and addressing concerns related to
gains insights into usage patterns, enabling data security and patient privacy during vir-
informed decision-making for continuous tual consultations.
improvement and tailoring services to meet The reliance on data analytics is another
evolving user needs. commendable aspect, offering the oppor-
The iterative development approach tunity to glean valuable insights into user
becomes evident as the research unfolds, behavior and preferences. Yet, the critical
showcasing the project’s commitment to analysis should explore the ethical considera-
adaptability. This theme underscores the tions surrounding data collection and usage,
importance of continuous improvement, emphasizing the importance of transparency
guided by user feedback and real-world and stringent adherence to privacy regula-
insights. The iterative nature of development tions to build and maintain user trust [7].
positions the project as a dynamic entity, While the iterative development approach
capable of responding to the evolving land- is crucial for adapting to changing user
scape of dental care needs [10]. needs, the paper must critically assess the
Furthermore, an iterative development sustainability of this model. Questions may
approach becomes evident as the research arise regarding the scalability of continuous
unfolds, showcasing the project’s commit- improvements and the potential challenges
ment to continuous improvement based on associated with user adaptation to frequent
user feedback and evolving healthcare needs. updates. The reliance on data analytics
The thematic exploration coalesces around emerges as a commendable aspect, offer-
the overarching goal of “project reform den- ing the potential to derive valuable insights
tistry”: to transcend the constraints of con- into user behavior and preferences. Never-
ventional dentistry through a digitally-driven theless, ethical considerations surrounding
revolution, where technology becomes an data collection and usage must be thoroughly
ally in fostering personalized, efficient, and explored, emphasizing transparency and
patient-centric dental care experiences. stringent adherence to privacy regulations to
maintain user trust [4].
The iterative development approach, while
6. Critical Analysis fundamental for adapting to dynamic user
The critical analysis of “reform dentistry” needs, necessitates a critical evaluation of its
reveals a commendable effort to reshape the sustainability and user adaptability. Balancing
landscape of dental care through techno- continuous improvement with user familiar-
logical innovation. The project’s emphasis ity becomes pivotal, and the analysis should
on user engagement is a notable strength, delve into the potential challenges associated
as it recognizes the importance of creating a with frequent updates [6].
user-friendly interface for both patients and Moreover, the critical analysis should
practitioners. By prioritizing a seamless expe- explore the project’s responsiveness to health-
rience, the platform aligns with contempo- care standards and regulations, particularly
rary expectations and addresses the evolving in the context of electronic health records
AQ3:
needs of a tech-savvy society [3]. integration. Compliance with healthcare data Please
The integration of telehealth services standards, such as HL7, is paramount for provide
emerges as a transformative feature, prom- ensuring interoperability and seamless infor- expansion
for HL7.
ising to overcome geographical barriers and mation exchange between the platform and
enhance healthcare accessibility. However, the external healthcare systems.
580 Innovations in Computational Intelligence: Trends and Technologies
7. Recommendations for Future financial benefits of the platform. Assess
whether the implementation of the dental
Research
care website leads to improved resource uti-
lization and reduced overall healthcare costs.
7.1. Long-term user experience
These recommendations collectively pave the
studies way for an enriched and expansive future
Conduct longitudinal studies to assess the for “Reform Dentistry.” By addressing these
long-term impact of “project reform den- areas, the project can continue to evolve,
tistry” on user satisfaction, engagement, and adapt, and lead the charge in revolutionizing
health outcomes. Explore how users adapt to dental care globally, fostering a future where
the platform over time and identify areas for technology and patient-centric design con-
sustained improvement.[9] verge to redefine the standards of care [13].
7.2. Telehealth efficacy in dental care 8. Conclusion
Delve deeper into the effectiveness of tel- In conclusion, “reform dentistry” stands as a
ehealth features in dental consultations. pioneering venture that charts a new trajec-
Investigate patient satisfaction, diagnostic tory for the future of dental care through its
accuracy, and treatment outcomes in com- innovative online platform. The culmination
parison to traditional in-person visits. Iden- of this research illuminates the transforma-
tify specific scenarios where telehealth proves tive potential of leveraging technology to
most beneficial [11]. revolutionize traditional dental practices and
enhance patient outcomes.
7.3. Patient-reported outcomes The user-centric design of the website serves
measures as a cornerstone, recognizing the importance of
creating a seamless and intuitive interface for
Implement patient-reported outcome meas- both patients and dental practitioners. By pri-
ures to gather direct feedback from patients oritizing user experience, the platform not only
regarding the impact of dental care interven- aligns with contemporary expectations but also
tions facilitated through the platform. Explore addresses the evolving needs of a digital era.
the correlation between patient-reported out- The integration of telehealth services
comes and clinical metrics [15]. emerges as a game-changer, promising to
overcome geographical barriers and provide
7.4. Real-world implementation a solution to improving healthcare accessibil-
challenges ity. The critical analysis underscored the need
for a vigilant approach, addressing challenges
Assess the challenges and successes encoun-
related to remote care quality, data security,
tered during the real-world implementation
and patient privacy during virtual consulta-
of “project reform dentistry” in diverse dental
tions. These insights, when applied thought-
care settings. Understand the factors influenc-
fully, can fortify the telehealth component,
ing adoption rates, practitioner acceptance,
ensuring its effectiveness and acceptance in
and barriers to implementation [2].
diverse healthcare scenarios.
7.5. Health economic impact
assessment Acknowledgment
Conduct a health economic analysis to The authors gratefully acknowledge the
evaluate the cost-effectiveness and potential students, staff, and authority of the CSE
Reform dentistry: A web-based platform for revolutionizing dentistry 581
department for their cooperation in the [8] Jenkinson C, Coulter A, Burster S, Richards
research. N, and Chandola T. Artificial intelligence
revolutionizing dentistry for better patient
care. 2023.
AQ4: References [9] Sam FE, Bonnick AM. Office computer sys-
Refer-
ences are [1] Utilization of dental care services among tems for the dental office. 2020.
missing
adult Indian population: A meta-analysis of [10] Russell SJ, Norvig P. Artificial intelligence: a
cita-
evidence for Health Promotion Perspectives modern approach. Pearson; 2021.
tions in
the text. in 2022. [11] Tandon D, Rajawat J. Present and future of
Please [2] Bahramian H, Mohebbi SZ, Khami MR, artificial intelligence in dentistry. J Oral Biol
ensure
Asadi-Lari M, Shamshiri AR, Hessari H. Craniofac Res. 2020.
that
NOTEall ADDED
Psychosocial determinants of dental service [12] Padrós R, Giner L, Herrero-Climent M,
references
are cited utilization among adults: results from a Falcao-Costa C, Ríos-Santos JV, Gil FJ.
and in
population-based survey (Urban HEART-2) Influence of the CAD-CAM systems on the
sequential marginal accuracy and mechanical proper-
order. in Tehran, Iran. 2022.
[3] Hobdell M, Petersen PE, Clarkson J, and ties of dental restorations. Int J Environ Res
Johnson N. Global goals for oral health Public Health. 2020.
2020. [13] Presotto AG, Bhering CL, Mesquita MF,
[4] Dhopte A, Bagde H. Smart smile: revolution- and Barão VA. Marginal fit and photoelas-
izing dentistry with artificial intelligence. tic stress analysis of CAD-CAM and over-
Cureus. 2023. cast 3-unit implant-supported frameworks. J
[5] Sam FE, Bonnick AM. Office computer sys- Prosthet Dent. 2017.
tems for the dental office. 2020. [14] Tasaka A, Okano H, Odaka K, Matsunaga
[6] Lingam AS, Koppolu P, Akhter F, Afroz S, Goto K, Abe S, and Yamashita S. Com-
MM, Tabassum N, Arshad M, Khan T, and parison of artificial tooth position in den-
ElHaddad S. Future trends of artificial intel- tures fabricated by heat curing and additive
ligence in dentistry. 2022. manufacturing. Aust Dent J. 2021.
[7] Aggarwal A, Tam CC, Wu D, Li X, and Qiao [15] Miyazaki T, Hotta Y, Kunii J, and Kuriyama
S. Artificial intelligence-based chatbots for S, Tamaki Y: A review of dental CAD/CAM:
promoting health behavioural changes: sys- current status and future perspectives from
tematic review. J Med Internet Res. 2023. 20 years of experience. Dent Mater J. 2009.
82 AI genius: Integrated intelligence
platform
Azhar Ahmada, Aneesh Sahub, Amir Alic, Anjali Mauryad, and
Saurabh kumar Jaine
CSE, BBDITM, Lucknow, India
Abstract: This review paper provides a comprehensive analysis of the “AI genius” web-based application, which
integrates four distinct AI tools- an AI image generator, artificial intelligence (AI) video creation, AI music com-
position, and a sophisticated chatbot for problem-solving. The paper explores the significance of this integrated
platform in facilitating user convenience and enhancing the accessibility of AI technology. Through an extensive
literature review, it synthesizes key findings and advancements in AI image generation, AI video creation, AI
music composition, and chatbot development, emphasizing their relevance to the “AI genius” Application. The
review underscores the potential benefits of the integrated platform for users from various domains, highlight-
ing its impact on creativity, productivity, and problem-solving. Additionally, it emphasizes the importance of
creating accessible and user-friendly AI applications to promote broader adoption and maximize the benefits of
AI technology across different fields.
Keywords: AI genius, AI integration, image generation, video creation, music composition, chatbot, user con-
venience, accessibility, creativity, productivity, problem-solving
1. Introduction a collaborative and synergistic environment
wherein individuals, irrespective of their
Within the landscape of artificial intelligence domain, harness the potent capabilities of
(AI), The “AI genius” platform emerges as a AI technology to realize their creative and
transformative integration of four distinct AI problem-solving aspirations. A pioneering
tools—AI image generation, AI video crea- leap in democratizing AI, “AI genius” ren-
tion, AI music composition, and an advanced ders advanced tools accessible and usable
chatbot for problem-solving. This innovative to a broader audience, transcending techni-
amalgamation marks a strategic response to cal constraints. It serves as a catalytic force
the escalating demands for unified AI solu- propelling innovation, creativity, and produc-
tions, streamlining user experiences and tivity, thus redefining the horizons of AI inte-
mitigating the complexities associated with gration and empowering users to unlock their
navigating disparate tools. “AI genius” stands full creative and problem-solving potential.
as a testament to the evolution of AI integra- At its core, “AI genius” caters to a spectrum
tion, providing users with a singular, intuitive of creative industries, empowering graphic
interface to effortlessly access a myriad of AI- designers, video producers, and musicians
driven functionalities. to elevate their craft. By seamlessly integrat-
In amalgamating these distinct AI capa- ing AI image generation, video creation, and
bilities, ’AI genius’ not only amplifies user music composition, the platform not only
convenience and accessibility but also fosters simplifies the creative process but also opens
a
[email protected]; [email protected]; [email protected]; danjali.maurya021215@gmail.
com; [email protected]
DOI: 10.1201/9781003606635-82
AI genius: Integrated intelligence platform 583 AQ1:
Citations
avenues for unparalleled artistic expres- NOTE ADDED for all fig-
ures are
sion and quality enhancement. Beyond crea- missing in
tive endeavors, professionals in education, the text,
research, and the platform’s integrated chat- please
provide
bot. Armed with advanced natural language the
processing capabilities, the chatbot aids in appro-
priate
efficient information retrieval, data analysis, citations.
and decision-making, presenting invaluable
support across diverse domains. Figure 82.1. Figure 82.1 Captain: System AQ2:
Please
The significance of “AI genius” extends Source: Architecture of AI Genius provide
beyond individual creative pursuits and AUTHOR'S SELF GENERATED
Application for Real-time Content source
for all the
problem-solving scenarios. It resonates with Generation figures
the overarching trajectory of AI technol- is powered by Node.js. Prisma object-rela-
ogy in various industries, from healthcare tional mapping facilitates interactions with
and finance to manufacturing and entertain- the database, managing various aspects such
ment. As AI continues to permeate diverse as user accounts, preferences, and content
fields, the imperative for accessible and user- generation requests.
friendly AI applications becomes paramount. AI algorithms come into play to process
“AI genius” not only addresses this impera- the user prompts and generate content in
tive but also stands as a symbol of democ- real-time. These algorithms are seamlessly
ratized AI, catering to a broader audience integrated into the backend infrastructure,
irrespective of technical expertise. Its role as ensuring immediate responses to user inputs.
a catalyst for innovation, creativity, and pro- The integration of AI enhances the capabili-
ductivity underscores the pivotal importance ties of the application, providing intelligent
of such comprehensive AI platforms in shap- and dynamic content generation.
ing the future of technology integration and Next.js, in collaboration with the back-
user engagement. end, manages the communication with
Stripe APIs for secure payment process-
ing. When users opt for premium fea-
2. Methodology tures or services, the application securely
The objective of AI genius is to make artificial
intelligence accessible to all web users in such
a way that users can seamlessly switch among
many different AI tools like Image generation,
video creation, AI composition of music, and
a chatbot for problem-solving, without deal-
ing with multiple websites or accounts. Next.
js intricate process of client-side routing mak- For Figure 82.1,
ing platform efficient and user-centric. The kindly add:
Stripe subscription system ensures smooth "As shown in Figure
and hassle-free billing. 82.1."
The user engages with the AI Genius
web application which is built using Next.
js and React.js. They provide input prompts
through the user-friendly interface. Next.js
manages and handles user requests, serving
Figure 82.2. Figure 82.2 Captain: Workflow
as the front-end framework. Communication
of AI-driven Content
with the backend server is established, which Source:
Generation in AI Genius
Application
AUTHOR'S SELF GENERATED
584 Innovations in Computational Intelligence: Trends and Technologies
communicates with Stripe to handle pay- 2.3. Making API call
ment transactions and subscription man-
The initiation of a programmatic interface
agement tasks. Stripe ensures a smooth
communication, where the application inter-
user experience during the upgrade pro-
faces with the chosen tool’s API through a
cess, handling the financial aspects securely
well-defined protocol.
and efficiently.
Kindly Add following
2.1. Prompt text here: 2.4. Request
"The detailed This phase involves the formulation and
The inception point is where user input, often
workflow of content transmission of structured instructions or
in the form of a command generation
or directive,ininiti-
AI parameters to the selected tool’s API, delin-
ates the workflow. Genius is depicted ineating the task objectives.
Figure 82.2,
2.2. Select any tool demonstrating the
process from user 2.5. Response
The decision node is where inputusers exercise the
to delivering The ensuing phase entails the generation of a
agency in choosing from an final output."tools, response by the tool’s API, manifesting out-
array of
each catering to distinct functionalities. tools
puts in adherence to the provided instructions.
within “select any tool”:
2.2.1. Image generation 2.6. Gathering response
An intermediate step encompassing the aggre-
Deploys an algorithmically driven image gen-
gation and processing of the API response,
eration tool, leveraging deep neural networks
involving data extraction and preliminary
or similar models to produce visually appeal-
manipulation.
ing and contextually relevant images.
2.2.2. Video creation 2.7. Desired output
The culminating stage is where the system
Engages a tool specialized in video process-
presents the end-user with the refined out-
ing and manipulation, utilizing codecs and
put or result derived from the selected tool.
algorithms to edit or generate video content
This could manifest as an image, video, music
as per user specifications.
composition, or a problem-solving response
from a chatbot.
2.2.3. Music composition This intricate workflow encapsulates user-
Leverages a music composition tool employ- driven interactions with diverse tools through
ing advanced algorithms, potentially based API calls, ensuring precise articulation of tasks
on machine learning, to compose musical and delivering nuanced outputs in alignment
pieces tailored to user preferences. with advanced technological paradigms.
2.2.4. Chatbot for problem-solving 3. Thematic Overview
Employs a sophisticated natural language AI Genius is a website providing easy and fast
processing-driven chatbot, incorporating artificial intelligence access to all users over
advanced dialogue systems and problem- the web. The use of React.js and Node.js for
solving algorithms to address user queries making the user interface (UI) understand-
intelligently. able and user-friendly. Until now there is no
AI genius: Integrated intelligence platform 585
website or application which provides such a of writing both front-end and back-end code
platform having different AI tools at a single in one location.
level. The thematic intersection of the research
Earlier, there was research on many AI works highlights the symbiotic relation-
tools and technology AI-based affective ship between AI and web development [17],
music generation (AI-AMG) systems [3], AI showcasing the transformative potential of
Image Producer [4], application of ChatGPT advanced technologies in shaping diverse
in addressing programming bugs[8], and One- fields, from programming to image and video
Shot Video Tuning [7]. The evolution of arti- generation [12]. The studies collectively
ficial intelligence as a service [16] should be emphasize the need for continuous explora-
provided to anyone who wants to use artificial tion, innovation, and interdisciplinary collab-
intelligence. AI is recognized as a transforma- oration to unlock the full potential of these
tive technology with significant potential to technologies for societal benefit.
enhance various sectors [16], so the focus is The literature tells about the continuous
to integrate AI capabilities [16] in a simple evolution of web development technologies,
way over the internet. Researchers delve into the transformative impact of AI across vari-
the application of AI in web development, ous domains, and the importance of secure
covering areas such as image generation, data-sharing models. There is a consensus
affective music generation, and AI-powered on the need for collaborative approaches to
programming bug-solving. Specific AI tools unlock the full potential of web technologies
and models like DALL-E [16], GitHub Copi- as well as AI technologies and address chal-
lot [5], and ChatGPT are explored for their lenges in the digital landscape.
contributions to creative processes, program-
ming assistance, and user interaction.
The researchers explore the dynamic land- 4. Critical Analysis
scape of web development technologies, with
a particular focus on frontend frameworks 4.1. Strength
like Next.js. Next.js is highlighted for its flex- The literature covers a diverse range of top-
ibility in building reusable UI [2] components ics, including web development, AI integra-
and its advantages in server-side rendering tion, secure data-sharing, and AI impact on
for improved search engine optimization visual arts and content generation.
(SEO). The studies acknowledge the evolu- Many studies focus on practical applica-
tion of web technologies and emphasize the tions, such as Next.js in web development,
need for compatible and efficient backend GitHub Copilot in programming, and AI
solutions. The research emphasizes Next.js’s models like DALL-E in content generation.
role in server-side rendering, addressing chal-
lenges related to the speed of page loading 4.2. Weakness
and enhancing SEO. Next.js simplifies the
development process by allowing developers The overview provided lacks specific details
to create both front-end and back-end code on the methodologies used in each study.
in one place. Next.js is specifically credited Methodological weaknesses or biases could
with solving the problem of slow page load- be present, affecting the reliability of the
ing through client-side rendering, and it aids findings.
in SEO [2] by facilitating better crawling by The ethical implications of AI, data-shar-
search engines it is a full-stack framework ing, and content generation are critical aspects
that requires compiling the entire code base that might not be thoroughly addressed in the
for every production build due to its structure overview.
586 Innovations in Computational Intelligence: Trends and Technologies
There appears to be a gap in addressing without compromising quality, empowering
the ethical considerations associated with AI users with limited budgets.
applications, especially in content generation Economic barriers hinder the widespread
and data-sharing. adoption of transformative AI applications.
While there’s recognition of AI’s impact AI Genius aims to break these barriers,
on professional productivity, the overview making AI accessible to users from diverse
doesn’t delve deeply into potential biases or backgrounds.
ethical concerns arising from AI models like Seamless integration of AI tools into exist-
GitHub Copilot. ing workflows is a challenge. AI Genius pri-
oritizes compatibility and easy integration,
4.3. Quality and reliability ensuring businesses can adopt AI technolo-
gies without disruption.
The reliability and quality of the studies may Personalization and user-friendly inter-
vary. It’s crucial to assess the rigor of meth- faces are vital for enhancing engagement with
odologies, the representativeness of samples, AI systems. AI genius focuses on tailoring AI
and the generalizability of findings in each experiences and creating intuitive interfaces
paper. for a positive user experience, fostering a
The reliability of the studies also depends more accessible AI environment.
on the reputation of the publication sources.
Peer-reviewed journals and conferences are
generally considered more reliable. 6. Recommendations for Future
Seamless integration of AI tools into exist- Work
ing workflows is a challenge. There is a lack
Adding an educational dimension to the sys-
of effort and research for finding out a more
tem, enhancing versatility and engagement.
user-friendly and integrated platform.
This feature could broaden the utility of the
AI Tool Hub, providing users with an interac-
5. Synthesis and Implications tive and educational experience.
Continuously updating integrated AI tools
To democratize AI technologies by tackling
to incorporate the latest advancements in
issues related to complexity, accessibility, and
machine learning and artificial intelligence.
affordability. Simplifying intricate algorithms
Ensures that users have access to cutting-edge
and interfaces to make AI more user-friendly
creative capabilities, keeping the platform rel-
and accessible to a broader demographic.
evant in a rapidly evolving AI landscape.
Current AI platforms often present com-
Enabling multiple users to work together in
plexity barriers, limiting access to a broader
real-time, share projects seamlessly, and foster
demographic. AI Genius recognizes the need
a sense of creative community. Enhances the
to simplify AI tools, making them more
platform’s collaborative capabilities, making
approachable for individuals and businesses
it more conducive for teamwork and collec-
without specialized technical expertise.
tive creative endeavors.
The fragmented nature of existing AI solu-
Implementing advanced analytics tools
tions hampers creative exploration. AI Genius
to gain valuable insights into user behavior
offers an integrated service that allows users
and preferences. Informs ongoing platform
to seamlessly combine various AI function-
improvements and enables greater customi-
alities without the need to switch between
zation of the user experience based on data-
platforms.
driven insights.
Affordability is a critical concern, espe-
Developing dedicated mobile applica-
cially for startups and small businesses. AI
tions for both iOS and Android platforms.
Genius provides cost-effective AI access
AI genius: Integrated intelligence platform 587
Enhances accessibility and usability for users in the evolving landscape of AI-powered cre-
on the go, expanding the reach of the AI ativity. The vision extends beyond a static
Genius to a broader audience. solution, emphasizing adaptability and acces-
Exploring partnerships and integrations sibility for users.
with popular creative platforms, content AQ3:
management systems, and social media net- Refer-
works. Allows users to directly incorporate References ences are
missing
AI genius-generated content into their exist- [1] Smith J, Johnson A, Davis M, and White S. cita-
ing digital workflows and online presence, AI Genius: Unifying creativity and innova- tions in
the text.
fostering seamless integration with estab- tion. AI Genius Research Report. 2023.
Please
lished platforms. [2] Bui D and Mynttinen T. Next.js for front- ensure
end and compatible backend solutions. that all
XAMK University. 2023. references
are cited
7. Conclusion [3] Dash A and Agres KR. AI-based affective and in
music generation systems: A review of meth- sequential
Envisaged as a centralized platform to address ods and challenges. National University of order.
the challenges posed by fragmented AI tools. Singapore. 2023.
Aims to integrate diverse AI tools, including [4] Sharma H and Jain K. An AI image genera-
image generators, video creators, music com- tor using Open AI and NODE.JS. Ind Eng J.
position engines, and poster creation tools. 2023.
Establishes robust APIs to facilitate seam- [5] Peng S, Kalliamvakou E, and Cihon P. The
less communication between integrated tools. impact of AI on developer productivity:
Implements a single sign-on mechanism to Evidence from GitHub Copilot. Microsoft
enhance both security and user convenience. Research. 2023.
Demonstrates a commitment to ongoing [6] Zhu Y, Baca J, Rekabdar B, and Rawassiza-
improvement and expansion. Acknowledges deh R. A survey of AI music generation tools
and models. Boston University. 2023.
the dynamic nature of the creative landscape
[7] Wu JZ. One-shot video tuning for text-to-
and the rapid evolution of AI technology. video generation. Comput Video Gener Res.
Pledges to remain at the forefront of innova- 2023.
tion by incorporating the latest advancements [8] Fariz MFS and Lazuardy DA. Modern front
in AI. end web architectures with React.js and
Positions the AI genius as a dynamic eco- Next.js. Int Res J Adv Eng Sci. 2022.
system, not a static solution. Aims to empower [9] Ding M and Zheng W. CogVideo: Large-
individuals and businesses to fully leverage AI scale pretraining for text-to-video genera-
for creative expression. Extends accessibility tion via transformers. Tsinghua University.
through the development of dedicated mobile 2022.
applications. [10] Merino-Gomes E. Text to image generation:
Envisions a future where AI-driven crea- Leaving no language behind. Univ Antonio
tivity is accessible and empowering for a de Nebrija. 2022.
[11] George AS and George AH. Data sharing
broad user base. Portrays the AI genius as an
made easy by technology trends. Partners
evolving solution that adapts to the changing Universal Int Res J. 2022.
landscape of AI and creative technologies. [12] Reviriego P. Language diversity in text-to-
The AI genius aims to solve the challenges image generators: Current state and future
associated with fragmented AI tools by pro- implications. Artif Intell Res. 2022.
viding a centralized, integrated platform. It [13] Deng J. Utilizing Chat GPT for program-
not only addresses current issues but also ming bug solutions. Program Bug Res. 2022.
demonstrates a commitment to continuous [14] Zhou Y. Training text-to-image generation
improvement, innovation, and empowerment models without text data. Cross Modal Sem
Space Res. 2022.
588 Innovations in Computational Intelligence: Trends and Technologies
[15] Morrical N, Tremblay J, Lin Y, and Tyree S. recent advances and their integration in
NViSII: A scriptable tool for photorealistic smart industries. Partners Universal Int Res
image generation. Synth Data Gener Work- J. 2020.
shop (SDG) at ICLR. 2021. [19] Zulić H. How AI can change/improve/influ-
[16] Lins S, Pandl KD, Teigeler H, and Thiebes ence music composition, performance, and
S. Artificial intelligence as a service. Institute education: Three case studies. INSAM J
AIFB, Research Group Critical Information Contemp Music Art Technol. 2019.
Infrastructures, Karlsruhe Institute of Tech- [20] Avdeef M. The audio uncanny valley:
nology. 2021. Exploring AI in popular music production
[17] Ballamudi VKR. Increasing production through the case study of SKGGE’s ‘Hello
build efficiency in Next.js applications. Web World’ album. Partners Universal Int Res J.
Dev. 2021. 2019.
[18] Devagiri JS. Augmented reality and arti-
ficial intelligence: A systematic review of
83 Agriculture: Soil analysis for suitable
crop and fertilizer prediction
Shreya Singha, Saurabh Kumar Jainb, and Stuti Rastogic
Department of Computer Science and Engineering, Babu Banarasi Das Institute of Technology and
Management Lucknow, India
Abstract: This review paper considers new developments in agriculture with respect to utilizing ML and DL
models for CFRS and PDD. The combination of ML and DL in agriculture has led to its metamorphosis. The AQ1:
paper explores the basic foundations, approaches, and applications of CFRS and PDD in agriculture. The paper Please
entails various aspects of the development of the CFRS, and it explains how ML and DL have assisted it in provide
NOTE ADDED
becoming more accurate and effective in offering personalized advice on crop choice as well as best fertilization the
practices. It also delves into modern PDD that applies machine learning and deep learning solutions for timely expan-
sion for
and accurate detection of plant disease thereby preventing loss on crops and maximizing yield. This article ML, DL,
examines different popular ML and DL models and tools for applying them in CFRS and NPD. The study CFRS,
examines the use of various systems in case studies and real-world data sets, clarifying their practical implica- PDD,
tions and problem areas. It discusses how IoT and remote sensing can be used together to collect agricultural PPP and
data bearing on quality and quantity. Finally, this review paper presents the transformative power of ML and DPDD.
DL for agriculture where PPP and DPDD models help farmers to make the right choices and avoid harvest
losses. This gives room for the identification of upcoming research areas and areas that need a review so as to
help achieve the sustained growth of the agriculture industry as a whole.
Keywords: Crop yield prediction, machine learning, fertilizer prediction, plant disease detection
1. Introduction should be planted on particular areas. In these
concerns, there is the presentation of a desk-
According to the census done in 2011, there top application where that applies various
are 118.6 million farmers who depend on data mining technologies so as to determine
agriculture to earn their living (agriculture). the optimal production of the most profitable
These include among others understanding plant species depending upon the prevailing
of the soil conditions, application of manure, climate charts and type of soils of the region.
management, and control of quality of crops The key aim of the reviewed paper is to
and difference between zones of the same provide a comprehensive and up-to-date
zone from same farm. Such decisions tend to analysis of the integration of machine intelli-
become overwhelming considering numer- gence and deep learning in crop yield predic-
ous variables and measures to be taken into tion with respect to developing more accurate AQ2:
NOTE ADDED
Please
account. This makes a farming strategy of and effective crop yield forecasting models. It provide
monitoring farmlands towards enhancing is envisaged that by exploiting the capabili- the
expansion
farmer’s productivity the focal point of this ties of MI and DL, more resilient and adap- for MI
paper. With access to online weather records tive models would be formulated that could and DL
like rainfalls and soil limits in different loca- accurately predict crop yields under divergent
tions, one is able to determine the plants that environmental and management conditions
[email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-83
590 Innovations in Computational Intelligence: Trends and Technologies
with a reasonable degree of accuracy. This crop and nutrient values, which were trained
would help farmers and other agricultural utilizing support vector machines (SVM)
stakeholders in effective decision-making algorithm to develop our system. We use
with regard to resource supply, irrigation the data provided by users and match them
management, pest control, and crop insur- against our trained dataset to give appro-
ance—thereby enhancing agricultural pro- priate and credible suggestions. The paper
ductivity and sustainability. Another slice of provides valuable findings, while introduces
the big development underway pertains to the concept of majority voting on a group
soil. One of the everyday problems that these of decision-making algorithms. We examine
farmers face is unawareness about the selec- how these methodologies contribute to data-
tion of the right crops and their proper plan- driven crop recommendations. For improved
tation. Our project works on the analysis of user experiences, crop recommendation can
major parameters like sodium, potassium, benefit from incorporating location data
phosphorus, pH value of the soil, rainfall, and using GPS technology. The versatility of this
locality. Considering all these, one can easily approach is explored, highlighting its poten-
determine the soil type for various kinds of tial impact on modern agriculture.
crops. This review paper seeks to contribute to
This paper examines different approaches the ongoing discourse on crop prediction by
of crop recommendation as outlined in examining innovative solutions. With this
another study to help us attain our objec- approach to soil investigation, crop recom-
tives. We delve into the use of support mendations and location basis information,
vectors, artificial neuron networks, ran- we anticipate giving relevant contributions
dom forest among many more suitable to agriculture best practices. In essence, this
approaches used to come up with the best project was set out to help our clients make
crop recommendation methods. This paper better decision which will in turn translate
is equally useful since it has given results in into higher output, efficiency and durability
addition to introducing the idea of major- in our sector.
ity voting for decision-making. We analyze
how these approaches help in enhancing the
user experience of data-driven crop advice. 2. Methodology
The addition of location data to improve
crop recommendation. In addition, such 2.1. Formulating a research question
processes as a combination of GPS tech- We started by formulating a clear and well-
nology with location-based data will allow defined questions to guide our review, as:
crop recommendations more efficient. This
information allows us to forecast harvesting • What are the recent advancements in crop
outcomes, and advice on appropriate crops yield prediction using machine learning
considering land and climate specifics. This and deep learning techniques?
approach has a variety, and it explores some • How do these innovations impact the
of its applications in modern agriculture. accuracy and efficiency of crop yield
The objective of this review paper is to par- forecasts?
ticipate in the current debate on crop fore- • What is the role of emerging technologies,
casting, focusing on novel approaches. Our such as AI, machine learning, and IoT, in
system is especially helpful for farmers with- enhancing “crop yield prediction?”
out an understanding of their soil’s fertility • Assess the potential of MI and DL to inte-
levels and best crops under their circum- grate with IoT and other sensor data sources
stances. We used widely available datasets of for comprehensive crop yield prediction.
Agriculture: Soil analysis for suitable crop and fertilizer prediction 591
These questions set the stage for a compre- from English, without any peer review, or
hensive and systematic review of the relevant about their direct relation to our topic core
literature. Maintaining the integrity of the did not provide a good understanding.
specifications.
2.4. Machine learning and data-
2.2. Search criteria, databases, and driven crop yield prediction
sources Employment and application of machine
Our search criteria included keywords and learning and data-driven ways of predicting
phrases related to the topic of review. These crops yield can be seen as a big step toward
criteria were carefully chosen to cover vari- sustainable development of agriculture.
ous aspects of the subject, ensuring a compre- Those are important for decision-making in
hensive search. We used Boolean operators the agricultural sector. Proper crop yield esti-
(e.g., AND, OR) to combine search terms mation is actually of utmost importance that
effectively and maximize the retrieval of rel- is because the agricultural industry is shift-
evant studies. ing its reliance on data-driving technology in
We conducted searches in several reputable order to be able to optimize farm operations,
databases, including Web of Science Google make efficient use of resources, and in the
Scholar, Springer Link, IEEE Xplore, Science end, have a higher production yield.
Direct, and ACM Digital Library. Each data-
base was chosen for its relevance to the field
and its coverage of academic literature. This 3. Thematic Overview
multi-database approach helped us capture a
broad spectrum of sources. 3.1. Support vector machine (SVM)
The main concentration was peer-reviewed SVM is a powerful machine-learning tech-
journal articles, conference reports, and schol- nique that effectively addresses the challenges
arly books. These references were judged as of classification and regression. Instead of
the ones having the highest degree of credibil- dealing directly with the initial feature space,
ity and the most reliable for our review. the SVM shows the data items in an N-dimen-
sional space and establishes a hyperplane that
2.3. Search inclusion and exclusion would separate the classes more clearly in
the Linear Kernel. Since SVM takes the bur-
criteria for selecting studies
den of specifying the optimal hyperplane, it
Now, to come up with precise and recent data, also demonstrates its talent in being able to
we based our evaluation on studies that were showcase the training data points, along with
made just before 2020. For that reason, they the differentiation of the classes, in space. A
also have to be in English since this way we compelling study has come out with the SVM
can ensure the review is done with the same application in crop categorizing using macro
accuracy of study. Theorists permitted to us, and micro-nutrient status, the study implies
which in an upright way examine all of the the principles that underpin this technology.
assistive technologies for web access that have
made Advancements directly. More than that, 3.2. Random forest
the research work had to get verified by the
test group and availability in the full text was Random Forest, a machine learning system
a must for in-depth analysis. We excluded the that uses an ensemble of decision trees, is
references that were not legitimate to the cur- able to mature through a number of rounds
rent time e.g. which were published in years to build a solid model. This method is of
before 2020, written a different language exceptional usefulness in the situation when
592 Innovations in Computational Intelligence: Trends and Technologies
farmers are to be supported in the assessment (IoT). A project that could be proposed in
of soil quality and crop growing suitability. the future, researchers consider the develop-
When Random Forest is included in Crop ment of an Agricultural Information System
Recommendation Systems and Fertilizer Rec- for improvement in Indian agriculture based
ommendation Systems, the performance of on real-time rural information by the use of
those systems is mostly ramped. IoT. Through this hi-tech device, farmers are
able to facilitate informed decisions and act
3.3. Machine learning-based crop accordingly as well as minimize the negative
impacts of climate change on their agricul-
recommendation
tural activities due to receiving the necessary
Crop recommendation systems, this is where data from the soil monitoring system. The
machine learning comes into play, which is idea of the creators of the system is that the
the idea of using powerful data analysis to productivity of soil has a direct relation with
farmers for the first time. They also use infor- crop production.
mation such as historical crop yields, soil
characteristics, weather patterns, and market 3.6. XGBoost
trends. One thought-provoking pitch is the
implementation which targets the establish- One of the most widely applied supervised
ment of a system tailored to agricultural and machine learning methods is XGBoost in the
horticultural crops. form of a gradient-boosting implementation.
A study in this respect will explore the appli-
cability and enhancement of an XGBoost
3.4. Soil analysis and fertility
algorithm through the optimization of its
prediction parameters to gain better learning accuracy.
Soil health is noted as the top priority for the Much emphasis in this paper is on the aspect
optimal performance of the crops. Crop yield of feature importance for the sake of optimiz-
prediction is done usually by using soil sam- ing the performance of this model.
ples from the analyzed area, and usually, the
plant is destroyed. Agricultural technology 3.7. Gaussian Naive Bayes
provides real-time information on the condi-
tions of the soil through the management of The Gaussian Naive Bayes approach assumes
several processes such as information collec- the features to be independent. It is the sim-
tion and monitoring. The use of technologies, plest and easiest approach in machine learn-
particularly artificial intelligence and machine ing. Finally, Gaussian Naive Bayes can aid
learning, is very popular in the process of soil in enhancing smart agriculture farming sys-
analysis. Through Artificial Intelligence and tems. Crop and fertilizer recommendations
Machine Learning enabled precision agricul- can be enhanced. A mixture of Gaussian
ture we are now able to detect crop growth Naive Bayes classifiers with local independ-
stages, plant stress, and pests. ent features combined with semi-AdaBoos is
another non-parametric approach that con-
siders performance improvement in terms of
3.5. Soil analysis for crop prediction local features.
Soil analysis is the most important activity
that is performed for the guidance of a farmer
on crop productivity. Soil monitoring and rec- 4. Critical Analysis
ommendation systems that can be designed A. Wenjie Yang et al., 2023: Under fallow
are the source where Nutrient measurement season precipitation, optimum fertiliza-
data are derived from the Internet of Things tion decreased environmental risks but
Agriculture: Soil analysis for suitable crop and fertilizer prediction 593
increased economic benefits for dryland the assessment of the reliability of such
wheat-based Nutrient Expert.[1] recommendations.
Strength: D. Sachin Kapoor, et. al (2022) Smart Agri-
Practical Application: The following culture Farming Using Harvestify Web
paper discusses optimized fertilization App. [4]
for dryland wheat. It presents an oppor- Strength:
tunity for practical application and pro- Innovative Application: The present paper
vides experience in agriculture, based on provides an application of the web appli-
fallow-season precipitation, as well as by cation based on smart agriculture, Har-
using the Nutrient Expert system. vestify, hence opening an innovative way
Weaknesses: toward modernization in farming.
Generalization: The findings of this study Weakness:
are only of confirmatory values to some Limited Detail: This might lack the techni-
conditions of dryland wheat farming, cal details of Harvestify, its functionalities,
and they can never be extrapolated to all or the technologies inherent within them.
circumstances. More details of a technical nature can be
B. Biplob Dey, et. al. (2023) Machine Learn- added to increase the worth of the paper.
ing Based Recommendation of Agricul- E. Nishit Jain, et. al. (2022) Farmer’s Assis-
tural and Horticultural Crop Farming in tant: A Machine Learning Based Applica-
India Under the Regime of Npk, Soil Ph tion for Agricultural Solutions. [5]
and Three Climatic Variables.[2] Strength:
Strength: Focus on a solution: Agriculture The paper
Innovative Approach: The paper employs treats a solution to a practical problem in
machine learning for crop recommenda- agriculture and proves that, yes, it is quite
tions, showcasing an innovative approach possible that machine learning can be one
to optimize agricultural and horticultural of the solutions for farmers’ problems.
practices. Weakness:
Weaknesses: Limited Information: In the event that
Data Quality and Quantity: The effective- information is not provided on the
ness of machine learning models heavily machine learning algorithms used and
relies on the quality and quantity of data. datasets employed, features of Farmer’s
The paper’s robustness may be affected if Assistant; in that case, that is going to
there are limitations in the dataset used be quite a bit of problem for readers to
C. Maaz Patel, et. al. (2023). Crop Recommen- make judgments like this application on
dation System using Machine Learning.[3] the basis of strength and effectiveness.
Strength: F. Sadia Afrin, et. al. (2018) Analysis of Soil
Relevance: The paper is very relevant, Properties and Climatic Data to Predict
dealing with a contemporary and impor- Crop Yields and Cluster Different Agri-
tant subject. Some type of crop recom- cultural Regions of Bangladesh [6]
mendation system, aided by Machine Strengths:
Learning, has a great relevance in the case Methodological Rigor: Given the associa-
of precision agriculture. tion with IEEE, there is an expectation of
Weakness: high standards in terms of research meth-
Methodological transparency: The devel- odology, analysis, and reporting.
opment and testing of a machine learning Weakness:
model have been done without details. Data Quality: The reliability of the pre-
To that respect, it significantly impairs dictions heavily depends on the quality
594 Innovations in Computational Intelligence: Trends and Technologies
of soil and climatic data. If the data used soil test analyses. This embraces explanations
lacks representativeness or accuracy, it of the effects of specific nutrient levels in soils
could undermine the robustness of the as well as recommendations for crop pro-
study. duction and fertilizer usage. Training farm-
G. S.Jeyalakshm, et. al. (2019) Data Mining ers about updating themselves with their soil
in Soil and Plant Nutrient Management, status data as well as identifying any abnor-
Recent Advances and Future Challenges mality can contribute towards sustainable
in Organic Crops [7] farming practices.
Strength:
Timely Focus: The paper’s focus on data 5.2. Implications
mining in the context of soil and plant
nutrient management aligns with the 5.2.1. Monitoring tools for farmers:
growing interest in precision agriculture Like accessible tools for parents in the case
and data-driven farming practices. of cyberbullying, some straightforward moni-
Weakness: toring tools should be available to farmers.
Lack of Specifics: The title provides These can be able to provide timely informa-
an overview, but without access to the tion on soil conditions, nutrient levels, and
detailed content, it’s challenging to evalu- suitability for various crops. User-friendly
ate the specific methodologies used, the applications can empower farmers with data
depth of the analysis, or the novelty of the that may help them make decisions about
findings. crop and fertilizer applications.
5. Synthesis and Implication 5.2.2. Backtracking and feedback
Agriculture: Soil analysis for suitable crop loop:
and fertilizer prediction The process of backtracking entails maintain-
ing a history of soil analysis results and the
5.1. Synthesis corresponding yield outcomes over a given
period of time. This feedback loop preserves
5.1.1. Real-time monitoring: historical data that enables farmers learn
When it comes to soil analysis is important from their previous choices and actions taken.
in order to determine whether the soil is suit- Moreover, soil condition changes can prompt
able for planting crops or not, and also helps alerts suggesting modifications in plant spe-
to predict what kind of fertilizers are needed. cies or fertilization programs.
In this manner, a farmer can be able to exam- To sum up, this synthesis stresses real-time
ine their soil condition in time and therefore monitoring, and education as touchstones of
make prompt decisions on which plant to use successful farmer activities within the domain
and type of fertilizer to apply. Consequently, of soil analysis. Implications are made con-
farmers often have real-time data that ena- cerning the need for farmer-centered simple
bles them to adjust strategies and practice monitoring tools and backward bending
agriculture effectively. curves through historical data leading to con-
stant progression in farming practices.
5.1.2. Farmers’ education:
This also makes it possible for farmers to 6. Conclusion
understand the results obtained from machine We are painting a bright image of changing
learning as applied in the interpretation of agricultural technology in our review article.
Agriculture: Soil analysis for suitable crop and fertilizer prediction 595
The incorporation of high-tech methods and are more accurate and precise in search of
techniques as reviewed here is more about precision farming.
moving away from conventional farming to The role of IoT and remote sensing
precision agriculture that will ensure sustain- integration:
able development. Real-time data collection definitely has
An example of an optimization and fer- huge potential to influence future research on
tilizer-based nutrient system suggested by this subject matter. In other words, it would
one research paper reveals a smart blending be fitting for future researchers to focus more
of traditional agricultural approaches with on how IoT devices could be made to interact
technology which can help decrease the risks with remote sensors in enhancing agricultural
of environmental pollution while increasing systems. Such integration has wide poten-
their economic benefits in dryland, and win- tial to enhance accuracy in data entry and
ter wheat production. This new approach is unleash related valuable information in pre-
driven by emerging agrarian needs and the cision farming. Integration of IoT and remote
place for technology towards achieving bal- sensing enables better data-driven decisions
ance between food productivity and guarding and therefore increases farming efficiency.
environment against degradation. A comparison between machine learning
The future of farming seen through intel- algorithms is also included:
ligent agricultural solutions entails not only Future studies that will look holistically
efficient processes enabled by technology into comparing different types of machine
but also real-time data for farmers to make learning algorithms that are employed in
effective decisions. Further emphasis on this predicting products must be undertaken to
transformative ability is evident through ensure the emergence of robust predictive
studies done on soil monitoring and recom- models. Such analysis would be very mean-
mendation systems, stressing the role played ingful in terms of finding the best suitable
by understanding soil health in adopting sus- algorithm for particular data sets. Knowledge
tainable customized crop plans. about the strengths and the shortcomings of
Moving forward, principles related to different algorithms helps to increase the pre-
data-driven decision-making and collabora- dictive efficiency for the specific purpose of
tion among diverse disciplines need to guide agricultural systems.
us. Agriculture entered the modern age by Validity of scalability and generalization:
ensuring food security, sustainable farming Therefore, for technological development
practices, and better livelihoods through the to proceed, researchers must assess the scal-
power of artificial intelligence. This review ability and generalizability of these systems in
paper will present a blueprint for creating a various locations, types of crops, and agricul-
more resilient and more productive agricul- tural operations. Performing thorough assess-
tural sector for the researcher, policymakers, ments will allow us to make sure that our
and agricultural stakeholders. solutions are universally applicable. This will
enhance the implementation of these technol-
ogies as well as their spread worldwide and
7. Recommendations for Future their advantages with ease across all nations.
Research
We want other areas of research in web acces- References
sibility and assistive technology to grow fur-
ther in their progress. Proposals are to make [1] Yang W, Lie Y, Jia B, Lu L. An optimiza-
tion and fertilizer-based nutrient system
existing agricultural systems better so that
for dry land, winter wheat production
data collected and analyzed or data predicted reduces environmental risks and increases
596 Innovations in Computational Intelligence: Trends and Technologies
economic benefits. Available at: https:// for agricultural solutions. Available at:
www.sciencedirect.com/science/article/ https://www.academia.edu/80649608/
pii/S0378377423004948?via%3Dihub. Farmers_Assistant_A_Machine_Learn-
Accessed October 8, 2024. ing_Based_Application_for_Agricultural_
[2] Dey B, Ahmed R. Machine learning-based Solutions?f_ri=2008. Accessed October 8,
recommendation of agricultural and hor- 2024.
ticultural crop farming in India under the [6] Saurabh Salvi, Adut Chaudhari, Pranit Shelkr,
regime of NPK, soil pH, and three climatic Namtra Ansari.(2021)”Soil Monitoring and
variables. Available at: https://papers.ssrn. Recommendation System.”ICAST. https://
com/sol3/papers.cfm?abstract_id=4437863. papers.ssrn.com/sol3/papers.cfm?abstract_
Accessed October 8, 2024. id= 3868690
[3] Patel M, Rane A, Patni V. Crop recommen- [7] Sadia A, Talha K, Mahia M, Wasit A, Rashe-
dation system using ML. Devang Patel Insti- dur MR. Analysis of soil properties and cli-
tute of Advanced Technology and Research, matic data to predict crop yields and cluster
Charhot University of Science and Technol- different agricultural regions of Bangladesh.
ogy. Available at: https://www.researchgate. IEEE ICIS. Available at: https://www.seman-
net/publication/370056714_Crop_Recom- ticscholar.org/paper/Analysis-of-Soil-Prop-
mendation_System. Accessed October 8, erties-and-Climatic-Data-to-of-Raka-Khan
2024. /4c996df02f238fb553022af398c1205dc86
[4] Kapoor S, Aggarwal I, Rav AK. Smart agricul- 15f53. Accessed October 8, 2024
ture farming using Harvestify app. Department [8] Jeyalakshmi S, Rana V, Suseendra G. Data
of Computer Science and Engineering, Sharda mining in soil plant nutrient management.
University. Available at: https://papers.ssrn. IJRTE. 2019. Available at: https://www.ijrte.
com/sol3/papers.cfm?abstract_id=4157630. org/wp-content/uploads/papers/v8i2S11/
Accessed October 8, 2024. B10350982S1119.pdf. Accessed October 8,
[5] Jain N, et al. Farmer’s Assistant: A 2024
machine learning-based application
84 Comprehensive examination of
system surveillance method employing
keystrokes logging technology
Bajrangi Vishkarmaa, Kamlesh Kumarb, Aman Vermac, Atul Singhd,
and Abhishek Singhe
Computer Science and Engineering, Babu Banarasi Das Institute of Technology and Management,
Lucknow, India
Abstract: In many company, data protection and recovery is immediately ultimate main determinant. In many
cases place data improvement is necessary. For these types of questions, a keylogger is individual of the best
resolutions, that is frequently refer to as keylogging or row of keys capture. Key capping is the act of record-
ing keystrokes on a row of keys, commonly clandestine because the man using the row of keys does mix up
that their conduct are being listened. With the help of keylogger request, users can receive the dossier when the
occupied file is debased due to various reasons like capacity misfortune etc. This is a pursuing use used to track
the consumers that records the keystrokes; uses record files to repair news. Using this application, we can recall
a erased electronic mail or URL. In this keylogger project, at whatever time the consumer types entity on the
keyboard, the keystrokes are picked up and shipped to the admin electronic mail address outside the consumer’s
knowledge inside a set period. Keywords—Company, Dota protection, recovery, keylogger, resolutions, keylog-
ging, recording keystrokes, occupied file, capacity misfortune, pursuing use, track consumers, record files.
Keywords: company, Data protextion, recovery, keylogger, resolution, keylogging, recording keystrokes, oc-
cupied file, capacity, misfortune, pursuing use, track consumers, record files
1. Introduction power, etc. Keyloggers are specifically profit-
able in observing advance misconducts.
In numerous IT endowment organizations This is frequently an observation request
immediately a days, facts security and facts employed to track the customers that log
recovery are the foremost alive variables that keystrokes, engage record records to restore
are basically sent in Computer Forensics. data, and capture a record of all composed
Computer forensics forms the skill of analyz- answers. The collected dossier is forgiven on
ing electronic publishing to protect, recovery the foundation as a hide record or to fore-
and analyze the facts in a successful way. shadow to the admin or the determinable
There are abundant cases where news accountant.
Recovery is required. Keylogging is one of Keyloggers are a in a way malware (we
the first-rank, famous spying calculating pro- are illustrating for moral purposes as vari-
grams in calculating history, so by utilizing ous educators are handling the unchanging
keylogger uses, clients can restore news within for foundation observant) that with deter-
the moment of truth of calamity and injuring mination path customer assurance from
of working records on account of deficit of the comfort inside the endeavor to heal
a
[email protected]; [email protected]; [email protected]; dasa.atul8127@gmail.
com; [email protected]
DOI: 10.1201/9781003606635-84
598 Innovations in Computational Intelligence: Trends and Technologies
individual and private facts. Key lumberjacks projected up until now to overcome the ques-
can electronic mail or ftp the report holding tion and also the troubles of the projected
keystrokes record, back to the surveillance whole. Key record is a stock business prac-
woman. These keyloggers work cautiously tice that endure be attainable from diversi-
out of point to capture the customer exercise fied points of view. When the aggressor gains
on the support, so all of the keystrokes are set tangible approach to your estimating ploys,
missing in a well-hide document. they may eavesdrop on physical hardware
We are making the program keylogger, such as the keyboard to collect valuable user
skilled are many sorts of keyloggers but we data. This strategy is completely dependent
are focused on the program. on some real properties, either sound trans-
mission created when the client composes or
electromagnetic remote console propagation
2. Problem Definition (Santripti Bhujel | Mrs. N. Priya 2022).
Keyloggers are, for the most part, created In [2], we found out that external keylog-
only for a particular performance of logging gers or hardware keyloggers are small elec-
and do not cause damage by polluting the tronic devices that are placed between the
plans the way viruses do. Alternatively, Key- keyboard and motherboards. This procedure
logging programs monitor the keystrokes of requires attackers to have physical access to
the victim, so that underhandedly they snag the system to compromise. Keyloggers are
all information by rounding up the endeav- performed on targeted machines for record-
ors acted on a calculating. Keyloggers capture ing client keystrokes finally, the transfer
the strokes fashioned by answers and sustain of this private information to a third party
aforementioned information in hidden record (kavya. C, Suganya.R 2021).
files, that is before shipped to the admin. In the In [3], we observed that keyloggers are used
process, Keyloggers wind up leaving behind a for both legal and illegal purposes. Attackers
limited footprint in terms of thought and ven- usually use keyloggers to obtain private infor-
dor exercise. Most of the ruling class cannot mation of an individual or an association.
be visualized in the ‘Task Official’ nor be seen In the past, a lot of credit card information
among processes. It is frequently disputing to has been misused by attackers using keylo-
equate the record files and the OS files, even ggers. Keyloggers therefore are individual of
subsequently slanting the whole guide. Many ultimate hazardous types of spyware to date
keyloggers primarily devote effort to hiding (Devashree Kataria, Manan Kalpesh Shah, S
keystrokes, but accompanying the exercise Bharath Raj, Priya 2020).
of rounding up screenshots and mouse clicks In the study conducted by Kataria et al.
can further increase the district of following (2020) [3], a discerning examination into the
of the system. multifaceted utility of keyloggers reveals their
pervasive presence in both lawful and illicit
contexts.
3. Literature Review Echoing these findings, attackers exploit
In [1], we observe that to make recognizing keyloggers as insidious tools to surrepti-
keyloggers more conceivable, it is main for tiously acquire private information, whether
things to take a handle in detail news about pertaining to individuals or organizations.
what keyloggers genuinely are, in what way Notably, historical incidents document the
or manner they work, implement, and com- widespread misuse of credit card information
prehend a various approach to it. On answers orchestrated by attackers leveraging keylog-
to this kind of questions, we will debate the gers. This underlines keyloggers as among the
various types of algorithms that have existed most perilous forms of spyware to date, as
Comprehensive examination of system surveillance method employing keystrokes logging technology 599
highlighted by Kataria et al. Their research commit remove the proneness, even though
positions keyloggers at the pinnacle of cyber it would possibly improve security in vari-
threats, emphasizing the critical need for pro- ous ways. He also proposed an extension of
active defensive measures against these highly the permissible lengths authentication codes
hazardous tools. could put the attack to sleep, but they could
In [4], we discover that malicious programs not change the straightforward situation. In
using the keystroke logging feature an exam- total, the central issue is the enemy of keylog-
ple of a real-time online banking system. If ging systems done in that particular way will
some of the frame functions were erroneously reasonably invalidate their entire considera-
updated, it may allow an attacker to gain tion (Jia Wang, Brent Lagesse 2020).
access to the client ledger. The escape of these In [6] the study by Johnson et al. (2018)
attacks maybe surely detached if the gadget [4], an exploration into the multifaceted use
keeps querying for a new set of personalities of keyloggers underscores their dual role in
or alphabets although either the login is prof- both legal and illicit domains. The research
itable (Disha H. Parekh, Nehal Adhvaryu, elucidates that, akin to the findings of Kataria
Vishal Dahiya 2020). et al. (2020) [3], attackers deploy keyloggers
In the investigation carried out by Parekh as potent tools to illicitly acquire private
et al. (2020) [4], a detailed exploration into information from individuals and organi-
the realm of malicious programs utilizing key- zations. Instances of unauthorized access
stroke logging unveils a concerning scenario, leading to the misuse of sensitive data, par-
particularly in the context of real-time online ticularly credit card information, have been
banking systems. The researchers highlight a documented throughout the history of cyber
significant vulnerability: the potential com- threats.
promise of a client’s ledger if certain frame The researchers highlight keyloggers
functions undergo erroneous updates. This as a persistent and formidable category
revelation emphasizes the critical interplay of spyware, emphasizing their continuing
between system functions and security, where relevance and potential harm in the con-
inadvertent updates may inadvertently grant temporary cybersecurity landscape. Such
unauthorized access to sensitive financial revelations accentuate the critical need for
information. robust protective measures against these
The authors underscore a potential mitiga- insidious tools.
tion strategy in the form of frequent querying
for new sets of characters or alphabets, even
after a successful login. This constant renewal 4. Assumptions and Dependencies
of security measures may effectively thwart Assumption: The following assumption was
the persistence of attacks. Parekh et al.’s taken into consideration:
research sheds light on the nuanced dynam-
ics of keystroke logging threats within online • The user has a reliable internet connec-
banking systems, urging for proactive meas- tion to upload the keystroke and screen-
ures to fortify against potential breaches and shot data to the cloud.
ensure the robust security of sensitive finan- • The cloud storage service is secure and
cial data. can protect the data from unauthorized
In [5], we discovered that as test depends access.
on individual positions and not possible the • The consumer has built an report on the
specific personality styles that are admitted cloud depository help and has given the
inside the proof rule, admitting codes to con- inevitable login attestations to the keylog-
tain a more different variety of personalities ger and screenshot modules.
600 Innovations in Computational Intelligence: Trends and Technologies
Dependencies: The dependencies are as user activities, and conducting regular secu-
follows: rity audits can help mitigate insider threats.
• The keylogger and screenshot modules 5.3. Data encryption
rely on the Python programming lan-
guage and specific libraries such as pyn- Protecting data through encryption is essen-
put. These dependencies must be installed tial. If hackers gain access to confidential
on the computer for the modules to func- information, encryption can make it difficult
tion properly. for them to make sense of the data even if
• The keylogger and screenshot modules they manage to extract it. Employing strong
must have access to system resources, encryption algorithms is crucial for safe-
such as the keyboard and screen, to cap- guarding sensitive information.
ture data.
5.4. Incident response plans
5. Gap Identification Organizations should have well-defined inci-
dent response plans in place to address cyber-
Hackers and different after second parties
attacks promptly. This includes processes
are forever expect exposures present inside
for identifying, containing, eradicating,
bureaucracy. To gain knowledge about what
recovering from, and learning from secu-
they demand from arrangings, they either
rity incidents. Regular testing and updating
attain to secret data stocked in bureaucracy
of these plans are essential to ensure their
and either cause harm to the purity of dos-
effectiveness.
sier or may cause dossier misfortune. Another
question is that cyber evil are growing con-
tinually. If we have the chat logs or keystroke 5.5. Endpoint security
logs of the fatality’s laptop, we can surely Given the proliferation of remote work,
resolve the complete preparation of the vic- securing endpoints such as laptops and other
tim, that will support high- quality answer to devices is critical. Endpoint protection tools,
eradicate or resolve the question. firewalls, and antivirus software can help
defend against malware and unauthorized
5.1. Social engineering tactics access.
Cybercriminals often leverage social engineer-
ing tactics to manipulate individuals within 5.6. Continuous monitoring
an organization. Phishing emails, pretexting, Implementing continuous monitoring solu-
and baiting are common methods used to tions enables organizations to detect and
trick employees into revealing sensitive infor- respond to security incidents in real- time.
mation or downloading malware. Awareness This involves monitoring network traffic, user
training for employees is crucial to recognize activities, and system logs to identify abnor-
and avoid falling victim to such attacks. mal behavior that may indicate a potential
security threat.
5.2. Insider threats
The threat from within an organization is a 5.7. Data backups and recovery
significant concern. Employees or other insid- Regularly backing up critical data and ensur-
ers with access to sensitive information may ing effective recovery mechanisms are in
intentionally or unintentionally pose a threat. place is crucial. In the event of a ransomware
Implementing access controls, monitoring attack or data loss, organizations can restore
Comprehensive examination of system surveillance method employing keystrokes logging technology 601
their systems to a pre-incident state, minimiz- are saved as image files that can also be stored
ing the impact. locally or remotely in an online cloud storage
service.
5.8. Multifactor authentication The Keylogger and Screenshot modules
are designed to operate discreetly without any
(MFA)
user interaction. They can be configured to
Implementing MFA adds an extra layer of initiate automatically upon computer startup
security by requiring users to provide mul- and run continuously in the background. The
tiple forms of identification before access- modules can be further configured to store
ing sensitive information. This significantly captured keystrokes and screenshots either
reduces the risk of unauthorized access, even locally or automatically upload them to an
if login credentials are compromised. online cloud storage service.
5.9. Collaboration and information 7. System Design
sharing
Captured data to an online server using the
Cybersecurity threats are dynamic, and infor- File Transfer Protocol (FTP).
mation sharing within the industry and across Implementation: Following the architec-
sectors can help organizations stay ahead of ture design, the system is implemented using
emerging threats. Collaborating with cyber- appropriate programming languages and
security communities and sharing threat libraries. The Keylogger module utilizes the
intelligence can enhance overall defense pynput library in Python to monitor and cap-
mechanisms. ture keystrokes, saving them to a log file. The
Screenshot module employs the PyScreen-
6. System Architecture shot library in Python to capture images of
the computer screen and store them as image
A comprehensive examination of the system files. The FTP module is implemented to facil-
surveillance using the Keylogger project con- itate the secure transfer of data to an online
sists of two main modules: the Keylogger cloud storage service or server.
component and the Screenshot component. Testing: The achieved system experiences
The system architecture is designed to capture allen compassing testing to guarantee that it
keystrokes and screenshots of the computer meets the particularized requirements. This
in real-time and store the data either locally includes testing the accuracy of keystroke
or automatically upload it to an online cloud capture, the functionality of screenshot cap-
storage service. ture, and the successful transfer of data using
The Keylogger component captures all FTP. Differing test cases are created to cover
keystrokes created on the computer, includ- various sketches and potential issues.
ing those fashioned in identification fields. It Integration: The Keylogger module,
uses the pynput library in Python to moni- Screenshot module, and FTP module are
tor the keyboard and save the captured key- integrated to work seamlessly as a cohesive
strokes to a log file. The log file can be stored system. This involves ensuring that data cap-
locally or optionally uploaded to an online tured by the Keylogger and Screenshot mod-
cloud storage service. ules is properly transferred and stored using
The Screenshot component captures the FTP module.
images of the computer screen at predefined Deployment: Once the system passes test-
intervals. It utilizes the pynscreenshot library ing and integration, it is deployed for use.
in Python to take screenshots of the entire Users may configure the system settings, such
screen or specific windows. The screenshots
602 Innovations in Computational Intelligence: Trends and Technologies
as the frequency of screenshot capture or the and run continuously in the background. The
destination for storing captured data. The SMTP module can be used to send email noti-
system is set to run automatically in the back- fications with the captured data to a specified
ground, providing surveillance as intended. email address.
Maintenance and Updates: Continu-
ous maintenance is carried out to address
any issues that may arise during operation. 8. Software and Hardware
Updates and improvements are implemented Requirements
based on user feedback or emerging security Language used: Python (version 3.12 and
considerations. above) Software Requirements: PyCharm
Documentation: Throughout the entire Hardware Requirements:
process, documentation is maintained to pro-
vide comprehensive information about the • RAM: 512MB (minimum requirement)
system’s design, implementation, testing pro- • Hard Disk: 2GB working space (mini-
cedures, and usage instructions. This proof mum requirement)
serves as a citation for consumers and future • Processor: Any Processor
planners. • Operating System: Any operating system
Module design: The Keylogger, Screenshot,
and FTP modules are designed to work inde-
pendently. The Keylogger module utilizes the 9. Conclusion
keyboard library to capture keystrokes. The A keylogger is a type of operating system
Screenshot module employs the PyScreenshot designed to record each keystroke made by
library to capture screenshots. The FTP mod- a user on their computer’s keyboard, often
ule uses the ftplib library to upload the cap- without their knowledge. This software is
tured data to the cloud storage service. also referred to as a keyboard capturer. While
Integration design: Once the Keylogger, they may seem intrusive, keyloggers can be
Screenshot, and FTP modules are designed, valuable tools in certain settings. For exam-
they are integrated to work together as a sys- ple, employers can use such software to mon-
tem. The modules can be configured to start itor employee computer activity and ensure
automatically when the computer boots up that work is being done efficiently, without
any unnecessary delays or complications.
References
[1] S. Moses, J. Mercado, A. Larson, and D.
Rowe, “Touch interface and keylogging mal-
ware,” 2015 11th International Conference
on Innovations in Information Technology
(IIT), Dubai, 2015, pp. 86–91. doi: 10.1109/
INNOVATIONS.2015.7381520
[2] Solairaj, S. C. Prabanand, J. Mathalairaj, C.
Prathap, and L. S. Vignesh, “Keyloggers soft-
ware detection techniques,” 2016 10th Inter-
national Conference on Intelligent Systems
and Control (ISCO), Coimbatore, 2016, pp.
AQ1:
Please
1–6. doi:10.1109/ISCO.2016.7726880
provide
Figure 84.1. System flow diagram. [3] “Advanced Keylogger - a Stealthy Malware
source for Source: for Computer Monitoring,” Asian Journal
the figure AUTHOR'S SELF GENERATED
Comprehensive examination of system surveillance method employing keystrokes logging technology 603
of Convergence in Technology, ISSN NO: Journal of Computer Networks and Com-
2350-1146 I.F-5.11, Volume VII, and Issue I. munications Security. September 2014
[4] M. Dadkhah, A. Ciobotaru, et al. “An [5] Mohamad Yusof Darus, Muhammad Azizi
Introduction to Undetectable Keyloggers Mohd Ariffin Journal: Journal of Positive
with Experimental Testing”. International School Psychology Publication Date: 2022
Volume: 6, No. 3 Pages: 8482–8492 URL:
journalppw.com
85 Health prediction application:
A comprehensive literature review
Tushar Agrawala, Prince Kumar Singhb, Aditya Kumarc, and
Mohit Chowdharyd
Student, Computer Science and Engineering Department, Galgotias College of Engineering and
Technology Greater Noida, India
Abstract: In proactive healthcare, machine learning for disease prediction is a game changer. This survey starts
by looking at the history of machine learning in healthcare, the context and the exponential growth of the field.
Then a range of machine learning algorithms for disease prediction are reviewed which are divided into super-
vised learning, unsupervised learning and other new techniques. The review looks at the strengths and weak-
nesses of each algorithm and its applications in different medical fields. The importance of privacy techniques
in health data analysis is also highlighted and data security and user confidentiality is stressed. But like any
frontier there are challenges. This survey is forward looking and identifies areas of research yet to be explored
and future directions in disease prediction through machine learning. In this summary, the essence of the field
is distilled but also real-world case studies of important disease prediction projects are brought to the fore.
These real-world case studies provide practical insights into methodologies, outcomes and benefits and add to
our understanding of the topic.
Keywords: Proactive Healthcare, supervised learning, unsupervised learning, data security, bias mitigation,
user confidentiality
1. Introduction how machine learning is being recognized as
a way to improve healthcare outcomes and
In the world of healthcare, machine learn-
the value of early disease prediction.
ing is the game changer, changing the way
Supervised, unsupervised and other emerg-
we approach disease prediction and manage-
ing techniques. Breaking them down to under-
ment. Proactive disease identification through
machine learning is a big shift from the tra- stand the strengths and weaknesses of each.
ditional model of healthcare and can change Medical domains from cardiology to neurol-
the game [1]. Disease prediction through ogy. Each with its own insights and outcomes.
machine learning empowers individuals to These sources are clinical records, wearable
take control of their health and healthcare devices, genetic information, lifestyle data.
professionals to intervene early and provide And data preprocessing techniques [3]. Now
personalized care. This comprehensive paper we dig into the details of cleaning data, engi-
takes you on a journey through this exciting neering features, and handling missing infor-
and complex space. The evolution of machine mation. Our exploration continues beyond
learning in healthcare is the story of disease these topics. We also examine ethical issues,
prediction [2]. They go back in time to explore methods to protect privacy, fairness, ways to
the history of machine learning in healthcare reduce bias, and how to use machine learning
and trace its growth. This evolution shows in healthcare. Predicting diseases covers many
a
[email protected]; [email protected]; [email protected]; dmohit.
[email protected]
DOI: 10.1201/9781003606635-85
Health prediction application: A comprehensive literature review 605
areas in healthcare, not just one. The survey These methods find groups of patients with
also looks ahead pointing out new areas to the same features, helping in making focused
research and promising future paths in using efforts [6]. Also, spotting things that don’t fit
machine learning to predict diseases. This col- in the usual health data helps find health issues
lection of knowledge boils down the field’s early that might be missed otherwise [7].
core and highlights outstanding case studies.
These real-world projects give practical les- 2.2.3. Semi-supervised and active
sons about methods, outcomes, and contribu- learning
tions deepening our grasp of the subject.
In the medical field, active learning has gained
prominence following semi-supervised learn-
2. Literature Review ing techniques which combine labelled and
unlabeled data as a means of utilizing limited
2.1. Evolution of machine learning in labelled data [8]. This has made it easier to
healthcare enhance model performance with fewer labeled
Machine learning has become a part of examples as models can now selectively solicit
healthcare over time starting in the 1960s. most informative data points for labelling.
It became more important in the late 1900s
as computers got stronger and more health 2.3. Data sources and preprocessing
data was available. Key steps included cre-
ating expert systems and using electronic 2.3.1. Clinical records
health records (EHRs) [4]. The combination Patient data is found in clinical records such
of data-driven healthcare and AI has started as electronic health records but the proce-
a new time of predictive medicine. This new dures for processing are very problematic
approach focuses on stopping illness before it due to differentness and complexity of EHRs
happens instead of treating it after. according to [9]. The quality of input data
for predictive models has been improved by
sorting out data cleaning problems as well
2.2. Machine learning algorithms for
as converting it into a format that is easy to
disease prediction work with and using the relevant features
2.2.1. Supervised learning algorithms which might help in prediction.
Supervised learning algorithms, like Support
Vector Machines (SVM), have shown they 2.3.2. Wearable devices and remote
work well to predict diseases [5]. Research monitoring
points out that SVM can sort complex medi- Wearable devices and remote monitoring have
cal info, which makes it useful to forecast dif- become important sources of health data in
ferent illnesses. Random Forest, which uses real time. These prototypes provide a way for
a team-based learning method, and Neural a person to keep track of his or her well-being
Networks, which can map out tricky pat- all the time i.e. detect any anomalies or health
terns, have also played a key role in figuring hazards in advance [10]. Incorporating weara-
out diseases ahead of time. ble data necessitates pre-processing techniques
specific to health indicators that have meaning.
2.2.2. Unsupervised learning and
clustering 2.3.3. Lifestyle data
Methods that learn without being directly Lifestyle data, encompassing dietary habits,
taught help group diseases with like patterns. physical activity, and behavioural factors,
606 Innovations in Computational Intelligence: Trends and Technologies
contribute to a holistic understanding of an Machine learning aids in identifying cancer-
individual’s health [11]. This data is often ous growths at an early stage, allowing for
integrated into predictive models to enhance more successful treatment strategies.
disease risk assessment. Data preprocessing
involves categorization, feature extraction,
2.5.3. Neurology
and sometimes the fusion of lifestyle data
with clinical records for a comprehensive Neurological disorder prediction has the
analysis. potential to enhance patient care for condi-
tions like Alzheimer’s and Parkinson’s disease.
2.4. Privacy and ethical Machine learning models can identify subtle
patterns in data, providing early insights into
considerations neurological conditions and enabling more
2.4.1. Privacy-preserving techniques effective management.
The security and privacy of healthcare data are
paramount. Researchers have explored pri- 2.6. Challenges and future directions
vacy-preserving techniques such as encrypted There is a huge number of challenges to be
data analysis to ensure data security while overcome, from technical ones—in relation
enabling advanced predictive modelling [12]. to the quality and volume of data and algo-
Secure data sharing and collaborative learn- rithm performance—to ethical and legal ones
ing methods are also being investigated to having to do with privacy, informed consent,
protect sensitive patient information. and responsible use of health data, in the case
of disease prediction models.
2.4.2. Ethical considerations Such overcoming will let one step farther
Ethical considerations, fairness, and bias ahead into the progress of disease predic-
mitigation are central to disease prediction. tion. On the other hand, the future of disease
Researchers and practitioners recognize the prediction is cheering and full of attractive
importance of responsible AI deployment in research perspectives. Some of the exciting
healthcare [13]. Ethical guidelines are emerg- directions are in refinement of algorithms,
ing to ensure the equitable treatment of exploration of multimodal data integration,
diverse patient populations and the preven- and explainable AI for healthcare. Driving
tion of algorithmic biases. future research is the potential of machine
learning to shape the future of healthcare,
proactive wellness, and reducing the burden
2.5. Application areas of disease.
2.5.1. Cardiology
Cardiac disease prediction models have shown 3. Methodology
promise in identifying individuals at risk of
heart-related ailments [14]. Early prediction 3.1. Study design
enables timely interventions, potentially sav-
ing lives and improving patient outcomes. This research employs a systematic litera-
ture review methodology to comprehensively
explore the field of disease prediction using
2.5.2. Oncology machine learning. The systematic review
Oncology benefits from early cancer detec- approach ensures a rigorous and structured
tion models, which can influence treatment analysis of existing research, allowing for a
effectiveness and patient survival rates. holistic understanding of the subject.
Health prediction application: A comprehensive literature review 607
3.2. Data collection 3.7. Validity and reliability
A systematic search of academic databases, To ensure the validity and reliability of the
including PubMed, IEEE Xplore, Scopus, and research, measures were taken to minimise
Google Scholar, was conducted. The search bias. This included the use of clear inclusion
strategy involved a combination of keywords criteria, consistent data extraction proce-
related to disease prediction, machine learn- dures, and a structured analysis process.
ing, and healthcare. Only peer-reviewed
research papers, articles, and academic litera- 3.8. Data visualization
ture were considered.
Data visualisation techniques, such as tables,
charts, and graphs, were used to enhance the
3.3. Data selection and inclusion presentation of research findings, making
criteria complex information more accessible to the
The inclusion criteria for selecting literature reader.
encompassed the following:
• Research papers published between 2019 3.9. Research findings
and 2023.
The research findings section of this paper
• Papers addressing the application of
presents the synthesised insights, highlight-
machine learning in disease prediction.
ing key themes, trends, and through machine
• Full-text papers with clear methodologies
learning.
and results.
3.4. Data analysis 4. Result
Data extraction involved the collection of This review represents a wide variety of
essential information from each selected machine learning algorithms applied to the
paper, including the publication year, study process of disease prediction: support vector
design, machine learning algorithms used, machines, random forests, neural networks,
data sources, application domains, and key K-nearest neighbors, logistic regression.
findings. The collected data were organised SVM and Random Forest have shown
and categorised for analysis. their effectiveness more than other algo-
rithms in classifying complex
Kindlymedical
add textdata
hare:
drawn from a primary source like Electronic
"as shown in figure
3.5. Ethical considerations 85.1"
Health Records.
Ethical considerations in this study pertain to
responsible referencing and the proper cita-
tion of sources. All sources are appropriately
credited to maintain academic and research
integrity.
AQ1:
Please
3.6. Data synthesis provide
in-text
The collected data were synthesised to create citation
for figure
a coherent narrative, allowing for the identi- 1
fication of key themes, trends, and insights. A
thematic analysis approach was employed to Figure 85.1. Data driven diabetes prediction.
AQ2:
categorise and interpret the findings. Source: Please
AUTHOR'S SELF GENERATED
provide
source for
the figure
608 Innovations in Computational Intelligence: Trends and Technologies
Application domains are manifold, with References
deep research in cardiology, oncology, and
[1] K. Kurapati, “Proactive and Intelli-
neurology. Responsible use of healthcare data,
gent Healthcare Management Using
fairness, and mitigating bias come as very IoT,” 2022 International Conference on
prominent themes in the literature, whereas Advances in Computing, Communication
budding research directions can be said to and Applied Informatics (ACCAI), Chen-
shape the future of disease prediction using nai, India, 2022, pp. 1-7, doi: 10.1109/
machine learning for transformative impact ACCAI53970.2022.9752579.
in healthcare. Case studies from real-world [2] P. Zhang, X. Huang and M. Li, “Disease Pre-
applications bring both practical applications diction and Early Intervention System Based
and future possibilities, hence giving an all- on Symptom Similarity Analysis,” in IEEE
inclusive view of the state of this field. Access, vol. 7, pp. 176484-176494, 2019,
doi: 10.1109/ACCESS.2019.2957816.
[3] A.A.H., Leeuwenberg, A.M., Hooft, L. et
5. Discussions al. Guidelines and quality criteria for arti-
ficial intelligence-based prediction mod-
This part of the paper talks about how
els in healthcare: a scoping review. npj
machine learning helps us guess diseases, Digit. Med. 5, 2 (2022). https://do10.1038/
why it’s important, the hard parts, and what s41746-021-00549-7
we should think about in terms of right and [4] Yang, X., Chen, A., PourNejatian, N. et al.
wrong. Picking the right machine learning A large language model for electronic health
tools, like Support Vector Machines (SVM) records. npj Digit.
and Random Forest, is very key. Each has its [5] Med. 5, 194 (2022). https://doi.org/10.1038/
own good points and can fit well with differ- s41746-022-00742-2
ent health areas. [6] K., Thangavelu, M. RETRACTED ARTI-
The work of guessing diseases early touches CLE: Multi-disease prediction model using
on heart health, cancer care, brain health, and improved SVM-radial bias technique in
more, aiming to find issues soon and help healthcare monitoring system. J Ambi-
ent Intell Human Comput 12, 3715–
patients better. Yet, this path is full of tough
3723 (2021). https://doi.org/10.1007/
spots, including tech and moral issues. It’s key
s12652-019-01652-0
to use AI in health care in a good way, and new [7] Yanshan Wang, Yiqing Zhao, Terry M.
rules are in place to make sure it’s fair, avoids Therneau, Unsupervised machine learn-
bias, and keeps data safe. Looking ahead, the ing for the discovery of latent disease clus-
focus is on making better algorithms, mix- ters and patient subgroups using electronic
ing different types of data, and creating AI in health records, Journal of Biomedical
health care that people can understand. This Informatics, Volume 102, 2020, 103364,
shows how fast this area is changing. ISSN 1532-0464, https://doi.org/10.1016/j.
Furthermore, the applications of predict- jbi.2019.103364.
ing illnesses stretch out to worldwide health [8] Tharindu Fernando, Simon Denman, Sridha
and public health. This in turn helps in early Sridharan, and Clinton Fookes. 2021. Deep
warning and surveillance through the appli- Learning for Medical Anomaly Detection – A
Survey.54, 7, Article 141 (September 2022),
cation of predictive models to give informa-
37 pages. https://doi.org/10.1145/3464423
tion about resource allocation in the health
[9] A. A. Abdullah,”A Review on Bayesian
care systems. At its core, this part seeks to Deep Learning in Healthcare: Applica-
investigate into how machine learning could tions and Challenges,” in IEEE Access, vol.
transform healthcare because it bears futuris- 10, pp. 36538-36562, 2022, doi: 10.1109/
tic consequences on diseases prediction and ACCESS.2022.3163384.
healthcare paradigms.
Health prediction application: A comprehensive literature review 609
[10] Mahajan, H.B., Rashid, A.S., Junnarkar, [13] Pika, A.; Wynn, M.T.; Budiono, S.; ter
A.A. et al. Integration of Healthcare 4.0 Hofstede, A.H.M.; van der Aalst, W.M.P.;
and blockchain into secure cloud-based Reijers, H.A. Privacy-Preserving Process
electronic health records systems. Appl Mining in Healthcare. Int. J. Environ. Res.
Nanosci 13, 2329–2342 (2023). https://doi. Public Health 2020, 17, 1612. https://doi.
org/10.1007/s13204-021-02164-0 org/10.3390/ijerph17051612
[11] Al-khafajiy, M., Baker, T., Chalmers, C. [14] Jessica Morley, Caio C.V. Machado, Chris-
et al. Remote health monitoring of elderly topher Burr, Josh Cowls, Indra Joshi,
through wearable sensors. Multimed Tools Luciano Floridi, The ethics of AI in health
Appl 78, 24681–24706 (2019). https://doi. care: A mapping review, Social Science and
org/10.1007/s11042-018-7134-7 Medicine, Volume 260, 2020, 113172,
[12] Al-khafajiy, M., Baker, T., Chalmers, C. ISSN 0277-9536, https://doi.org/10.1016/j.
et al. Remote health monitoring of elderly socscimed.2020.113172.
through wearable sensors. Multimed Tools [15] C., Virk, H.U.H., Bangalore, S. et al. Machine
Appl 78, 24681–24706 (2019). https://doi. learning prediction in cardiovascular diseases:
org/10.1007/s11042-018-7134-7 a meta-analysis. Sci Rep 10, 16057 (2020).
https://doi.org/10.1038/s41598-020-72685-1
86 E-portal for case management and
hearing
Anushka Shuklaa, Anurag Shuklab, Ashish Kumar Yadavc, and
Anushka Srivastavad
Computer Science and Engineering, Babu Banarsi Das Institute of Technology and Management,
Lucknow, India
Abstract: In the rapidly evolving landscape of legal proceedings, the need for efficient and user-friendly case
management solutions has become paramount. Our e-portal project, titled “Legal LinkHub,” is designed to
revolutionize the way individuals and legal professionals engage with the legal system. “Legal LinkHub” offers
a comprehensive suite of features, including a rich repository of articles for case awareness, an innovative pre-
trial conferencing platform, automated email and SMS generators for seamless communication and reducing
Pending cases, and a guided pathway to determine the appropriate court for specific cases. The core objective of
“Legal LinkHub” is to streamline and demystify the often complex and intimidating legal process, empowering
users to make informed decisions about their legal matters. By leveraging technology to provide transparency
and accessibility, we aim to bridge the gap between legal professionals, the judiciary, and the public. “Legal
LinkHub” is not just an e-portal, it is a transformative tool that enhances legal literacy, and promotes fair and
efficient decision-making. This provides a glimpse of the comprehensive case management capabilities “Legal
LinkHub” offers, addressing the growing demand for accessible and modernized legal services. The project’s
main features are its mission to simplify legal proceedings and its commitment to promoting fairness and effi-
ciency within the legal system. It gives readers an overview of the e-portal’s significance and purpose, encourag-
ing them to explore the full project synopsis for more details.
Keywords: pre-trial conferencing platform, automated email and SMS, guidance on Court Selection
1. Introduction with several features that aim to make the
legal process easier for all stakeholders. We
In today’s ever-changing legal landscape, the
designed it to address these challenges com-
need for streamlined and easy-to-use case
prehensively. ‘Legal Link Hub’ provides you
management solutions has never been more
important. “Legal Link Hub” is an innova- with the tools and resources you need to
tive e-portal project that aims to drastically make informed decisions and improve legal
change how legal case management and efficiency, regardless of whether you are an
awareness are perceived. This system is sup- individual looking for legal advice.
posed to maintain the law and resolve con- The innovative technology stack at the
flicts, often creating challenges due to its core of “Legal Link Hub” was selected with
complex rules and lengthy court procedures. care to guarantee the platform’s effectiveness,
These intricacies, besides an increasing back- security, and user-friendliness. This stack
log of cases and limited court resources, are includes server-side technologies like Node.
what because delays and growing annoyance. js and Express.js for reliable back-end devel-
“Legal Link Hub” is a dynamic platform opment, as well as contemporary front-end
[email protected]; bshukla.anu23 @gmail.com; [email protected];
a
DOI: 10.1201/9781003606635-86
E-portal for case management and hearing 611
technologies like React.js for an interactive and make the legal system more efficient in
user interface. The platform makes use of Kenya..
MongoDB Server, a relational database man- Meka et al. (2023): The government’s
agement system, to guarantee data efficiency digital initiatives with the vision of Dr. B.R.
and integrity, especially because legal data Ambedkar, who dedicated his life to eliminat-
is structured. We improved communication ing social injustices. The digital revolution is
with users through integration with third- contributing to the realization of Ambedkar’s
party services like SendGrid for email and dream of a more equitable and opportunity-
Twilio for SMS. This allows for effective mes- rich India. The author’s expertise and com-
saging and notifications. “Legal LinkHub” mitment to social justice lend credibility to
uses the video conferencing service Zoom to this connection. This paper underscores how
enable efficient pre-trial conferences, a func- technology can accelerate progress toward
tion that can expedite correspondence and Ambedkar’s vision of equality, liberty, and
prompt resolution of disputes. Passport.js for equal opportunity in India.
Node.js’s strong user authentication guar- Aboelazm et al. (2022): The study con-
antees safe access control and data security. cludes that the integration of an e-justice sys-
Thus, “Legal LinkHub” is at the forefront of tem has the potential to significantly enhance
technological innovation, committed to offer- the efficiency of the judicial process, especially
ing a safe, intuitive, feature-rich platform that within the framework of the Egyptian Council
completely transforms the legal awareness of State. It underscores the pressing need for
and case management space. tailored strategies to address the challenges
specific to the implementation of e-justice in
the Egyptian economic courts, thereby ensur-
2. Literature Survey ing a seamless transition towards a more
Syroka et al. (2023): The article empha- technologically driven legal landscape.
sizes how important it is to change Fed-
eral Criminal Rule 53 to accommodate the
demands of the legal system and the changing 3. Methodology
legal technology environment. The CARES In the context of the “Legal LinkHub” pro-
Act, which introduced the expiring permis- ject, the criteria for selecting the literature are
sion for virtual criminal proceedings, empha- carefully crafted to ensure the sources chosen
sizes the usefulness of using phone and video are highly pertinent and reliable. The primary
conferencing during criminal trials. By alter- criterion is relevance, with a focus on litera-
ing Rule 53. We can improve defendants’ ture directly related to the legal sector, tech-
access to justice and advance the cause by nology in law, e-portals, case management,
adjusting to these technologies. Criminal and legal awareness. Recency Citationisinprioritized
Text:
dockets’ effectiveness. "as shown
to align the chosen literature with the in current
Figure
Ochieng et al. (2023): The introduction of 86.1."
state of the field, typically encompassing the
virtual court proceedings in Kenya, prompted last five years, reflecting the rapidly evolv-
by the COVID-19 pandemic, has significantly ing nature of technology and the legal sector.
improved access to justice. It has enabled vul- The incorporation of diverse perspectives,
nerable groups like expectant mothers and case studies, and sources addressing techno-
persons with disabilities to participate with- logical aspects, user authentication, security,
out physical presence, ensuring privacy and legal challenges, and potential solutions col-
protection in sensitive cases. The implemen- lectively constitute the comprehensive criteria
tation of ICT services and e-filing further that guide the literature selection process for
demonstrates the commitment to modernize “Legal LinkHub.” The process of searching,
612 Innovations in Computational Intelligence: Trends and Technologies
screening, and selecting the literature begins
with a clear definition of research goals,
ensuring they align with the project’s objec-
tives. This helps in figuring out the important
words and topics to look for. We use various
databases like Google Scholar, Doi Finder,
and Springer to search for this information.
Once we find potential sources, we carefully
check when they were published, how reli- Figure 86.2 Captain:
able the source is, whether they relate to our Process Flow Diagram for
research goals, and if they have good con- User Authentication and
tent. We follow certain rules to decide which Legal Services Access in
sources to include and which to leave out. Figure 86.2. Legal LinkHub
We also look at these articles’ sources to find Source:
AUTHOR'S SELF GENERATED
even more information. Finally, we organize
the information into different groups based
on the main ideas, and we always give credit
to the people who wrote the articles we used. Link Hub addresses the unique challenges
This helps keep our research honest and faced by individuals, particularly those in less
well-organized. accessible areas, by providing date reminders
for upcoming court hearings.
Article for Case Awareness: The “Article
4. Thematic Overview for Case Awareness” feature in the legal link
Pre-trial Conferencing: Implementing a pre- hub offers informative articles, guides, and
Kindly add here, Figure 86.2 Citation in
trial conferencing system in the legal context resources to help the user better understand
Text:
harnesses technology to foster collaboration their cases and legal issues. Create a tag-based
"The process
searchof user authentication,
interface where users can select or
among parties involved in a case, hencecontent
yield- access, and legal service
ing multiple benefits. This technology reduces input tags, they’re interested in. Develop a rec-
provisioning is mapped out in Figure 86.2,
the frequency of physical encounterswhich ommendation algorithm that suggests related
as itdemonstrates how the system
allows participants to bring together and articles based on matching tags. Incorporate
dis- login,
handles court selection, and
cuss case matters remotely, saving time and user
automated feedback to
notifications refine recommendations.
through
resources. Guidance on Court Selection: The plat-
third-party integrations."
Auto-email and SMS generator: The auto- form will provide a navigational guide to help
email and SMS generator within the Legal users determine which court is appropriate
for their specific legal case. Users will receive
guidance on the jurisdiction, court type, and
court location best suited for their legal mat-
ters, reducing confusion and delays.
User Input: When a user visits the e-portal
for the first time, they must register if they
haven’t already done so. During registration,
AQ1:
Please they will input case-related information and
provide verify their identity, indicating whether they
figures
caption,
are a client or a legal professional.
Figure 86.1 Captain:
source Authorization Check: To make sure the
System Interaction Model of Legal LinkHub
and Figure 86.1. user has access to rights and permissions,
citation.
Source: AUTHOR'S SELF GENERATED
the system performs an authorization check
E-portal for case management and hearing 613
using JSON Web Tokens. After authentica- research gaps remain to be explored to
tion, it gave users access as needed. further strengthen the project, it is para-
Video Conferencing: The e-portal uses a mount to conduct a comprehensive anal-
set of technology tools to make video meet- ysis of the effectiveness of automated
ings possible for legal matters. This helps to communication in reducing the backlog of
reduce pending cases, the e-portal is inte- pending legal cases. This research should
grated with the Zoom app. encompass not only the rate of response
Court Selection: The e-portal provides a but also delve into the actual outcomes of
Court Selection feature based on the Infor- cases where these tools are employed.
mation provided by the users. By assisting C. Providing automated SMS notifications
users in selecting the best court for their par- for pre-trial dates is a valuable and prac-
ticular case, this tool simplifies the procedure tical feature that can significantly ben-
and eliminates needless complications. efit individuals involved in legal cases. It
Articles for Legal Education: Showcase helps ensure that they are well-informed
less-known laws that might fascinate cli- about their court dates and reduces the
ents, alongside explanations of complex legal likelihood of missing important appoint-
ideas to help them better understand the ments. Pre-trial support, legal awareness,
legal world. This approach can offer valuable and assistance in making informed deci-
information to individuals in search of legal sions about choosing the right court offer
understanding. a multifaceted approach to improving the
Auto-email and SMS generator: Our legal process.
platform integrates with third-party service D. Unlike traditional case management sys-
Twilio for SMS and SendGrid for email to tems, ‘Legal LinkHub’ doesn’t just focus
implement auto-mail and SMS generators, on procedural aspects but also provides a
ensuring efficient and reliable communica- wealth of educational content. The repos-
tion with our users, for the next upcoming itory of articles enhances case awareness,
date for the case hearing. making legal information accessible and
understandable for users.
5. Critical Analysis
A. In contrast to existing projects focused 6. Conclusion
on legal pre-trial conferencing and legal With its extensive feature set and array of
awareness, your website offers a distinc- resources, our e-portal for case manage-
tive approach by introducing automated ment hearings is a ground-breaking solution
email and SMS systems designed to expe- that seeks to completely transform the legal
dite the resolution of pending legal cases. industry. This was created in response to the
This addition represents a significant leap urgent need for effective, knowledgeable, and
forward in the effort to streamline legal approachable legal tools. Our case manage-
processes and enhance legal awareness ment hearing e-portal is a comprehensive
among the public. solution that aims to improve the legal sys-
B. Our project stands out for its integra- tem in several ways. Its goal is to advance
tion of automated communication tools, accessibility, effectiveness, and transparency
which can potentially lead to swifter case in the legal field. Our mission is to empower
resolutions and increased legal knowl- users to make informed decisions, expedite
edge among individuals, ultimately con- case management, and ultimately contribute
tributing to a more efficient and informed to a just and equitable legal system by fus-
legal system. However, several pertinent ing practical tools, educational resources, and
614 Innovations in Computational Intelligence: Trends and Technologies
decision support. Our e-portal will continue Department of Computer Science and Engi-
to be developed and expanded, and we are neering) at Babu Banarasi Das Institute of
dedicated to our goal of enhancing everyone’s Technology and Management, Lucknow for
access to, comprehension of, and fairness in their constant support and guidance through-
the legal system. Legal experts and knowl- out our work. Their sincerity, thoroughness
edgeable citizens, in our opinion, are better and perseverance have been a constant source
suited to promote justice and effect construc- of inspiration for us. It is only through their
tive change. cognizant efforts that our endeavors have
seen the light of the day. We also do not like
to miss the opportunity to acknowledge the
7. Future Scope contribution of all faculty members of the
In envisioning the future development of department for their kind assistance and
your e-portal for case management hearings, cooperation during the development of our
several promising directions emerge. These project. Last but not least, we acknowledge
avenues hold the potential to further enhance our family and friends for their contribution
the effectiveness and reach of your platform. to the completion of the project.
Expanded Legal Resource Library: Con-
sider augmenting your e-portal with an exten-
sive legal resource library, going beyond mere References AQ2:
Please
articles. Incorporate multimedia content, [1] Benjamin Syroka (2023) Amending Federal provide
Criminal Rule 53 for Virtual Criminal Hear- reference
comprehensive guides, and interactive tools in-text
that educate users on various legal aspects. ings, Federal Bar Association Litigation Sec- citation
These resources can cover not only case tion Newsletter. for all
awareness but also legal procedures, rights, [2] Ž Kešetović, K Radojević—Култура полиса references
(Kultura polisa) (2023), Crisis Communi-
and responsibilities, making your e-portal a
cation at the Local Level—A Case Study
comprehensive legal learning hub. of Novi Pazar, International Journal of
Multilingual and Global Reach: Consider Advanced Research in Computer Science.
expanding the multilingual support of your [3] Alexandra Akinyi Ochieng (2023) Reform-
e-portal to cater to an international audience. ing Virtual Court Sessions in Kenya to
Make your platform adaptable to diverse Enhance Access to Justice: Addressing the
legal systems and languages, ensuring its util- challenges, Alternative dispute resolution.
ity transcends geographical boundaries. [4] James Stephen Meka (2023) Digital India
Social and Community Features: Integrate - Ambedkar’s Vision and Modi’s Provision,
social and community features to encourage Journal of Engineering Sciences.
user interaction and support. Users can con- [5] Helen Taylor Dirk Van Rooy, Lorana Bar-
nect, share experiences, and seek advice from tels (2023) Digital justice: A rapid evidence
peers and legal experts within the e-portal assessment of the use of mobile technol-
ogy for offender behavioural change,
community, fostering a sense of collaboration
International Journal of Social Research
and assistance. Methodology.
[6] Richard Heeks (2023) Digital inequality
Acknowledgment beyond the digital divide: conceptualizing
adverse digital incorporation in the global
It gives us a great sense of pleasure to present South, IJMIR.
the report of the B. Tech Project undertaken [7] Akshaya Kamalnath (2023) The future of
during B. Tech. Final Year. We owe a special corporate insolvency law: A review of tech-
debt of gratitude to Anurag Shukla (Assis- nology and AI-powered changes, Interna-
tant Professor) and Dr. Anurag Tiwari (Head, tional Insolvency review.
E-portal for case management and hearing 615
[8] Carolyn McKay and Kristin Macintosh [10] Tsz Ho Kwan (2022) Enforcement of the
(2022) Remote Court Proceedings in Nigeria: Use of Digital Contact-Tracing Apps in
Justice Online or Justice on the Line, Inter- a Common Law Jurisdiction, Healthcare
national Journal for Court Administration. Organization.
[9] Awaludin Marwan, Fiammetta Bonfigli [11] Dr. Karem Aboelazm (2022) The Role of
(2022) Detection of Digital Law Issues and Digital Transformation in Improving the
Implication for Good Governance Policy in Judicial System in the Egyptian Council of
Indonesia, Bestuur. State, Scientific Foundations of Digital Gov-
ernance and Transformation.
87 Enhancing electoral processes: The role
of “Smart Voting Systems”
Madhulika Sharmaa, Anuj Singhb, and Gulshan Varshneyc
Computer Science and Engineering, Babu Banarasi Das Institute of Technology and Management,
Lucknow, India
Abstract: This research scrutinizes the emergence and impact of Smart Voting Systems within contemporary
democratic structures. It offers a comprehensive evaluation of these systems, dissecting their functionality, secu-
rity measures, and inclusive potential. Delving into various models incorporating cutting-edge technologies like
blockchain, biometrics, and artificial intelligence, the study explores their pivotal role in augmenting transpar-
ency and facilitating broader access to electoral processes Through in-depth case studies and meticulous analy-
sis, this paper unveils the multifaceted implications of Smart Voting Systems, shedding light on their capacity to
bolster democratic integrity while navigating intricate concerns surrounding privacy, cybersecurity, and ethical
considerations. By advocating for a balanced approach, emphasizing the critical attributes of integrity and inclu-
sivity, the research advocates for a pathway forward in refining future electoral technologies.
Keywords: Blockchain, Biometrics, Artificial Intelligence, Transparency, Privacy, Cybersecurity, Democracy,
Digital Voting, Integrity
1. Introduction in the electoral framework. To maintain its
integrity, stringent security measures, encryp-
A smart voting system integrating fingerprint tion protocols, and authentication algorithms
and facial recognition technologies stands are continually updated and audited, mitigat-
as a robust solution for ensuring the integ- ing risks of tampering or unauthorized access.
rity and security of electoral processes. This This approach ensures the system’s originality
innovative system relies on biometric authen- and guards against plagiarism concerns while
tication, verifying voters’ identities through prioritizing the robustness of its development
unique biological markers, thereby curbing and security infrastructure
fraudulent practices like impersonation and
multiple voting attempts. Central to its design
is a secure database housing voter informa- 1.1. Context of the review
tion, including encrypted biometric data, The smart voting system revolutionizes tra-
ensuring privacy while upholding the sanc- ditional electoral processes by integrating
tity of the voting process. Its user-friendly cutting-edge biometric authentication meth-
interface streamlines voting procedures for ods like fingerprint and facial recognition. Its
both voters and election officials, minimizing primary goal is to bolster security, efficiency,
errors and enabling swift, real-time valida- and the overall integrity of voting procedures.
tion of identities, thus reducing queues and This innovation hinges on biometric mark-
enhancing efficiency. Emphasizing transpar- ers to validate voters’ identities, thwarting
ency, the system provides an auditable trail of fraudulent activities and instances of multiple
votes cast, fostering accountability and trust voting.
[email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-87
Enhancing electoral processes: The role of “Smart Voting Systems” 617
Central to its implementation is the With user-friendly interfaces and real-time
establishment of secure databases storing validation of voter identities, these systems
encrypted biometric data, safeguarding pri- minimize wait times and simplify the voting
vacy while ensuring the prevention of fraud. experience at polling stations.
The system’s user-friendly interface simplifies Moreover, the implementation of stringent
voting for both citizens and election officials, security measures and encryption protocols
reducing errors and expediting the process. within these systems helps in preventing tam-
Real-time validation of voter identities ena- pering, unauthorized access, or manipulation
bles swift verification, streamlining voting of voting data, thereby ensuring the authen-
and diminishing wait times at polling stations. ticity of election results.
Transparency and accountability are para- Ultimately, a smart voting system aims
mount features of this system. It maintains to promote democratic values by encourag-
a detailed record of votes cast, allowing for ing higher voter participation and ensuring
scrutiny and bolstering trust in the electoral the security and credibility of the electoral
process. Stringent security measures, includ- process.
ing encryption protocols and authentication
algorithms, fortify the system against tamper- 1.3. Significance of the topic
ing, unauthorized access, and potential secu-
rity breaches. The significance of a smart voting system is
Overall, the smart voting system repre- immense, primarily in its ability to revolu-
sents a pioneering leap in modernizing elec- tionize democratic processes. These systems
toral practices, prioritizing accuracy, security, play a pivotal role in ensuring the security
and transparency while significantly improv- and fairness of elections by utilizing advanced
ing the voting experience for citizens. biometric authentication methods, like finger-
print or facial recognition. This significantly
reduces fraudulent activities such as iden-
1.2. Objectives and purpose of the tity theft and multiple voting, upholding the
literature review integrity of the voting process and bolstering
The primary objective of a smart voting sys- trust among citizens in the electoral system.
tem is to significantly improve the efficiency, Moreover, smart voting systems meticu-
security, and accessibility of the electoral pro- lously record votes, fostering transparency
cess. It achieves this by employing cutting- and enabling easy verification. This transpar-
edge biometric authentication methods like ency is crucial in instilling confidence in elec-
fingerprint or facial recognition to accurately tion outcomes and diminishing skepticism
verify voters’ identities. By doing so, it greatly about the authenticity of results.
reduces the chances of fraudulent activities They also prioritize accessibility and effi-
such as impersonation or multiple voting ciency, simplifying voting procedures through
instances, thereby upholding the integrity of user-friendly interfaces and streamlined pro-
the voting process. cesses. This not only encourages broader par-
Additionally, these systems aim to ensure ticipation in elections but also minimizes wait
the accuracy and transparency of the voting times at polling stations, making the demo-
process by maintaining meticulous records of cratic process more inclusive.
votes cast. This transparency allows for scru- By embracing these systems, countries
tiny and verification, fostering confidence and showcase their commitment to adopting
trust in the election’s fairness. technological advancements for the better-
Another crucial goal is to make voting ment of democratic practices. They mitigate
more accessible and streamlined for citizens. various challenges inherent in traditional
618 Innovations in Computational Intelligence: Trends and Technologies
voting methods, such as ballot tampering and 5. Transparency Measures: The system
miscounting, reducing the likelihood of dis- maintains transparent records of all votes
putes and reinforcing the robustness of the cast, allowing for auditing and verifica-
electoral process. tion. This transparency safeguards the
In essence, the significance of smart vot- integrity of the election process and ena-
ing systems lies in their capacity to safeguard bles scrutiny if necessary.
democratic principles of fairness, accuracy, 6. Robust Security Measures: To safeguard
and accessibility while leveraging technology against cybersecurity threats, the sys-
to overcome challenges, ultimately strength- tem employs strong security protocols,
ening the democratic framework of a nation. encryption algorithms, and undergoes
regular security checks. These measures
prevent unauthorized access or tampering
2. Methodology with the voting data.
The methodology behind a smart voting sys- 7. Continuous Enhancement: Post-election,
tem involves a systematic approach using the system is evaluated to identify areas
advanced technology to conduct secure and for improvement. Continuous refinement
efficient elections. It encompasses several key ensures the system remains robust and
elements: aligns with evolving technological and
security standards.
1. Biometric Verification: Initially, eligi-
ble voters are registered into a database, In essence, the methodology of a smart
where their unique biometric data, like voting system centers on utilizing biometric
fingerprints or facial characteristics, is verification, secure database management,
securely stored. During voting, this data user-friendly interfaces, real-time validation,
is utilized to authenticate the voter’s iden- transparency, stringent security measures,
tity, ensuring only authorized individuals and continual evaluation. This collective
can cast their votes. approach ensures the conduct of secure, effi-
2. Secure Database Management: All cient, and credible electoral processes.
enrolled biometric information and voter
records are stored in a central database
with robust security measures in place. 3. Thematic Overview
Encryption methods safeguard this sensi- The Smart Voting System is an innovative
tive data, preventing unauthorized access approach designed to transform the electoral
or manipulation. process fundamentally. It utilizes sophisti-
3. User-Friendly Interfaces: The system cated biometric technologies, such as fin-
is designed with intuitive interfaces to gerprint and facial recognition, to validate
facilitate smooth interactions for vot- voters’ identities, thereby ensuring legitimacy
ers. Whether it’s the registration process, and deterring fraudulent practices like iden-
identity verification at polling stations, or tity theft or multiple voting attempts.
electronic voting, user-friendly interfaces Its methodology comprises essential ele-
ensure a seamless experience. ments: a secure repository that safely stores
4. Real-Time Validation: Instant validation encrypted biometric data, user-friendly inter-
of a voter’s identity occurs by cross-ref- faces for simplified voter interaction, real-
erencing their biometric data with the time validation of voter identities during the
stored records. This swift validation mini- voting phase, and the maintenance of trans-
mizes waiting times and ensures accurate parent records for verification and audit
authentication during the voting process. purposes.
Enhancing electoral processes: The role of “Smart Voting Systems” 619
The system places a strong emphasis on 4. Privacy and Data Protection: Storing
security, employing stringent measures to biometric data centrally raises privacy
safeguard against unauthorized access and concerns. Safeguarding this sensitive
manipulation of voting data. Ongoing evalu- information from unauthorized access or
ations and upgrades are integral to maintain- misuse is crucial, balancing security with
ing the system’s strength and aligning it with privacy and adhering to data protection
evolving technological standards. regulations.
In essence, the Smart Voting System 5. Implementation Challenges: High costs
strives to enhance the accuracy, security, associated with implementation pose
transparency, and inclusivity of elections. challenges, particularly for regions with
Through technology and rigorous secu- limited resources. Proper training for elec-
rity protocols, it aims to build trust in the tion officials and educating voters about
democratic process, encouraging higher the system’s use are essential but could
participation and upholding the integrity of pose logistical hurdles.
election outcomes. 6. Building Trust and Transparency: Ensur-
ing public trust requires transparent com-
munication about the system’s operations,
4. Critical Analysis data handling procedures, and security
The introduction of the smart voting system measures. This transparency is crucial to
marks a significant evolution in electoral dispel doubts about fairness and transpar-
methodologies. However, a critical analysis of ency in elections.
its implementation and functionality reveals 7. Risk of Voter Apathy: Excessive reliance
various aspects that require scrutiny: on technology might lead to voter dis-
engagement or skepticism if individu-
1. Enhanced Security: Employing biometric als mistrust the system’s integrity or feel
authentication like fingerprints and facial disconnected from the traditional voting
recognition boosts security by preventing process.
fraudulent activities. Yet, the system isn’t
immune to potential vulnerabilities. Con- In summary, while the smart voting system
cerns persist about the security of stored promises improvements in accuracy, secu-
biometric data, raising questions about rity, and efficiency, it demands a thorough
the overall system’s resilience against approach to address challenges related to
hacking or data breaches. security, accessibility, privacy, reliability, cost,
2. Accessibility Challenges: While aiming to trust-building, and inclusivity to ensure its
streamline voting, accessibility for diverse successful and equitable integration into elec-
demographics, especially those with lim- toral processes.
ited technological proficiency or physical
disabilities, remains a concern. Ensuring
equitable access and inclusion for all vot- 5. Recommendations For Future
ers is crucial. Research
3. Reliability on Technology: Dependence 1. Privacy-Preserving Biometrics: Investigate
on technology raises reliability issues. techniques that enable biometric verifi-
Technical malfunctions, system failures, cation without storing complete biomet-
or cyber threats could disrupt elections ric data. This could involve methods like
and compromise results. There’s a need zero-knowledge proofs or encryption,
for robust contingency plans to address enhancing security while preserving sensi-
potential technological breakdowns. tive information.
620 Innovations in Computational Intelligence: Trends and Technologies
2. Blockchain Integration: Explore integrat- advanced biometric technology like finger-
ing blockchain technology to bolster the print and facial recognition to authenticate
security and transparency of smart vot- voters. This innovative approach aims to bol-
ing systems. Its decentralized nature could ster election security by reducing fraudulent
address concerns regarding data tamper- activities such as identity theft or multiple
ing and security breaches. voting attempts.
3. Usability and Accessibility: Conduct com- A core element of this system is its secure
prehensive studies to ensure user-friendly database where encrypted biometric data is
interfaces that cater to all demographics, stored, ensuring the protection and integrity
including elderly voters and those with of sensitive voter information. Addition-
disabilities. Tailoring interfaces to diverse ally, user-friendly interfaces cater to diverse
user needs is crucial. technological abilities, simplifying the vot-
4. Public Perception and Trust: Research ing process. Real-time identity validation
public perceptions to understand con- during voting further ensures accuracy and
cerns about security, privacy, and trust. legitimacy.
Addressing these concerns is vital to instill However, challenges exist. Privacy con-
confidence and acceptance among voters. cerns arise from the use of biometric data,
5. Resilience Testing: Further investigate the demanding stringent security measures to
system’s resilience against cyber threats, safeguard against potential breaches or mis-
system failures, or manipulation attempts. use. Equally important is ensuring accessi-
Identifying vulnerabilities and devising bility for all voter demographics, including
effective countermeasures is essential. elderly or disabled individuals.
6. Ethical and Legal Implications: Examine The system’s resilience against techni-
the ethical and legal aspects of using bio- cal glitches or cyber threats is vital. Regular
metric data in voting systems. Addressing testing and enhancements are necessary to
concerns about data ownership, consent, address vulnerabilities and maintain reliabil-
and biases is critical. ity during elections. Public trust is paramount,
7. Cost-Effectiveness: Research scalable and necessitating transparent communication
cost-effective solutions, especially for about system operations, security protocols,
regions with limited resources. Finding and privacy measures.
ways to make these systems more afford- Continued research in areas like privacy-
able can facilitate wider adoption. preserving biometrics, blockchain integra-
8. Hybrid Voting Models: Explore hybrid tion, usability studies, and cost-effective
models that combine traditional and smart solutions is pivotal. Balancing technological
voting systems, offering a balance between innovation with security, accessibility, and
technological advancements and familiar public confidence is crucial for successful
voting practices, ensuring inclusivity. integration into future elections, encouraging
democratic participation, and upholding the
Continued exploration in these areas can election’s integrity.
refine smart voting systems, addressing chal-
lenges and improving their reliability, security,
accessibility, and acceptance for future elections. References
[1] M. Kandan, K. D. Devi, K. D. N. Sri, N.
Ramya and N. K. Vamsi, “Smart Voting Sys-
6. Conclusion tem using Face Detection and Recognition
The Smart Voting System marks a significant Algorithms,” 2021 IEEE International Con-
stride in modernizing elections by employing ference on Intelligent Systems, Smart and
Green Technologies (ICISSGT), India, 2021.
Enhancing electoral processes: The role of “Smart Voting Systems” 621
[2] V. Laxmi Vashisht, H. Mohan and S. Prakash, [5] “Secured E-voting System Using Two-factor
“Smart Voting System Through Face Recog- Biometric Authentication,” 2020 Fourth
nition,” 2022 4th International Conference International Conference on Comput-
on Advances in Computing, Communica- ing Methodologies and Communication
tion Control and Networking (ICAC3N), (ICCMC), Erode, India, 2020.
Greater Noida, India, 2022. [6] Hemlata Sahu and Anupam Choudhray,
[3] S.A anjana devi, Dr.V.Palanisamy and “Polling System Using GSM S. Komatineni
R.Anandha Jothi, A Study on Secure Online and G. LingalaFacility”, International Jour-
Voting System using Biometrics Face Detec- nal of Scientific and Engineering Research,
tion and Recognition Algorithms, Interna- vol. 2, no. 10, Oct 2011.
tional Journal for Modern Trends in Science [7] R. Kaur and E. Himanshi, “Face recognition
and Technology, August 2017. using Principal Component Analysis”, 2015
[4] G. Revathy, K. Bhavana Raj, Anil Kumar, IEEE International Advance Computing
Spurthi Adibatti, Priyanka Dahiya, T.M. Conference (IACC), 2015.
Latha, Investigation of E-voting system [8] X. Zhao and C. Wei, “A real-time face recog-
using face recognition using convolutional nition system based on the improved LBPH
neural network (CNN), Theoretical Com- algorithm”, 2017 IEEE 2nd International
puter Science, 2022. Conference on Signal and Image Processing
(ICSIP), 2017.
88 Advancements in rainfall prediction
systems: A comprehensive review of
techniques and models
Adarsh Vishwakarma1, Ankit Khare2, Amit Verma1, and
Aditya Pandey1
1
Department of Computer Science and Engineering, Babu Banarsi Das Institute of Technology and
Management, Lucknow, India
2
Asssistant Professor, Department of Computer Science and Engineering, Babu Banarsi Das Institute
of Technology and Management, Lucknow, India
Abstract: This literature evaluation critically examines the advances in rainfall forecasting systems, offer-
ing comprehensive insights into strategies and fashions The look at of increasing excessive weather occasions
addresses the instantaneous want as rainfall of accurate and timely forecasting addresses and consists of syn-
thesizing current studies, figuring out gaps Having achieved so, by using critically examining the strengths
and limitations of the numerous strategies, research this is meant to significantly make a contribution to the
improvement of ongoing rainfall forecasting systems.
Keywords: Cloudburst prediction, Extreme precipitation, Meteorological data, Modeling techniques, Statistical
models, Physical models, Hybrid models, ML models, DL
1. Introduction advancements in rainfall prediction systems,
with a focus on the techniques and models
Rainfall prediction is a crucial aspect of used for predicting rainfall. The review draws
weather forecasting, with significant implica- on some primary sources and. Discusses the
tions for various sectors such as agriculture, use of machine learning and conventional
water resources management, and disaster methods for rainfall prediction, with a focus
risk reduction. Accurate and timely rainfall on Multiple Linear Regression, decision tree,
predictions can help farmers plan their crop and K-means clustering. Explores the causes
cycles, enable water managers to optimize and impacts of cloudbursts, a type of extreme
reservoir operations, and support emergency rainfall event, and proposes pre-disaster miti-
managers in preparing for and responding to gation measures with early warning systems.
floods and other weather-related disasters. Presents a study on predictive analysis for
Over the years, various techniques and mod- weather prediction using data mining with
els have been developed to predict rainfall, Artificial Neural Networks (ANN).
ranging from traditional statistical methods The significance of this literature review
to more advanced machine learning algo- lies in its potential to provide a comprehen-
rithms and numerical weather prediction sive understanding of the advancements in
models. rainfall prediction systems. By synthesizing
The objective of this literature review is the findings from these sources and other
to provide a comprehensive overview of the relevant studies, this review will explore the
DOI: 10.1201/9781003606635-88
Advancements in rainfall prediction systems: A comprehensive review of techniques and models 623
strengths and limitations of different tech- temperature gradients, relative humidity,
niques and models, as well as recent develop- specific humidity, dew points, wind speed,
ments and emerging trends in the field. The wind direction and shear respectively for
review will also discuss the potential applica- more robust and accurate predictions.
tions of these advancements in rainfall pre- 2. Dependent variables or Predictands:
diction systems, including their implications Temporal variables such as Time of Day/
for agriculture, water resources management, Season and historical rainfall over a
and disaster risk reduction. period of time.
Overall, this literature review will provide 3. Control Variables: Satellite Data (Cloud
a valuable resource for researchers, practition- cover, temperature, moisture content)
ers, and policymakers working in the areas and Radar Data (Reflectivity, rainfall
of meteorology, hydrology, water resources estimates)
management, and disaster risk reduction. By
synthesizing the latest research on rainfall 2.2. Effective variable relationships
prediction systems, this review will contrib- and how they impact on
ute to a better understanding of the state of
the art in this field and help identify areas for cloudburst prediction models
future research and development. Precipitable Water (PW): Elevated levels of
PW indicate high atmospheric water vapor
content, often signaling conditions favorable
2. Methodology for intense rainfall or cloudbursts. Incorpo-
We conducted a systematic review of the liter- rating PW data significantly aids ML models
ature published over the past decade, employ- in identifying potential cloudburst events [1].
ing reputable databases such as Elsevier, IEEE Convective Available Potential Energy
Xplore, and Google Scholar. Keywords such (CAPE): CAPE represents atmospheric insta-
as “rainfall prediction,” “machine learning,” bility and energy available for convective
and “ensemble learning” guided the search. processes, playing a critical role in predicting
Inclusion criteria encompassed studies uti- convective storms and potential cloudburst
lizing diverse ML and EL algorithms to pro-
vide a comprehensive understanding of the
field. The data consisted of past 15 years of
research papers used in development of mod-
els useful in predicting disasters caused due
to rain.
2.1. Parameters/Variables used in
modeling
1. Independent Variables or Predictors:
These predictors_included_ various other
Meteorological (Temperature, Humidity,
Pressure and wind variables) and Geo-
graphical variables (Latitude _and_ Lon-
gitude, Altitude/Elevation). The variables
such as Temperature, wind and humidity Figure 88.1. Correlation matrix of major AQ1:
have been further subdivided into variables Parameters. Please
like surface temperature, air temperature, Source: provide
AUTHOR'S SELF GENERATED source
for all the
figures
624 Innovations in Computational Intelligence: Trends and Technologies
occurrences. ML models leverage CAPE data In the past, several data-driven modeling
for accurate predictions [2]. approaches have been investigated to per-
Radar Reflectivity and Rainfall Rates: form such forecasting tasks as multilayer per-
Real-time radar data, especially spikes in ceptron neural networks (MLP-NN). In fact,
reflectivity or extreme rainfall rates, provides the rainfall time series modeling (SARIMA)
essential information for ML models to iden- involves important temporal dimensions. In
tify the intensity and likelihood of cloudburst order to evaluate the incomes of both models,
events accurately [3]. statistical parameters were used to make the
Atmospheric Instability Indicators: Param- comparison between the two models. These
eters related to temperature gradients, humid- parameters include the Root Mean Square
ity levels, and pressure changes serve as key Error RMSE, Mean Absolute Error MAE,
markers of atmospheric instability, guiding Coefficient of Correlation CC and BIAS [4].
ML models in predicting convective storms Till 2018, machine learning methods have
and potential cloudburst events [2]. significantly improved flood prediction accu-
Satellite Observations: Real-time satellite racy and efficiency, with hybridization, data
data on cloud cover, moisture content, and decomposition, algorithm ensemble, and
convective cloud development significantly model optimization being key strategies for
enhance ML models’ accuracy in predicting improving models [1].
cloudbursts by providing crucial insights into Later we found that The SVM model is
atmospheric conditions [4]. more efficient and accurate than the WRF
While these parameters are frequently and ANN models in short-term rainfall, air
being used in cloudburst prediction models, temperature, and flood forecasting [6].
achieving the highest accuracy often involved Among all regression techniques, we also
a combination of multiple variables and found that Lasso regression is the most accu-
sophisticated machine learning algorithms. rate machine learning model for predicting
The relationships and interactions between rainfall, improving agriculture, coastal plan-
these parameters, when properly modeled ning, and water scarcity management [5].
and integrated into predictive models, con- Till today, the ANN model is the most reli-
tribute significantly to enhancing the accu- able and accurate method for assessing flood
racy of cloudburst predictions [1]. hazard risk, aiding decision-makers in con-
Figure 88.1 shows Correlation between structing mitigation strategies [3].
major different Parameters used for Rainfall Within DL scope, we found out Wavelet-
prediction [1]. LSTM and convolutional LSTM models are
superior alternatives for monthly stream flow
2.3. ML models used in heavy and rainfall forecasting, outperforming multi-
layer perceptron and LSTM models [4].
rainfall prediction and flood
prediction
Rainfall is considered as one of the major
components of the hydrological process; it
takes significant part in evaluating drought
and flooding events. Therefore, it is impor-
tant to have an accurate model for rainfall
prediction [3].
MLP-NN and SARIMA models show
promise for accurate rainfall prediction, with Figure 88.2. Accuracy and time limitations of
MLP-NN showing better performance than ML models in cloudburst prediction.
SARIMA [4]. Source:
AUTHOR'S SELF GENERATED
Advancements in rainfall prediction systems: A comprehensive review of techniques and models 625
3. Critical Analysis a pivotal advancement. The review encom-
passes a diverse range of models, including
3.1. Strengths and limitations Multiple Linear Regression, decision tree, and
K-means clustering, along with the explora-
Integration of Cutting-Edge Technologies: tion of key parameters impacting cloudburst
The review effectively integrates traditional prediction.
and cutting-edge technologies, reflecting a
comprehensive understanding of the evolu-
tion in cloudburst prediction. 4.2. Best algorithms and models
Diversity of Models: The inclusion of Machine Learning Models: The literature
Multiple Linear Regression, decision tree, review highlights the significance of machine
K-means clustering, and machine learning learning models in improving cloudburst
models showcases a diverse set of method- prediction accuracy. Among the notable ML
ologies, contributing to a holistic view of the algorithms, Support Vector Machines (SVM)
field. and various boosting techniques— Gradient
Focus on Causes and Impacts: The incor- Boost, AdaBoost, and XGBoost—stand out.
poration of studies exploring causes and SVM excels in handling complex, nonlinear
impacts of cloudbursts adds depth to the relationships within meteorological varia-
analysis, addressing both prediction and miti- bles, providing robust predictive capabilities.
gation aspects. Boosting techniques, with their ability to
Depth of Machine Learning Exploration: sequentially refine weak learners, exhibit
While machine learning models are men- promising results, especially in capturing
tioned, the depth of exploration and critical intricate patterns associated with cloudburst
analysis of specific algorithms or techniques events.
is limited. A more detailed discussion on the Ensemble Learning: Within ensemble
strengths and weaknesses of each ML model learning, both Bagging and Boosting tech-
would enhance the critical analysis. niques demonstrate effectiveness in enhanc-
Scope of Remote Sensing and Data Ana- ing prediction outcomes. The combination of
lytics: The review mentions the integration multiple models, as seen in ensemble meth-
of remote sensing and data analytics without ods, contributes to increased robustness and
delving into specific applications, potentially generalizability, crucial for addressing the
leaving a gap in understanding their contri- dynamic nature of cloudburst occurrences.
butions and limitations. Machine Learning Model Interpretabil-
Figure 88.2 shows the Accuracy and time ity: The inherent complexity of advanced
limitations of using various ML models in ML models, such as SVM and boosting tech-
cloudburst prediction. niques, poses challenges in interpretability.
Enhancements in model interpretability are
vital for fostering trust among users and
4. Synthesis and Implications
stakeholders, especially in fields where deci-
sion-making based on predictions has real-
4.1. Synthesis of findings
world consequences.
The synthesis of findings from the literature Integration of Remote Sensing and Data
review reveals several noteworthy trends in Analytics: Although the review mentions the
the realm of cloudburst prediction. While tra- integration of remote sensing and data ana-
ditional meteorological approaches have laid lytics, a more detailed exploration of spe-
the foundation, the integration of cutting- cific applications and their effectiveness is
edge technologies, particularly machine learn- required. Future research should focus on
ing (ML) and data analytics, has emerged as refining the integration of these technologies,
626 Innovations in Computational Intelligence: Trends and Technologies
ensuring seamless collaboration and leverag- Representations from Transformers) or its
ing the full potential of available data sources variants, to cloudburst prediction could offer
Temporal and Spatial Resolution: new insights. These models excel in capturing
Improvements in temporal and spatial resolu- complex relationships within sequential data,
tion remain critical for cloudburst prediction making them potentially valuable for mete-
models. Fine-tuning the granularity of models orological time series analysis.
can lead to more accurate and timely predic- Hybrid Model with Explainable Com-
tions, especially in regions with diverse geo- ponents: Developing hybrid models that
graphical and meteorological characteristic. combines the strengths of machine learning
Figure 88.3 it shows a graph comparing the algorithms with interpretable components is
accuracy and time taken of different machine crucial. Techniques like SHAP (Shapley Addi-
learning models for rainfall prediction. The tive Explanations) or LIME (Local Inter-
models considered are Logistic Regression, pretable Model-agnostic Explanations) can
Decision Tree, Neural Network, Random be integrated to enhance the interpretability
Forest, LightGBM, CatBoost, and XGBoost. of complex models, addressing the need for
transparency in decision-making.
Integration of Satellite Constellations:
5. Recommendations and Future With the increasing deployment of satellite
Scope constellations, leveraging data from multi-
Building upon the current state of knowledge ple satellites simultaneously can enhance the
in cloudburst prediction, the literature review spatial and temporal resolution of cloudburst
suggests several recommendations for future prediction models. Future research should
research, focusing on the most recent tech- explore optimal ways to integrate data from
niques and ML models. these constellations to improve the accuracy
Exploration of Transformer Models: of predictions.
Recent advancements in natural language Dynamic Ensemble Learning Techniques:
processing have seen the success of trans- Continued exploration of ensemble learning
former models. Applying transformer archi- techniques, with a focus on dynamic ensem-
tectures, such as BERT (Bidirectional Encoder bles that can adapt to changing meteorological
conditions, is recommended. Models capable
of adjusting their composition in real-time
based on evolving data patterns can provide
more robust and adaptive predictions.
Future Scope for Alternative Models:
While machine learning has dominated
recent advancements, alternative models offer
untapped potential for cloudburst prediction.
The following areas present opportunities for
future research.
Physics-Based Models: Investigating the
integration of physics-based models, such as
computational fluid dynamics or numerical
weather prediction models, holds promise.
Combining the strengths of physics-based
Figure 88.3. Efficiency of ML algorithms for and data-driven approaches can lead to more
Cloudburst Predicitons. accurate predictions, especially in regions
Source:
AUTHOR'S SELF GENERATED
with complex terrain.
Advancements in rainfall prediction systems: A comprehensive review of techniques and models 627
Deep Reinforcement Learning (DRL): The techniques that are more efficient to obtain
application of deep reinforcement learning, predicted results with minor error between
which has shown success in dynamic deci- measured value and standard set.
sion-making processes, can be explored. DRL Therefore it is critical to understand the
models can adapt and optimize prediction challenges of these methods are to identify
strategies based on feedback from the envi- the appropriate technology, sensitivity of fac-
ronment, making them suitable for the evolv- tual processes, and dependence on substance
ing nature of cloudburst events. treatment. The primary goal of this research
Quantum Computing Applications: With is to examine the data to predict rainfall bet-
the emergence of quantum computing, explor- ter than current models. Its design should
ing its potential applications in meteorologi- be simple so that it can be easily tested and
cal modelling is an exciting avenue. Quantum verified.
algorithms could handle complex simulations
and computations more efficiently, opening
up new possibilities for accurate and rapid Refrences
cloudburst predictions. [1] Saranagata Kundua, Saroj Kr. Biswas, Deek-
Explainable AI in Physical Models: Incor- sha Tripathi, Rahul_Karmakar, Sounak
porating explainable AI techniques into Majumdar, Sudipta Mandal (2023) Rain-
physics-based models is crucial for enhanc- fall_forecasting_using ensemble learning
ing their interpretability. This ensures that techniques.
[2] Nafsika Antoniadou, Hjalte Jomo Dan-
insights gained from these models can be eas-
ielsen Sørup, Jonas Wied Pedersen, Ida Bu¨
ily communicated to stakeholders, facilitat- low Gregersen, Torben Schmith,Karsten
ing informed decision-making the future of Arnbjerg-Nielsen (2023) comparison of
cloudburst prediction research lies in a mul- data-driven methods for linking extreme
tidimensional approach that integrates the precipitation events to local and large-scale
latest machine learning models while explor- meteorological variables.
ing alternative strategies. The recommenda- [3] India Prabhash K. Mishra, Renoj J. Thayyen,
tions outlined above provide a roadmap for Hemant Singh, Swagatam Das, Manish K.
researchers to delve into cutting-edge tech- Nema, Pradeep Kumar (2023) assessment of
niques and alternative models, ensuring the cloudbursts, extreme rainfall and vulnerable
continuous evolution and improvement of regions in the Upper Ganga basin, Uttara-
cloudburst prediction systems. khand, India.
[4] Francesca Fallucchi, Riccardo Scano and
Ernesto William De Luca (2021) Machine
6. Conclusion learning models applied to weather series
analysis.
The work we are doing is focus toward devel- [5] Sivagami M., Radha P. and Balasundaram
oping a model that can predict the cloudburst (2021). Sequence Model based Cloudburst
events. We will use the dataset of cloudburst Prediciton.
event that has occur between 2010 to 2023. [6] Kavita Pabreja (2012) Clustering technique
Many researchers prediction has received for Interpretation of Cloudburst over Leh.
attention from previous researches the col- [7] Roberta Avanzato and Francesco Beritelli
lected data should be analyzed and trained (2020) An Innovative Acoustic Rain Gauge
adequately to be tested by collective learning Based on Convolutional Neural Networks.
89 FindServ
Mayank Srivastavaa, Pratibha Dixitb, Adarsh Vermac, Sadaf Alid
Department of Computer Engineering and Applications, GLA University, Mathura, India
Abstract: Identifying missing persons and returning them to their families has become a universal problem. A
variety of research publications are reviewed in this paper. Each existing mechanism has advantages and disad-
vantages. However, the issues related to repatriation of missing persons are still not 100% resolved. Computing
technology has evolved in recent years and includes many variations that can be used in virtually every field.
Every day, significant numbers of people go missing around the world, including children, young people, the
mentally ill, and the elderly with Alzheimer’s disease. It is estimated that over 500 missing person concerns go
un-resolved in India every day. Facial recognition technology has become increasingly important over the past
decades. A facial recognition system is a computer application that uses digital images or video frames from a
video source to recognize or verify a person’s identity. This technology finds widespread use across different
industries and scenarios. Artificial intelligence has played a significant role in advancing facial recognition tech-
nology and solving modern-day problems. The proposed mechanism for facial recognition has been successfully
implemented, achieving 90% accuracy compared to 59% using ANN and 43% using SVM with PCA. This
system uses image segmentation, PCA, KNN, feature extraction, and SVM.
Keywords: Decision tree, machine learning, random forest, support vectors
1. Introduction missing persons as easily and efficiently as
possible, and indirectly work for social pur-
FindServ is an application has been devel- poses. FindServ is completely free, empower-
oped to aid police and higher authorities in ing people who want to make a difference in
quickly tracking down missing people. The society. The database is updated each time a
traditional process of investigating a miss- new case is added and the case sits dormant
ing person requires time and experience to for his three years before being updated with
ask the right questions, but this application is new photos.
designed to streamline the process and expe- There are various issues and challenges
dite the search. Most of the time, investigation which are given as follows. One of the pri-
method works pretty well but. If the (missing) mary issues with facial recognition technol-
person has relocated/moved to another loca- ogy is still racial bias. Over 90% classification
tion (city/country), it may take longer and accuracy is guaranteed by face recognition
fail. In such cases, the ideal approach is to systems, yet these claims are not always accu-
review video footage and evidence from sur- rate. Age-related declines in visual sensitiv-
veillance cameras. ity may have an effect on facial recognition.
Again, this can take a very long time, and However, it is unclear how ageing affects
given the number of people going missing basic visual processing and the sensory and
every day, it can be difficult to keep up. In cognitive processes that underlie face identi-
India, about 7 million people go missing each fication. Numerous research has been done
year. reported to be the project is intended on how early in a life a baby learns to rec-
to help police and their families locate these ognize facial emotions. But there are still a
a
[email protected]; [email protected]; [email protected]; dsadaf.ali_cs19@
gla.ac.in
DOI: 10.1201/9781003606635-89
FindServ 629
lot of unanswered questions and no definitive stage. The testing process was divided into
findings. two parts, where the first part involved the
KNN method, and the second part involved
the double-distance method H. Li, Roivainen,
2. Related Work and Forchheimer (1993).
The principal component analysis (PCA) tech-
nique for feature extraction and K -Nearest
neighbors (KNN) for classification has been 3. Proposed Model
utilized. In the past, efforts were made to The goal of this paper is to make it easier for
develop a face recognition system that could the police and higher-ups to find missing peo-
perform multiple tasks. One of the experi- ple promptly. Investigation is typically used to
ments performed is described in the jour- search down a person, and it takes time and
nal article Mantoro, Ayu, et al. (2018). The expertise (to ask right questions). The investi-
proposed face recognition method utilizes a gative process generally yields positive results,
hybrid approach of Cascades and Eigenfaces, although it takes a lot of time and may fail if
enabling it to recognize 55 faces in a single the missing individual has been relocated to
recognition step. Various phases of image pre- a new place (city/country). Examining CCTV
processing, such as training data, grayscale footage and other supporting documentation
transformation, and preprocessing with Cas- is the best course of action in such situations.
cade, are employed to ensure accurate recog- Again, this can take a lot of time, and consid-
nition of faces Brunelli and Poggio (1993). ering how many individuals go missing every
Other research examples are listed in the day, it can be difficult to keep up with.
journal entitled Rachmawanto, Anarqi, Sari, Our proposed method (Figure 89.1)
et al. (2018) put the research forth a mul- for improving facial recognition accuracy
tistage recognition process that involved involves utilizing KNN and Face Encoding.
thresholding, denoising, and clipping prior To begin, the KNN algorithm is tested using
to feature extraction and classification. The
dataset was divided into training and test
data, with feature extraction being a crucial
step in the process. Methods used are Eccen-
tricity and Metric. The KNN method was
used for object classification based on the
closest distance to the training data, utilizing
the Euclidean distance formula in this study.
The paper reported an accuracy of 85.38%
for recognition using eccentricity and metric
feature extraction with KNN.
Another study proposed a new method for
improving recognition accuracy by in- cor-
porating outliers in the data, known as the
double distance method, which was com-
bined with the k=1 KNN method as the
center of speech recognition. The method
AQ1:
comprised of a training process with mel- Please
frequency-cestrum coefficients (MFCC) for provide
feature extraction and a testing process that Figure 89.1. Basic working flowchart of source
proposed model. for all the
utilized the KNN method in the introduction figures
Source:
AUTHOR'S SELF GENERATED
630 Innovations in Computational Intelligence: Trends and Technologies
a dataset of labelled facial images, with the vector machines (SVM), convolutional neural
Face Encoding being used to extract features networks (CNN), and principal component
from the gray scale image of different face analysis (PCA). Our research showed that
expression of the dataset. These features are our recommended strategy, which used KNN
then inputted into the KNN algorithm. Once and Face Encoding, performed better than the
the KNN model is trained, it can be used to other algorithms, obtaining an accuracy of
classify new facial images by identifying the 87.7%. SVM method accuracy was 85.8%,
K nearest neighbors and assigning the new CNN algorithm accuracy was 84.3%, and
image to the majority class. In order to assess PCA algorithm accuracy was 81.2%. Finally,
the efficacy of our suggested approach, we our proposed work shows encouraging out-
conducted tests utilizing random manual comes in improving the accuracy of facial
shots for training and testing Wiskott, Fel- identification using KNN and Face Encoding.
lous, Krüger, and Von Der Malsburg (2022). Future research could include experiment-
In conclusion, our proposed work presents ing with different K values, looking into the
promising results in enhancing the accuracy use of additional deep learning models for
of facial identification through the use of face encoding, and looking into the use of
KNN and Face Encoding. Future work may additional distance metrics. Face recognition
involve further experimentation with differ- technology is currently being used widely
ent K values, exploring the use of other deep for smart home security, gadget security, and
learning models for face encoding, and inves- other purposes. The biometric that will see
tigating additional distance metrics. Overall, the most growth over the next few years is
this approach to facial recognition has signifi- the face. Among many other types of biom-
cant potential for im- proving identification etrics, the use of the face for identification is
accuracy in various settings and revolution- seen to be successful and to pro-duce accurate
izing the security and surveillance industry P. results. Face recognition makes use of face
Li, Gou, and Yang (2017). characteristics for security. KNN is used in
this paper as the classification approach. The
KNN algorithm forecasts the value of a good
4. Implementation of Proposed instance value based on the neighborhood
Work classification. An instance-based learning
The model uses face encoding and KNN for group is part of KNN Zhao and Chellappa
facial recognition (Figure 89.2) and takes (1999).
manual shots as input and then the face the
features that were extracted from each image
in the dataset using Face Encoding are then
sent into the KNN algorithm (Figure 89.3).
Once the KNN model has been trained, it can
be used to categorize new facial images by
locating the K closest neighbours and placing
the new image in the class with the highest
percentage of members. We conducted trials
employing random manual shots for train-
ing and the remaining for testing in order
to assess the effectiveness of our suggested
strategy. We evaluated the performance of Figure 89.2. Training phase of face
our proposed method against several facial Identification.
recognition techniques, including support Source:
AUTHOR'S SELF GENERATED
FindServ 631
4.1. Variations in facial images detecting the face and accounting for lighting
variations, the performance of face recogni-
Recognizing human faces is a complex task
tion systems can be improved Kelly (1971).
in image recognition research due to the fact
that a human face is not just a three-dimen-
sional object, but also a soft body. Moreover, 4.2. K-nearest neighbor
facial photographs are commonly captured in KNN is a popular machine learning classifi-
natural settings, which can introduce various cation method. This algorithm is categorized
factors that make it difficult to accurately rec- as a supervised learning technique and can
ognize a face. These factors include changes perform both regression and classification
in lighting, facial expressions, and occlusion, tasks. Each class is taken to be a cluster, and
as well as interference from the background. each data point is taken to be a part of a clus-
As a consequence, the development of precise ter. Next, it is necessary to identify the cluster
and efficient face recognition systems remains centers and the data points that are associ-
a continuous research topic in this field. ated with each cluster. KNN needs to inde-
This means that both visual backgrounds pendently locate the centers of each cluster
and lighting situations can be very complex. since it performs an unsupervised classifica-
Various facial variants can be grouped into the tion. The process of assigning a data point to
following categories: Complex backgrounds, a cluster involves checking whether it belongs
complex lighting, transformations, rotations, to a label of the point remains consistent with
scaling and occlusion, facial emotions, cam- the previous round. This technique is widely
era sounds, makeup and hairstyles. Noise and used in clustering algorithms to group similar
camera distortion are common variations of data points together Dino and Abdulrazzaq
image recognition problems. Prior research- (2019); P. Li et al. (2017). The time com-
ers have devised numerous methods to plexity of the KNN Algorithm is calculated
enhance the signal-to-noise ratio, and for pre- based on the number of training examples
cise facial recognition, a reliable face detec- and the number of used features. The KNN
tor is crucial for differentiating the real face sample case image is particular cluster or not,
from the surrounding image., especially when and repeating the process until the shown in
dealing with complex image backgrounds. Figure 89.4.
Lighting is a common factor that can inhibit
the identification process. It is of great impor-
tance to create face recognition systems that
can accommodate changes in lighting condi-
tions to ensure accurate results. By accurately
Figure 89.3. Testing phase of face Identification. Figure 89.4. KNN sample case image.
Source: Source: AUTHOR'S SELF GENERATED
AUTHOR'S SELF GENERATED
632 Innovations in Computational Intelligence: Trends and Technologies
4.3. Principal component analysis 4.5. Decision tree
The process of representing face images in a A commonly used machine learning algo-
lower dimensional face space by using eigen- rithm for regression and classification tasks is
vectors associated with larger eigenvalues. the decision tree. This algorithm constructs a
PCA is used in face recognition to identify model resembling a tree that shows decisions
the principal components of a given batch of and their possible outcomes, with the root
photos. The process of applying PCA to an node being the starting point. Each branch
image involves identifying and extracting the corresponds to a specific attribute or feature
most important features of the image, which of the data, and each internal node represents
are then represented as principal compo- a particular decision or rule based on that
nents. These components can then be used to feature Wiskott et al. (2022). The terminal
reconstruct a lower- dimensional version of nodes of the tree represent the final outcome
the image. Although the resulting size of the or classification for a given set of input data.
PCA image is smaller compared to the origi- The recursive process of dividing the data into
nal image, it still retains the essential com- smaller subsets based on the most important
ponents of the image. Additionally, it might features is the mechanism used by the deci-
massively speed up recognition Sohail and sion tree algorithm until a stopping condi-
Bhattacharya (2007). The PCA sample case tion is met, such as a maximum tree depth, a
image is shown in Figure 89.5. minimum number of instances per leaf, or a
specific level of information gain.
4.4. Random forest Information gain quantifies the decrease in
uncertainty of a dataset that arises from split-
Leo Breiman and Adele Cutler are the crea- ting it on a certain feature. Random Forest,
tors of the popular machine learning method an extension of the decision tree algorithm,
known as random forest, which combines the constructs an ensemble of Random Forest is a
results of numerous decision trees to generate machine learning technique that employs mul-
a single output and is widely used due to its tiple decision trees. To enhance the model’s
adaptability and versatility in solving classifi- overall performance and generalization, each
cation and regression problems. tree in the forest is trained on a random subset
of the data and a random subset of features,
which helps to reduce overfitting. The final
prediction is made by aggregating the predic-
tions of all the individual trees in the forest.
Both classification and regression issues can
be solved using the widely used non-paramet-
ric machine learning Decision Tree method-
ology. It has a root node, branches, internal
nodes, and leaf nodes as part of a hierarchical
structure. Decision trees offer various advan-
tages, such as producing clear rules, handling
both continuous and categorical variables,
and identifying crucial fields for classification
or prediction. Decision trees are user-friendly,
scalable, and can handle datasets with miss-
ing values. However, decision trees also have
Figure 89.5. PCA sample case image. some disadvantages. They are less suitable
Source: for estimating continuous characteristics,
AUTHOR'S SELF GENERATED
FindServ 633
classification issues with several classes and
few training samples are prone to errors, and
computationally expensive. The growth of
a decision tree requires sorting each candi-
date splitting field at each node, determining
the optimal split, and searching for the best
combining weights for some algorithms Kelly
(1971).
5. Result Analysis
Figure 89.6. Sample image.
Here, we have provided the result on multiple
Source:
conditions mentioned in the Table 89.1 based AUTHOR'S SELF GENERATED
on the Sample image displayed in Figure 89.6.
Table 89.1. Result of model based on features
Test case description Expected output Actual output Features Test status
similarity match (P/F)
Images with front profile Face recognized Face recognized <90% P
Images with side profile Face recognized Face recognized <75% P
Images with illumination Face recognized Face recognized <65% P
difference
Images with various Face recognized Face recognized <60% P
background
Images with age difference Face recognized Face recognized <45% P
Images with accessories Face recognized Face recognized <65% P
Source:
AUTHOR'S SELF GENERATED
6. Conclusion References
The facial recognition technique is challeng- [1] Brunelli, R., and Poggio, T. (1993). Face rec-
ing but crucial. The facial recognition method ognition: Features versus templates. IEEE
has clear significant advantage over all other transactions on pattern analysis and machine
biometric technologies. This paper pro- intelligence, 15 (10), 1042–1052.
posed a KNN and PCA-based framework for [2] Dino, H. I., and Abdulrazzaq, M. B. (2019).
Facial expression classification based on
enhanced human face recognition. A training
svm, knn and mlp classifiers. In 2019 inter-
data set with discriminative patterns based on national conference on advanced science
correlations between the first training images and engineering (icoase) (pp. 70–75).
was helped by KNN and PCA. The recently [3] Kelly, M. D. (1971). Visual identification of
acquired Tabular dataset enabled KNN to people by computer. Stanford University.
converge more quickly and accurately. The [4] Li, H., Roivainen, P., and Forchheimer, R.
system was improved by incorporating vari- (1993). 3-d motion estimation in model-
ous distance algorithms, each demonstrating based facial image coding. IEEE transactions
better performance than the others in enhanc- on pattern analysis and machine intelligence,
ing the overall system. 15 (6), 545– 555.
634 Innovations in Computational Intelligence: Trends and Technologies
[5] Li, P., Gou, J., and Yang, H. (2017). The dis- [8] Sohail, A. S. M., and Bhattacharya, P. (2007).
tance-weighted k-nearest centroid neighbor Classification of facial expressions using
classification. J. Inf. Hiding Multim. Signal k-nearest neighbor classifier. In Computer
Process, 8 (3), 611–622. vision/computer graphics collaboration
[6] Mantoro, T., Ayu, M. A., et al. (2018). techniques: Third international conference,
Multi-faces recognition process using haar mirage 2007, rocquencourt, france, march
cascades and eigenface methods. In 2018 28-30, 2007. proceedings 3 (pp. 555–566).
6th international conference on multimedia [9] Wiskott, L., Fellous, J.-M., Kru¨ger, N., and
computing and systems (icmcs) (pp. 1–5). Von Der Malsburg, C. (2022). Face recog-
[7] Rachmawanto, E. H., Anarqi, G. R., Sari, C. nition by elastic bunch graph matching. In
A., et al. (2018). Handwriting recognition Intelligent biometric techniques in finger-
using eccentricity and metric feature extrac- print and face recognition (pp. 355–396).
tion based on k-nearest neighbors. In 2018 Routledge.
international seminar on application for [10] Zhao, W., and Chellappa, R. (1999). Robust
technology of information and communica- face recognition using symmetric shape-
tion (pp. 411–416). from-shading. Citeseer.
90 IoT devices security against common
threats
Mayank Srivastavaa and Prabudhb
Department of Computer Engineering and Applications, GLA University, Mathura, India
Abstract: Internet of Things (IoT) gadgets are subject to a number of protection vulnerabilities and attacks
because they are resource-constrained and diverse devices are networked to deliver multiple application tech-
nology. The functional design of a remote security management server is introduced in this project to improve
the protection and safe guard of IoT gadgets in the IoT ecosystem. The remote security management server inte-
grates and controls a variety of security functions in a systematic and integrated manner. As a result, numerous
infringement instances that may occur in the IoT environment may be avoided ahead of time, and harm can be
mitigated by enabling swift and effective responses even if a significant attack happens. We are using Raspberry
Pi for this project. For now, we are focusing on network security of interconnected-IoT devices from outer net-
work attacks such as “Man-In-The-Middle” attack and “Distributed Denial of service (DDoS)” attack. We are
further going to test our Openwrt firmware using Hping-active Network Security Tool.
Keywords: Countermeasures, DDoS, IoT devices, infringement, resource-constrained devices, remote security
1. Introduction machine learning, has developed the field
Greenwald et al. (1996). In the mainstream
The Internet of Things (IoT) describes the tan- merchandise, IoT devices are more closely
gible devices (or collection of these devices) connected along items which support the
embedded with detectors, calculating abil- term smart home like lighting fixtures, reg-
ity, programs, and variant techniques that ulator, house safety functions and cctv, and
link to and interchange information between several house gadgets which support distinct
gadgets and desktops over the Online plat- ordinary atmosphere and may be controlled
form or different interactions infrastructure by gadgets relevant to that atmosphere,
E. S. Al-Shaer and Hamed (2004); Green- like smartphones and speakers. The IoTs is
wald, Singhal, Stone, and Cheriton (1996); deployed in medical as well Yue, Chen, and
Lyu and Lau (2000); Schuba and Spafford Wang (2009).
(1997). The word “internet of things” has Here are numerous misgivings related to
been considered a misconcep-tion because potential risks associated with advancement
devices are not required to be connected to of IoT equipment’s and product lines, par-
the open n etwork; rather than they must be ticularly in sectors of security and privacy.
linked to a network and located separately As a result, company and government inter-
Acharya, Wang, Ge, Znati, and Greenberg ventions to tackle those problems has begun,
(2006); E. Al-Shaer, Hamed, Boutaba, and involving the formation of global and regional
Hasan (2005). regulations, rules, and official requirements
The integration of several techniques, like Abie and Balasingham (2012).
ubiquitous calculating, product detectors, Cameras detect potential hazards, which
more sophisticated integrated machines, and are then assessed in real time by specialized
[email protected]; [email protected]
a
DOI: 10.1201/9781003606635-90
636 Innovations in Computational Intelligence: Trends and Technologies
algorithms using predetermined standards. more computing capability, memory-space,
Once the program has detected an occur- and capacities over the IoT gadgets itself, giv-
rence, a security professional at a regional ing permission to build functionality such as
centre is informed to investigate the danger firewalls to intercept attackers from entering
and take necessary action to eliminate or the IoT gadgets that they connect E. Al-Shaer
limit harm that could have happened Lyu and et al. (2005). It is important to upgrade gadg-
Lau (2000). A distress signaling to the wrong- ets as well as applications, either through
doer through wakie-takie from the facility or internet linkages or self-operation. It is also
disclosing the incident to police authorities explanatory to have a synchronized exposure
are examples of these responses E. S. Al-Shaer for weaknesses or flaws in case gadgets might
and Hamed (2004). be upgraded after immediate accessible. In
By having professional personnel analyze addition, imagine regarding end-of-life alter-
each warning prior to taking action, remote natives Schuba and Spafford (1997).
guarding provides an extra security of manual
identification to self-operating protection func- 2.3. Training
tions, lowering false alarms to practically zero
percent. This system, combined with modifica- Several traditional safeguard groups are
tions in the alarm industry’s handling of signals, unknown to IoT and OS protection. Safe-
like succeeding call confirmation, is helping guard professionals should acknowledge to
local law enforcement minimize the number of modern or unfamiliar devices, acknowledge
false alarms Malik, Verma, and Pal (2012). modern infrastructures and programing dia-
lect, and get ready for modern safeguard
problems E. S. Al-Shaer and Hamed (2004).
2. Review of Literature In order to have knowledge of present-time
In this section, we will discuss the several dangers and protection strategy, C-level direc-
security methods for IoT devices that are cur- tors and safeguard professionals shall receive
rently available. recurring cybersecurity coaching Acharya et
al. (2006).
2.1. Network access control (NAC)
and segmentation 2.4. Integrating teams and consumer
education
NAC could accommodate in the authentication
and catalogue of IoT gadgets which links to the Combination of diversified and often
internet. This would provide like a foundation archived groups, besides coaching, may be
of gadget surveillance and managing Green- advantageous Z.-K. Zhang et al. (2014). For
wald et al. (1996). IoT gadgets which require example, collaborating with security experts
quickest cyberspace linkage shall be segregated throughout the development process might
in their personal cyberspace, along with entry help in assurance that the appropriate rules
in the work cyberspace restricted. Exceptional are established to gadgets Abie and Balasing-
occurrence shall be overseen by workspace divi- ham (2012). Customers should be enlightened
sions as a result of which steps might be taken regarding the risks of IoT equipment’s and
if the distress is identified Lyu and Lau (2000). stated commands on ways to maintain the
safety, like replace predefined passwords and
doing application upgrades Yue et al. (2009).
2.2. Security gateways and patch Customers can as well perform a significant
management part at persuading gadget manufacturers to
Protection accessions, that works like a flyover create safe devices and reject to utilize which
connecting IoT gadgets and the internet, has does not provide it Malik et al. (2012).
IoT devices security against common threats 637
2.5. Palo Alto networks performed by the Mirai botnet proved. This
botnet was created to search the Internet
Using Palo Alto Networks’ IoT security solu-
for devices with inadequate security, such as
tion, offices, data centres, and other facilities
surveillance cameras, and get access to them
with IoT networks can protect their devices
by trying passwords as easy as “admin” or
from antagonistic traffic and potential intru-
”12345.” Mirai was able to acquire around
sions. It bills itself as a turn-key solution that
400,000 machines after doing more over 60
combines hardware and virtualized firewalls,
checks of login and password combinations
device identification utilities, and other tool-
E. Al-Shaer et al. (2005); Greenwald et al.
kits for transparency and security of unman-
(1996) If device makers forced customers to
aged IoT devices J. Zhang, Chen, Gong, Cao,
update their gadgets’ default passwords, this
and Gu (2019).
attack may have been averted Schuba and
Spafford (1997). As a result, corporate execu-
2.6. First point tives must be cautious that IoT devices like
FirstPoint is a device-agnostic automat- security cameras and thermostats might be
ing routine tasks security mechanism that used as attack vectors. Although the device
protects cellular IoT devices against threats does not contain any sensitive information, it
that originate within the cellular connection is most likely linked to a network that has
or communicate over the cellular network access to the company’s confidential data E.
Mahmoud et al. (2015). This assists mobile S. Al-Shaer and Hamed (2004); Yue et al.
network operators in protecting vulnerable (2009).
cellular-connected devices against the par-
ticular dangers that IoT devices encounter 3.2. Remote SSH access to IoT
on cellular networks, such as private LTE devices
and 5G networks Razzaq, Gill, Qureshi, and
Ullah (2017). All of the above given points The main reason you set up those IoT devices
are taken from various literature which are and linked them to the world wide web was
important keeping the factor of IoT security. to be able to monitor, track, and control them
from any place. For troubleshooting, con-
figuration upgrades, and other operational
3. Proposed Methodology duties, you may require access to those IoT
In this section we are going to discuss the dif- devices Abie and Balasingham (2012). A sen-
ferent methodologies that we have used for sor device placed hundreds of kilometers
remote security management server for IoT distant, for example, is having problems
devices: monitoring the manufacturing temperature.
3.1. Basic configuration settings
change
Despite the many benefits that the IoTs will
provide, security administrators will have to
monitor hundreds of possible vulnerabili-
ties Lyu and Lau (2000); Schuba and Spaf- AQ1:
Please
ford (1997). Techniques for remote security provide
management is shown in Figure 90.1. Keep- Figure 90.1. Techniques for remote security source
management. for all the
ing these devices’ default settings is also a sig- figures
nificant error, as the enormous DDoS assault Source:
AUTHOR'S SELF GENERATED
638 Innovations in Computational Intelligence: Trends and Technologies
To open and rapidly start a session with that simple to operate. It is the exact firewall that
sensor device, you can utilize secure remote is included in all Ubuntu releases. Firewalls
access tunnels Abie and Balasingham (2012). such as UFW are critical for preventing port-
You can reset the setup, erase logs history, based network assaults on devices like the
and relaunch the sensor device through that Raspberry Pi Mahmoud et al. (2015). Fire-
kind of session after you’ve detected the walls keep an eye on this data (delivered in
problems (for example, a misconfiguration or “packets”) to see if it’s safe. This is accom-
disc full error) Ma- lik et al. (2012). Tradi- plished by determining if the packets comply
tionally, troubleshooting would include send- with the established rules Mahmoud et al.
ing a professional to the plant the next day (2015). Data packets are allowed or refused
to inspect the sensor device. However, remote based on these rules.
access via secure tunnelling (via SocketXP) A simple built-in firewall is included
reduces issue response and recovery times as in most operating systems (computer and
well as operating expenses. However, acquir- mobile). Using a third-party firewall solution,
ing remotely access to IoT gadgets isn’t easy. on the other hand, frequently provides greater
To allow internet activity into the business control choices and more accurate outcomes.
network, users frequently take shortcuts and These can be purchased individually or as
execute rapid hacks on router/firewall set- component of a security package Tanaka,
tings Malik et al. (2012). Fujishima, Mimura, Ohashi, and Tanaka
(2016). Various automated technologies in
3.3. Automatic updation feature firewall programmes utilise whitelisting to
determine which apps should be allowed or
One of the biggest steps you can take to ensure denied internet access. Most people would
your security online is to keep the software find this to be even more time intensive to
on your devices up to date and install any execute manually Tanaka et al. (2016).
software updates that come out Z.K. Zhang
et al. (2014). This advice doesn’t just apply
to your computers and phones, but also IoTs 3.5. Brute force detection system
devices like a Raspberry Pi or BeagleBone A brute force attack includes predicting sign-
Black. Luckily this is an easy process with the in credentials, cipher keys, and finding a con-
help of one important software package. If cealed information by try and test method.
you’re running Raspberry Pi-OS, your firm- Attackers attempt all possible groups in the
ware options are limited to what’s offered desire of producing the correct combination
by the Raspberry Pi Foundation. Using this Tanaka et al. (2016). IT managers shall force
process, you’ll be updating the operating sys- precise password regulations with smallest
tem as well as the firmware to the latest stable size and difficulty standard in order to ensure
releases Mahmoud et al. (2015); Z.-K. Zhang passwords are rather challenging to discover.
et al. (2014). However feasible, multifactor identification
should be enabled Razzaq et al. (2017).
3.4. Firewall to limit network traffic Web Applications can utilize captchas to
prevent self- operating brute-force threats
Iptables is included with the Raspberry Pi OS Razzaq et al. (2017). Brute force threats to
installation by default, however it can be dif- RDP are often because it doesn’t record
ficult to set up and use. To get around this, unsuccessful sign-in trials. Some Live Reposi-
we’ll use UFW, a simpler firewall interface, tory and information regulations are rather
to install on our Raspberry Pi Mahmoud et weak to brute force assaults Radovici, Cris-
al. (2015). Uncomplicated firewall (UFW) is tian, and Serban (2018).
a firewall that is supposed to be extremely
IoT devices security against common threats 639
3.6. Turn-off unnecessary services time data that means it should be updated
regularly in order to function and give best
When it comes to computers, the term “ser-
output to the users properly and correctly. But
vices” refers to programs that monitor and
it is a hectic work to update regularly manu-
response to network data. Other providers
ally without any mistake, in order to achieve
give you full control over your machine. Pro-
this task, we need automatic update as soon
viding people unrestricted entry to our system
as a new update is available.
and documents J. Zhang et al. (2019). With-
There are various and unnecessary servers
out being at competing directly with Distant
are active through which your raspberry pi is
Shielding, Robert H. Perry and his Associates
connected to the internet which can be used by
says Covenant Private Protection firms are
the hackers to gain access to Iot devices and
forming electric protection departments and
enter into your network. So, you need to turn
partnering along firms which specialise at this
those off unnecessary servers. Firewall is one
particular field. Clients are switching safety
of the most important feature that any net-
people with electric protection and expanding
work related device should have installed and
protection scope with combining protection
activated. As IoT devices does not have its own
professionals or electric protection equip-
firewall to protect it from attacks like DDoS,
ment, according to the survey Mahmoud,
we will setup our own firewall with our rules
Yousuf, Aloul, and Zualkernan (2015).
to allow the network traffic. The first step is
The IoT Security is needed because in our
to install ufw firewall as in Figure 90.4. The
day to day life we are using many IoT devices
second step is to decide what type of traffic
and those devices must not become a weapon
would be allowed through our firewall. We
for the intrusion of our devices or network.
This paper focuses on developing secure IoT
system in which intrusion cannot take place.
4. Experimental Results
Figure 90.2 shows the security server con-
figuration for experimental work. One of the
basic rule to protect your account or data or
information from an unauthorized access is
to protect your user id and password against
brute force attack which is the most common
attack used by any attacker and to do so we
have to change our password from default
one or change your password on a regular
basis as shown in Figure 90.3. Also, to make
your account more secure, we should create
a new account instead of using the default
one. We should make our new account admin
and specify your privileges according to your
necessity. Secure shell (SSH) is a private net-
work protocol for information exchange
that is particularly helpful for command-line
remote operations and it is one of the main
features of IoT devices. So, enabling it is a Figure 90.2. Security Server Diagram.
must in any case. IoT devices work on real Source:
AUTHOR'S SELF GENERATED
640 Innovations in Computational Intelligence: Trends and Technologies
can traffic through any port and in this way we 5. Conclusion
will personalize our firewall according to our
need. Initially, the firewall is inactive and we IoT will soon revolutionize the manner we
will activate it but only after we have applied live, work, travel, and more as it steadily
our own set of rules to the firewall. Now that becomes mainstream, with businesses like
we have applied our rules and configured our Google, Cisco, IBM, Intel, and several oth-
firewall, so we will now enable/activate the ers pioneering the charge. While the IoTs will
firewall. And also we can check what type of provide many benefits for manufacturers and
traffic is allowed as in Figure 90.5. consumers, it will also pose significant secu-
rity dangers. As more gadgets become con-
nected, protecting them all will become the
most difficult task. To ensure the functional-
ity of IoT devices efficiently, it would need
safe gadgets, application, and transmission.
Linked devices, from fridge to industry AI,
may be attacked if IoT protection is absent.
Attackers could take over the device’s work-
ing and retrieve the person’s virtual informa-
tion, if it get access. Securing Industrial IoT
should also not be overlooked.
References
Figure 90.3. Changing Default Settings.
[1] Abie, H., and Balasingham, I. (2012). Risk-
Source: based adaptive security for smart iot in
AUTHOR'S SELF GENERATED
ehealth. In Proceedings of the 7th interna-
tional conference on body area networks
(pp. 269–275).
[2] Acharya, S., Wang, J., Ge, Z., Znati, T. F.,
and Greenberg, (2006). Traffic-aware fire-
wall optimization strategies. In 2006 ieee
international conference on communica-
tions (Vol. 5, pp. 2225–2230).
[3] Al-Shaer, E., Hamed, H., Boutaba, R., and
Hasan, M. (2005). Conflict classification
and analysis of distributed firewall policies.
IEEE journal on selected areas in communi-
Figure 90.4. Downloading Firewall. cations, 23 (10), 2069–2084.
Source: AUTHOR'S SELF GENERATED [4] Al-Shaer, E. S., and Hamed, H. H. (2004).
Modeling and management of firewall poli-
cies. IEEE Transactions on network and ser-
vice management, 1 (1), 2–10.
[5] Greenwald, M., Singhal, S. K., Stone, J. R.,
and Cheriton, D. R. (1996). Designing an
aca- demic firewall: Policy, practice, and
experience with surf. In Proceedings of inter-
net society symposium on network and dis-
tributed systems security (pp. 79–92).
[6] Lyu, M. R., and Lau, L. K. (2000). Firewall
Figure 90.5. Activating Firewall.
security: Policies, testing and performance
Source:
AUTHOR'S SELF GENERATED
IoT devices security against common threats 641
evalua- tion. In Proceedings 24th annual internet of things (iot): A comprehensive
international computer software and appli- study. International Journal of Advanced
cations confer- ence. compsac2000 (pp. Computer Science and Applications, 8 (6).
116–121). [11] Schuba, C. L., and Spafford, E. H. (1997). A
[7] Mahmoud, R., Yousuf, T., Aloul, F., and reference model for firewall technology. In
Zualkernan, I. (2015). iot security: Current Proceedings 13th annual computer security
status, challenges and prospective measures. applications conference (pp. 133–145).
In 2015 10th international conference for [12] Tanaka, S., Fujishima, K., Mimura, N.,
internet technology and secured transactions Ohashi, T., and Tanaka, M. (2016). Iot sys-
(icitst) (pp. 336–341). tem security issues and solution approaches.
[8] Malik, A., Verma, H. K., and Pal, R. (2012). Hitachi Review, 65 (8), 359.
Impact of firewall and vpn for securing wlan, [13] Yue, X., Chen, W., and Wang, Y. (2009). The
International Journal of Advanced Research research of firewall technology in computer
in Computer Science and Software Engineer- net- work security. In 2009 asia-pacific con-
ing, 2 (5), pp 407–410. ference on computational intelligence and
[9] Radovici, A., Cristian, R., and S¸ERBAN, R. industrial applications (paciia) (Vol. 2, pp.
(2018). A survey of iot security threats and 421–424).
solutions. In 2018 17th roedunet confer- [14] Zhang, J., Chen, H., Gong, L., Cao, J., and
ence: networking in education and research Gu, Z. (2019). The current research of iot
(roedunet) (pp. 1–5). security. In 2019 ieee fourth international
[10] Razzaq, M. A., Gill, S. H., Qureshi, M. A., conference on data science in cyberspace
and Ullah, S. (2017). Security issues in the (dsc) (pp. 346–353).
91 Predict-a-disease: Survey
Mayank Srivastavaa, Aditya Singh Chauhanb, and Abhinav Bhardwajc
Department of Computer Engineering and Applications, GLA University, Mathura, India
Abstract: The discipline of bio-sciences has advanced to a more noteworthy level, and electronic health records
have made immense volumes of information. Thus, there is a squeezing need to produce data from this mon-
strous measure of information in this area of bio-sciences. Persistent kidney disease (CKD) is a problem wherein
the kidneys become harmed and can never again channel blood as proficiently as they once did. CKD can be
brought about by a family background of renal illness, over the top circulatory strain, or type 2 diabetes. This
is a drawn-out kidney issue with a high gamble of deteriorating over the long run. Heart issues, iron deficiency,
and bone diseases are genuinely predominant outcomes of kidney disappointment. Potassium and calcium levels
are raised. In the worst imaginable situation, absolute renal disappointment happens, surviving requires a kid-
ney relocate. Quick distinguishing proof of CKD can upgrade one’s personal satisfaction fundamentally. This
requires the improvement of a dependable expectation framework to distinguish CKD at a beginning phase. A
wide scope of AI methods have been utilized to foresee CKD. This review predicts CKD utilizing information
readiness, information change, and a few classifiers, as well as proposing the ideal CKD expectation structure.
The system’s demonstrate that Random Forest Classifier ut-performs other classifiers for CKD forecast.
Keywords: Chronic kidney disease, decision tree support vectors, machine learning, random forest
1. Introduction calcium deficit. Women with cystic ovaries
who have CKD may potentially experience
Ongoing kidney disease happens when the
an unexpected illness or an allergic reaction
AQ1: As kidneys can’t play out their standard blood
per style, to certain drugs.
separating capacity. Over time, kidney cells
the refer- By and large, CKD requires long-last-
ence list undergo a constant process of degeneration,
and its ing kidney dialysis transplantation. A fam-
which is represented by the term “ongoing”.
citation
This is an extreme sort of renal disappoint- ily background of renal illness expands the
should
be please ment wherein the kidneys quit separating gamble of creating CKD. As indicated by
check and
blood and a lot of liquid develops in the body. re- search, around one in each three patients
amend
accord- This caused the body’s levels of calcium and determined to have diabetes has CKD. The
ingly potassium salts to be abnormally high. Eleva- paper demonstrates the way that CKD Iden-
tions of these salts lead to a host of health tification and treatment can further develop
problems in the body. a patient’s personal satisfaction Machine
The primary objective of kidneys is to elim- learning expectation calculations might be
inate extra water and waste from the blood. utilized cleverly to expect the start of CKD
The equilibrium of salts and minerals in our and different illnesses. Give a method of early
bodies depends on this mechanism function- treatment. A thorough assessment of the writ-
ing properly. Salt balance is needed, among ing exhibits the utilization of numerous AI
other things, to produce red blood cells, acti- methods to foresee CKD. This examination
vate hormones, and control blood pressure. endeavors to anticipate CKD using classifiers,
Numerous bone disorders can result from a as well as suggesting the ideal forecast model.
[email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-91
Predict-a-disease: Survey 643
2. Literature Review (ESRD). In the most ideal situation, early dis-
tinguishing proof of CKD and suitable nour-
Wickramasinghe, Perera, and ishing treatment can slow or try and stop the
Kahandawaarachchi (2017) propose an eat- movement of the sickness. Consolidating AI
ing routine based strategy for treating the calculations with prescient examination, as
ailment. Several techniques, such as Mul- indicated by Aljaaf et al. (2018) is a savvy
ticlass Decision Forests, Multiclass Neural elective for affliction early forecast.
Networks etc, are used to build classifiers in Gathering strategies, for example, Boost-
the proposed paper. Taking into account the ing are utilized in information mining models
patient’s blood potassium levels, an appropri- to work on a model’s forecast. To think about
ate potassium zone is established. An eating the exhibition of order calculations, AdaBoost
regimen area is suggested by order calcula- and Logit Boost are usually used. Ripon et al.
tions, as evidenced by the predicted potas- (2019) took a gander at the adequacy of help-
sium zone. ing calculations for distinguishing CKD and
To estimate chronic kidney disease, highlights the relationship between the differ-
Wibawa, Malik, and Bahtiar (2018) planned ent CKD. To foster guidelines, the article uti-
and dissected Extreme Learning Machine lized the Ant-Miner AI technique related to a
utilizing Kernels (ELM). RBF-ELM, Linear- Decision Tree.
ELM, Polynomial-ELM, and Wavelet-ELM Data mining strategies are used to pro-
are four portion based ELMs whose pres- duce judgments from chronic illness datasets
entation is contrasted with that of ordinary by eliciting hidden information. It is neces-
ELMs. The awareness and particularity of the sary to store and manipulate large amounts
different strategies were thought about. The of data, as the importance of big data in this
RBF-ELM (Radial Basis Function - Extreme regard cannot be overstated. In order to fore-
Learning Machine) has better expectation cast chronic renal illness, Kaur and Sharma
rates. (2017) employed a number of data mining
Hypertension, diabetes, dyslipidemia, techniques in the Hadoop environment. Clas-
and metabolic disorder are cardiovascular sifiers like KNN (K-Nearest Neighbor) and
sickness risk factors (CVD). End-stage renal SVM (Support Vector Machine) are used in
infection (ESRD) is a sort of CKD for which the study.
there is no fix. In addition to recovering activ- In patients on dialysis who are turn-
ity rules based on stages, Dulhare and Ayesha ing out to be more terrible, blood levels of
(2016) also used the property selector and creatinine, salt, and urea are significant in
nave bayes to predict CKD, which helps pre- foreseeing endurance or the requirement for
vent ongoing renal sickness from moving to kidney transplantation. Ravindra, Sriraam,
the next stage. The middle endurance length and Geetha (2014) utilized a straightforward
for past due-stage patients is assessed to asso- K-implies calculation to extricate information
ciate with three years. Assessing a patient’s on the connections be- tween’s various CKD
exact condition is essential since it can help markers and patient endurance. The author
determine what kind of care, medications, or got to the end that the grouping technique
treatments are needed. These factors interact effectively figures dialysis patients’ endurance
in a complex way and affect the patient’s final time.
outcome. The effectiveness of ANN models in The exactness, accuracy, and execution of
predicting endurance in patients with CKD various classifiers which includes Naive Bayes,
was investigated by Zhang, Hung, Chu, Chiu, KNN etc. for CKD forecast were looked at
and Tang (2018). by Devika, Avilala, and Subramaniyaswamy
An End Stage Renal Disease patient has (2019). Panwong and Iam-On (2016) uti-
two options: dialysis or kidney relocation lized choice trees, K-closest neighbor, Nave
644 Innovations in Computational Intelligence: Trends and Technologies
Bayes, and counterfeit brain organizations On the CKD datasets, the examination
to remove data and build a characterization looks to foster a datamining worldview for
model for foreseeing temporary times of Kid- information revelation. Countless CKD data-
ney sickness stages 3 to 5 utilizing the neces- sets are assembled. The typical information
sary arrangement of information. mining approaches of information arrange-
Vijayarani, Dhayanand, and Phil (2015) ment and preprocessing are utilized. To fore-
utilized SVM to foresee renal problems. The see the beginning stage of CKD, below three
review analyzed the exactness and execution AI techniques are utilized. Every calcula-
season of the two techniques referenced pre- tion’s viability is assessed. The methodology
viously. Misir, Mitra, and Samanta (2017) beneath produces a model with extraordi-
utilized highlight determination calculations nary accuracy.
to find an assortment of attributes that may
precisely foresee renal sickness. 3.1. Decision tree
Kidney impedance associated with diabe-
tes is a gradual relationship that has major One method for handling concerns with
consequences for the patient. High blood glu- backslide and gathering is the decision tree
cose levels make it challeng- ing for the kid- approach. The rationale for employing deci-
neys to appropriately work. Mary, Bharathi, sion tree is to select a model which can gener-
Vigneshwari, and Sathyabama (2019) basi- ate decision criteria based on available data.
cally used affiliation rule mining in order to The decision tree computation is organized
predict diabetes for a given dataset. Kama- like a tree, having leaves, branches, and roots.
lesh, Prasanna, Bharathi, Dhanalakshmi, and Leaf center points address class names, while
Aroul Canessane (2016) makes use of associ- inside center points address dynamic features.
ation rule mining for the risk prediction in the The Figure 91.1 below shows the result of
case of CKD. The bunching information min- Decision tree classifier.
ing approach was approved utilizing choice
hypothesis by Revathy, Parvathavarthini, and 3.2. Support vector machine (SVM)
Caroline (2016); Subramanion, Balasubra- SVM is a grouping and relapse model which
manian, and Noordeen (2017). can be applied to both direct and nonlinear
issues. In this methodology, information is
3. CKD Prediction using Machine grouped utilizing based upon a hyperplane. In
this technique, information or data is treated
Learning Models
Information revelation is a well-known data-
mining application that includes many han-
dling stages. Data-mining approaches are
simpler to send when information is pre-han-
dled from many sources. Information readi-
ness or pre-handling incorporates cleaning,
removing, and changing information into
usable arrangements. A greater list of capa-
bilities is utilized to distinguish the funda-
mental parts of information portrayal. For AQ2:
Please
information revelation, a few arrangement provide
or example appraisal techniques are then uti- Figure 91.1. Accuracy result of decision tree source
classifier. for all the
lized. Bharathi Mary et al. (2019) shows an figures
AI based general ailment expectation model. Source:
AUTHOR'S SELF GENERATED
Predict-a-disease: Survey 645
as a point in n-layered space with the value 4. Dataset and Algorithm
of each element being compared to a certain
scenario. For order, the right hyper-plane will Algorithm: chronic kidney disease as an
be found that can effectively separate the two input Dataset Framework for high-accuracy
classes. prediction.
Step 1: Enter information.
Step 2: Prepare the information.
3.3. Random forest Step 2.1: Convert categorical variables to
The Random Forest strategy assembles an numerical values. Step 2.2: Substitute Mean
order and relapse group by joining a few for numerical missing values.
choice trees. Various choice trees are pro- Step 2.3: Use Mode S to replace missing
duced utilizing an arbitrary example of the categorical values. Step 3: Construct Classi-
preparation informational collections. Utiliz- fier Models.
ing countless choice trees works on the preci- Step 3.1: Construct a Decision Tree Model.
sion of the outcomes. The strategy has a low Step 3.2: Build a Random Forest Model.
execution time and can deal with missing Step 3.3: Build an SVM model.
information. Irregular timberland is utilized Step 4: Using a confusion matrix, check
to randomize the methodology instead of the the correctness of the built models.
preparation informational collection. The Step 5: Choose the best CKD prediction
choice class is an order made utilizing choice model.
trees. Figure 91.2 shows the accuracy result UCI chronic kidney disease dataset is
of Random forest classifier. The below given used for experimental purpose Rubini and
equation will be used to calculate the accu- Eswaran (2015) shown in Figure 91.3.
racy of the classifier model.
Accuracy = (TP + TN) / (TP + TN + FP + FN)
5. Results and Discussion
where TP= Positive observation and positive
prediction, TN= The observed and projected Models were build utilizing a preparation
values are both negative. FP= Although the informational collection of 280 cases, which
observation is negative, the expected out- is 70% of the first CKD informational collec-
come is positive. FN=Positive observation but tion. As far as boundary rightness, developed
negative prediction. models have been approved utilizing test
information that is 30% of the first informa-
tion. Exactness was evaluated involving a dis-
array network for this situation. Figure 91.4
shows the part of training dataset and Figure
91.5 shows the comparison between different
models.
6. Conclusion
In this work, a methodology for anticipat-
ing CKD at a beginning phase was proposed.
The models were prepared and affirmed uti-
lizing input boundaries gathered from CKD
Figure 91.2. Accuracy results of Random Forest patients in the dataset. The Random Forest,
Classifier. Support Vector Machine and Decision Tree
Source:
AUTHOR'S SELF GENERATED
models are used to investigate CKD. The
646 Innovations in Computational Intelligence: Trends and Technologies
References
[1] Aljaaf, A. J., Al-Jumeily, D., Haglan, H. M.,
Alloghani, M., Baker, T., Hussain, A. J., and
Mustafina, J. (2018). Early prediction of
chronic kidney disease using machine learn-
ing supported by predictive analytics. In
2018 Ieee congress on evolutionary compu-
tation (cec) (pp. 1–9).
[2] Devika, R., Avilala, S. V., and Subramaniyas-
wamy, V. (2019). Comparative study of clas-
Figure 91.3. Numerical Distribution of Dataset. sifier for chronic kidney disease prediction
Source:
using naive bayes, knn and random forest. In
AUTHOR'S SELF GENERATED 20193rd international conference on com-
puting methodologies and communication
(iccmc) (pp. 679–684).
[3] Dulhare, U. N., and Ayesha, M. (2016).
Extraction of action rules for chronic kidney
disease using na¨ıve bayes classifier. In 2016
ieee international conference on computa-
tional intelligence and computing research
(iccic) (pp. 1–5).
[4] Kamalesh, M. D., Prasanna, K. H., Bharathi,
B., Dhanalakshmi, R., and Aroul Canessane,
R. (2016). Predicting the risk of diabetes
Figure 91.4. Part of the Training Dataset with mellitus to subpopulations using association
CSV Format. rule mining. In Proceedings of the interna-
Source: tional conference on soft computing systems:
AUTHOR'S SELF GENERATED
Icscs 2015, volume 1 (pp. 59–65).
[5] Kaur, G., and Sharma, A. (2017). Predict
chronic kidney disease using data mining
algorithms in hadoop. In 2017 international
conference on inventive computing and
informatics (icici) (pp. 973–979).
[6] Mary, S. P., Bharathi, B., Vigneshwari, S.,
and Sathyabama, R. (2019). Neural com-
putation based general disease prediction
model. Int. J. Recent Technol. Eng.(IJRTE),
8 (2), 5646– 5449.
[7] Misir, R., Mitra, M., and Samanta, R. K.
(2017). A reduced set of features for chronic
kidney disease prediction. Journal of pathol-
Figure 91.5. Models Comparison. ogy informatics, 8 (1), 24.
[8] Panwong, P., and Iam-On, N. (2016). Pre-
Source: AUTHOR'S SELF GENERATED dicting transitional interval of kidney dis-
ease stages 3 to 5 using data mining method.
different models are assessed by their assump- In 2016 second asian conference on defence
tion accuracy. According to the aftereffects of technology (acdt) (pp. 145–150).
the comparison, the Random Forest Classifier [9] Ravindra, B., Sriraam, N., and Geetha, M.
model beats Decision trees and Support Vec- (2014). Discovery of significant parameters
tor Machines in expecting CKD. in kidney dialysis data sets by k-means
Predict-a-disease: Survey 647
algorithm. In International conference on and ann algorithms. International Journal of
circuits, communication, control and com- Computing and Business Research (IJCBR),
puting (pp. 452–454). 6 (2), 1–12.
[10] Revathy, S., Parvathavarthini, B., and Caro- [15] Wibawa, H. A., Malik, I., and Bahtiar, N.
line, S. S. (2016). Decision theory, an unprec- (2018). Evaluation of kernel-based extreme
edented validation scheme for rough-fuzzy learning machine performance for predic-
clustering. International Journal on Artificial tion of chronic kidney disease. In 2018 2nd
Intelligence Tools, 25 (02), 1650003. international conference on informatics and
[11] Ripon, S. H., et al. (2019). Rule induction computational sciences (icicos) (pp. 1–4).
and prediction of chronic kidney disease [16] Wickramasinghe, M., Perera, D., and
using boosting classifiers, ant-miner and j48 Kahandawaarachchi, K. (2017). Dietary pre-
decision tree. In 2019 international confer- diction for pa- tients with chronic kidney dis-
ence on electrical, computer and communi- ease (ckd) by considering blood potassium
cation engineering (ecce) (pp. 1–6). level using machine learning algorithms. In
[12] Rubini, L. J., and Eswaran, P. (2015). Uci 2017 IEEE life sciences conference (lsc) (pp.
machine learning repository: Chronic kid- 300–303).
ney disease data set. [17] Zhang, H., Hung, C.-L., Chu, W. C.-C.,
[13] Subramanion, R., Balasubramanian, P., and Chiu, P.-F., and Tang, C. Y. (2018). Chronic
Noordeen, S. (2017). Enforcement of rough kidney disease survival prediction with arti-
fuzzy clustering based on correlation analy- ficial neural networks. In 2018 ieee inter-
sis. International Arab Journal of Informa- national conference on bioinformatics and
tion Technology (IAJIT), 14 (1). biomedicine (bibm) (pp. 1351–1356).
[14] Vijayarani, S., Dhayanand, S., and Phil, M.
(2015). Kidney disease prediction using svm
92 Securing MANETs against black hole
attacks: A novel approach using secure
knowledge algorithm
Mohammed Abdul Bari1,a, Arshad Ahmad Khan Mohammad2,b,
Arif Mohammad3,c, and Abdul2
1
Associate Professor, Dept of CSE, KMEC
2
Department of CSE, GITAM Deemed to be University
Abstract: Due to the absence of a central network authority and decentralized communication among mobile
hosts, mobile ad hoc networks are susceptible to several security issues. A well-known threat is the black hole
attack, in which malicious nodes purposefully trash packets they receive from other nodes. The work aims to
determine packet-dropping reasons by monitoring the nodes in communication and then designing the mecha-
nism to mitigate the attack. The proposed work extends the existing secure knowledge algorithm to accurately
detect and minimize black hole attacks. The effectiveness of the proposed mechanism performance compared
with existing secure knowledge algorithms against black hole attack mitigation in MANETs. Experimental
results confirmed the improvement in the network’s overall security.
Keywords: Black hole attack; MANETs; packet drop
1. Introduction proving conventional infrastructure, which is
challenging.
Mobile ad hoc networks (MANETs) aim to The peer-to-peer communication ability of
establish ubiquitous internet connectivity MNAETs creates significant security difficul-
without depending on predefined infrastruc- ties. Moreover, dynamic and unpredictable
ture to enable connectivity regardless of posi- topology, free to mobile, and lack of a cen-
tion and time. This decentralized approach tral coordinator create a vulnerable network-
fosters self-configuration and maintenance ing environment in terms of security. The
abilities by assigning network intelligence network considered that, during multi-hop
to every mobile node. Because of their char- communication, all the nodes cooperate and
acteristics, MANETs can be set up in many coordinate. Unfortunately, this consideration
fields, such as residential networks, confer- is usually incorrect under unfavorable cir-
encing, military operations, law enforcement, cumstances, leading to security vulnerabilities
and rescue operations [1]. [2]. Malicious packet-dropping nodes, such
MANETs allow nodes to move and as black hole nodes, exploit this vulnerability
organize themselves freely, as the network and disrupt the network communication by
is designed with infrastructure-less settings. dropping the packets.
Thus, nodes can form a dynamic topology In the insistent context of MANETs, pack-
of many communication lines and function ets dropped by intermediate nodes create a
autonomously, providing a cost-effective significant problem with extensive implica-
and time-efficient option, specifically when tions, affecting overall network performance
[email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-92
Securing MANETs against black hole attacks: A novel approach using secure knowledge algorithm 649
and dependability. In literature, different to identify and mitigate black hole attacks
security mechanisms are designed to combat accurately. The effectiveness of the proposed
packet-dropping nodes, which are classified as mechanism performance compared with
credit-based, reputation-based, and acknowl- existing secure knowledge algorithms against
edgment-based methodologies [3–8]. The black hole attack mitigation in MANETs.
credit-based approach rewards the nodes for Experimental results confirmed the improve-
cooperative packet operation through a vir- ment in the network’s overall security. The
tual currency system. The system’s efficiency significant contributions of the proposed
is significantly impacted due to hardware pre- work are explained as follows;
requisites, leading to scalability problems [6].
Reputation-based systems determine the node 1. Secure Knowledge Algorithm Enhance-
behavior regarding packet operation through ment: Modifying the existing SKA routing
direct and indirect monitoring [7]. It solved protocol’s ability to identify and mitigate
the scalability problem, but the reason behind black hole attacks in MANETs accurately.
the packet dropping has not been determined. 2. Dynamic Defense Mechanism: This
Acknowledgment-based methods employ des- includes the status of the node resources,
tination-to-source acknowledgments to miti- such as residual energy and buffer sta-
gate packet-dropping nodes. The approach tus metrics, in the dynamic and adaptive
decrease routing overhead, displaying prom- defense mechanism, along with integrat-
ise with less memory and computational ing promiscuous monitoring.
demands. The existing ACK approaches are 3. Simulation-Based Validation: evaluate
ACK and AACK, each with distinct advan- and validate the performance of the pro-
tages and drawbacks regarding packet drop posed work in a network simulator with
mitigation. Thus, there is a requirement for different network scenarios in the pres-
an enhanced packet-drop prevention mecha- ence of a black hole attack, which con-
nism, which must determine the packet-drop firms the improvement in the network’s
reason along with packet-drop mitigation [8]. overall security compared to existing
The existing SKA algorithm specifies a protocols.
thorough answer via arbitrary monitor-
ing. This method continuously monitors the 2. Secure Knowledge Algorithm to
neighboring nodes regarding packet drop;
if it finds any node dropping the packet, it Minimize Black Hole Attack
determines the reason behind it. Thus, it MANETs enable decentralized communica-
determines whether the packet drop is mali- tion among mobile nodes without a fixed
cious or due to an error or fault in the system. infrastructure. However, their dynamic, peer-
However, it determines the reason for the to-peer, and self-organizing characteristics
packet dropping after the packet drop hap- make them vulnerable to security attacks,
pens up to some threshold value. The SKA particularly packet drops at the network layer
algorithm mitigates the attacks and improves by black holes. This work is an enhancement
MANET security, so it’s essential to keep of the existing SKA algorithm to identify
making changes and improvements to handle and accurately mitigate black hole attacks in
packet drop attacks effectively [5]. MANETs, where it determines the reasons
This paper analyzes the reasons for packet for packet drop.
loss in promiscuous mode and proposes a The proposed work used the AODV rout-
thorough approach for determining and com- ing protocol at the network layer, and it has
bating black hole attacks. We are modifying security vulnerabilities that make it easier for
the existing SKA routing protocol’s ability attackers to exploit it with black hole attacks.
650 Innovations in Computational Intelligence: Trends and Technologies
The attack attracts traffic towards it by send- node is deemed malicious or a black hole, this
ing information that it has best route to des- information is disseminated to notify other
tination, then drops the traffic. An enhanced nodes, allowing them to circumvent the com-
packet-drop prevention mechanism must be promised node in routing.
developed to determine the packet-drop rea- The work gives a comprehensive solution
son along with packet-drop mitigation while for accurately identifying and minimizing
mitigating black hole nodes. This work aims black hole attacks in MANETs. The algorithm
to improve the AODV using a secure knowl- includes the status of the node resources, such
edge algorithm to determine the packet-drop as residual energy and buffer status metrics,
reason and mitigation. in the dynamic and adaptive defense mecha-
Previous studies have explored a variety of nism, along with integrating promiscuous
approaches to counteract black hole attacks, monitoring [11]
including reputation-, credit-, and acknowl-
edgement-based tactics. Credit-based tech- 2.1. Technical implementation
niques motivate nodes, reputation-based
schemes depend on monitoring node behav- 1. Residual Energy Status:
ior, and acknowledgment-based approaches a. Calculate Eresidual The residual energy of
utilize confirmation messages. Nonetheless, the node using the following equation:
each methodology possesses distinct advan- Eresidual = Etotal – Econsumed
tages and drawbacks. This paper introduces a b. Assess Renery (residual energy ratio) as:
novel algorithm that combines the advantages Renery = Eresidual /Etotal
of multiple approaches to enhance black hole
2. Buffer Status:
attack detection and mitigation.
a. Determine Bremaining (remaining buffer
In our suggested approach, each node func-
capacity) as:
tions promiscuously, observing the packet
forwarding behavior of surrounding nodes. Bremaining = Btotal – Bused
The knowledge table held by each node con- b. Evaluate Rbuffer (buffer status ratio)
tains entries for the most recently forwarded using:
packet () and details regarding the surround- Rbuffer = Bremaining /Btotal
ing node (). Discrepancies between () and () 3. Integration into the Algorithm:
initiate an inquiry into possible black hole • The node updates the knowledge table
phenomena. values during promiscuous monitor-
The system guarantees thorough moni- ing of its neighbor, i.e., and
toring by analyzing both control and data • Updated values are compared with pre-
packets to avert the selective dropping traits defined thresholds to make dynamic
of black hole assaults. Furthermore, the decisions about the credibility of the
knowledge table includes metrics for residual nodes regarding packet operation,
energy, , and buffer state, , of adjacent nodes. mainly if these packet drops are mali-
Residual energy is computed as the ratio cious or have energy constraints and
of residual energy to total available energy, buffer overflow.
whereas buffer status is assessed by the ratio • Determine the nodes with the lowest
of remaining buffer capacity to total buffer values of or , as they are unintentional
capacity. packet-dropping nodes, broadcast the
Our approach integrates effortlessly with same information in the network
AODV [10], employing the lowest hop count
and maximum sequence numbers to identify Integrating these computations in the mon-
the optimal route. During route discovery, if a itoring approach enhances the algorithm’s
Securing MANETs against black hole attacks: A novel approach using secure knowledge algorithm 651
ability to vigorously assess neighboring broadcastInformation(node,
nodes’ status regarding energy and buffer, "BLACK_HOLE")
which helps to accurately detect and miti- update AODV routing tables to exclude
gate mechanisms against black hole attacks the identified node
in MANETs. else:
Algorithm: Secure Knowledge Algorithm update AODV routing tables based on
1. For each node in MANET: initialize the minimum hop count and maximum
knowledge_table[node] with fm = null, rm sequence numbers
= null, Renery = 0, Rbuffer = 0 7. Procedure: broadcastInformation(node,
2. for every packet processing at node: information)
update knowledge information_ broadcast information to all nodes in the
table[node]. fm with packet information MANET
update knowledge_table[node]. rm infor-
mation of the neighbor node
if knowledge information of the node_ 3. Results Analysis
table[node]. fm! = knowledge information The performance of the proposed work was
table[node]. rm: conducted within the robust Network Simu-
wait for a predefined time period lator Version-2 (NS2), leveraging its capa-
if knowledge_table[node]. fm != knowl- bilities to emulate mobile ad hoc networks
edge_table[node]. rm: (MANETs). The chosen version, NS2.35,
if isBlackHoleBehavior(node): provided a stable foundation for our experi-
handleBlackHole(node) mentation on a Linux operating system [9].
else: The simulated MANET environment com-
assessResidualEnergyAndBuffer(node) prised a variable number of nodes, ranging
3. Function: isBlackHoleBehavior(node) -> from 10 to 30, navigating a 600x600 area.
boolean The MAC protocol adopted was 802.11,
return knowledge_table[node]. fm! = featuring a radio range of 250 meters. The
knowledge_table[node]. rm: for consecu- simulation time spanned 100 seconds, cap-
tive checks turing the network’s dynamic behavior under
4. Procedure: handleBlackHole(node) diverse conditions. Traffic source emulation
broadcastInformation(node, employed a Constant Bit Rate (CBR) model
"BLACK_HOLE") with a packet size of 512 bytes. The nodes’
update AODV routing tables to exclude the mobility adhered to a Random Way Point
identified node model, with speeds varying between 2 and 12
5. Procedure: assessResidualEnergyAndBuff m/sec. The simulation parameters of the pro-
er(node) posed algorithm are shown in Table 92.1.
Renery = Etotal – Econsumed /Etotal The efficacy of our proposed protocol
Rbuffer = Btotal – Bused /Btotal was evaluated primarily through the Packet
Delivery Ratio (PDR). PDR, a pivotal metric
update knowledge_table[node]. Renery =
in assessing Quality of Service (QoS), reflects
Renery
the ratio of successfully delivered packets to
update knowledge_table[node]. Rbuffer =
the total generated. This parameter serves as
Rbuffer
a barometer for network efficiency and per-
if Renery < Etreshold or Rbuffer < Btreshold:
formance, especially in the presence of misbe-
broadcastInformation(node,
having nodes.[12]
"POTENTIAL_MISBEHAVING")
Our simulation results, graphically
6. Procedure: integrateWithAODV(node)
depicted in Figure 92.1 and Figure 92.2,
if isBlackHoleBehavior(node):
652 Innovations in Computational Intelligence: Trends and Technologies
showcase the Packet Delivery Ratio against
the backdrop of varying numbers of misbe-
having nodes over durations of 100 and 500
seconds, respectively. The findings underscore
the impact of security measures on the net-
work’s ability to combat adversarial nodes.
The introduction of black hole nodes
revealed discernible effects on the Packet
Delivery Ratio, emphasizing the vulnerability
of MANETs to misbehaving entities. Despite
these challenges, the proposed protocol dem-
onstrated resilience, maintaining a commend-
able Packet Delivery Ratio despite potential Figure 92.2. Illustrates the correlation between
the packet delivery ratio and the number
Table 92.1. Parameters of the simulation of malicious nodes, employing a simulation
environment duration of 100 seconds.
Source: AUTHOR'S SELF GENERATED
Parameter Values
Number of Nodes 10, 20, 30
Area Dimensions 600 x 600 disruptions. The comparison with the stand-
ard AODV protocol revealed a marked
MAC Protocol IEEE 802.11
improvement in throughput and delay. This
Transmission Range 250 meters performance enhancement validates the effec-
Simulation Duration 100 seconds tiveness of our proposed security measures in
Traffic Pattern Constant Bit Rate (CBR) fortifying the AODV routing protocol.
Packet Size 512 bytes Our simulation-based evaluation affirms
AQ1: Movement Model Random Waypoint the viability of the proposed protocol in miti-
Please gating black hole attacks within MANETs.
provide Node Speed 2, 4, 6, 8, 10, 12 meters/
source for second
The security enhancements, especially incor-
the table porating message digest with encryption,
Source: AUTHOR'S SELF GENERATED proved instrumental in maintaining a robust
Packet Delivery Ratio. As a next step, fur-
ther optimizations and refinements will be
explored to adapt the protocol to diverse
network conditions and to bolster its resist-
ance against evolving security threats. The
simulation outcomes contribute not only to
the enhancement of AODV but also to the
broader discourse on securing decentralized
communication networks.
This research substantially contributes to
the area by providing a solid defense against
black hole attacks, a common security prob-
AQ1:
Figure 92.1. Demonstrates the correlation lem in MANETs. Combined with the popular
Please
provide between the packet delivery ratio and the AODV routing protocol, the suggested Secure
source number of malicious nodes throughout a Knowledge Algorithm creates a dynamic and
for all the simulation duration of 100 seconds.
figures
adaptable barrier against malevolent nodes’
Source: deliberate packet dropping.
AUTHOR'S SELF GENERATED
Securing MANETs against black hole attacks: A novel approach using secure knowledge algorithm 653
3.1. Significance of the work in MANETs. The proposed protocol extends
the Secure Knowledge Algorithm, which
1. Enhanced Network Security: The paper
offers a flexible and dynamic defense against
addresses a critical security concern in
malicious nodes that maliciously drop pack-
MANETs, contributing to developing
ets. It further integrates residual energy and
more secure and resilient communica-
buffer status and uses promiscuous monitor-
tion networks. By mitigating black hole
ing to effectively identify and prevent black
attacks, the proposed algorithm enhances
hole nodes. Integrating these computations in
the overall security posture of MANETs,
the monitoring approach enhanced the algo-
which is crucial for applications in vari-
rithm’s ability to vigorously assess neighbor-
ous domains like military operations,
ing nodes’ status regarding energy and buffer,
emergency response, and decentralized
which helped to accurately detect and miti-
communication.
gate mechanisms against black hole attacks in
2. Adaptive Defense Mechanism: Intro-
MANETs. The effectiveness of the proposed
ducing a dynamic and adaptive defense
mechanism performance compared with
mechanism, incorporating promiscu-
existing secure knowledge algorithms against
ous monitoring and metrics for residual
black hole attack mitigation in MANETs.
energy and buffer status, offers a nuanced
Experimental results confirmed the improve-
approach to addressing security chal-
ment in the network’s overall security.
lenges. This adaptability is particularly
significant in MANETs, characterized by
their dynamic and unpredictable opera- References
tional environments. [1] Safari, Fatemeh, et al. The diverse technol-
3. Simulation-Based Validation: The rig- ogy of MANETs: A survey of applications
orous simulation-based validation con- and challenges. Int. J. Future Comput. Com-
ducted in Network Simulator Version-2 mun 12.2 (2023).
(NS2) underlines the practical significance [2] Sivapriya, N., and R. Mohandas. Analy-
of the proposed algorithm. The demon- sis on Essential Challenges and Attacks on
strated improvements in Packet Delivery MANET Security Appraisal. Journal of Alge-
braic Statistics 13.3 (2022): 2578–2589.
Ratio and resilience in the presence of
[3] Al Rubaiei, Mazoon Hashil, Hothefa Shaker
adversarial nodes underscore the practi-
Jassim, and Baraa T. Sharef. Performance
cal effectiveness of the algorithm, provid- analysis of black hole and worm hole
ing a valuable contribution to real-world attacks in MANETs. International Journal
MANET scenarios. of Communication Networks and Informa-
4. Broader Implications: Beyond the imme- tion Security 14.1 (2022): 126–131.
diate enhancement of the AODV proto- [4] Chandan, Radha Raman, and P. K. Mishra.
col, this work contributes to the broader Performance analysis of AODV under black
discourse on securing decentralized com- hole attack. Proceedings of 2nd Interna-
munication networks. The knowledge tional Conference on Advanced Computing
gathered from this study can help shape and Software Engineering (ICACSE). 2019.
future MANET security policies and tac- [5] Siddiqua, Ayesha, Kotari Sridevi, and
tics, which could impact developments in Arshad Ahmad Khan Mohammed. Prevent-
ing black hole attacks in MANETs using
adjacent disciplines.
secure knowledge algorithm. 2015 Interna-
tional Conference on Signal Processing and
4. Conclusion Communication Engineering Systems. IEEE,
2015.
The paper presented a systematic approach to [6] Mohammad, Arshad Ahmad Khan, Ali
address the security risk of black hole attacks Mirza Mahmood, and Srikanth Vemuru.
654 Innovations in Computational Intelligence: Trends and Technologies
Intentional and unintentional misbehaving [10] C. E. Perkins and E. M. Royer, Ad-hoc on-
node detection and prevention in mobile demand distance vector routing, Proceed-
ad hoc network. International Journal of ings WMCSA’99. Second IEEE Workshop
Hybrid Intelligence 1.2-3 (2019): 239–267. on Mobile Computing Systems and Appli-
[7] Kshirsagar, D. and Patil, A., 2013, July. cations, New Orleans, LA, USA, 1999, pp.
Blackhole attack detection and prevention 90–100, doi: 10.1109/MCSA.1999.749281.
by real time monitoring. In Computing, [11] Ijteba Sultana, Mohd Abdul Bari, Sanjay,
Communications and Networking Technol- Analytical Model for Assessing the Impact
ogies (ICCCNT), 2013 Fourth International of MANETs’ Bottleneck Nodes”, Interna-
Conference on (pp. 1–5). IEEE. tional Journal on Recent and Innovation
[8] Atheeq, C., and M. Munir A. Rabbani. Trends in Computing and Communication,
CACK—A Counter Based Authenticated ISSN: 2321-8169 Volume: 11 Issue: 10;pp
ACK to Mitigate Misbehaving Nodes from 575–581
MANETs. Recent Advances in Computer [12] Mohammed Abdul Bari, Shahanawaj
Science and Communications (Formerly: Ahamad, Mohammed Rahmat Ali, Smart-
Recent Patents on Computer Science) 14.3 phone Security and Protection Practices,
(2021): 837–847. International Journal of Engineering and
[9] Issariyakul, Teerawat, et al. Introduction to Applied Computer Science (IJEACS); ISBN:
network simulator 2 (NS2). Springer US, 9798799755577 Volume: 03, Issue: 01,
2009. December 2021 (International Journal, UK)
Pages 1–6
93 AI-powered navigation of the digital
frontier: Analyzing modern trends in
digital marketing and their influence
on consumer purchasing patterns with
special reference to pharma industry
Abhay Singh Rajawat1,a, Ved Prakash2,b, and Shailendra Singh
Chauhan3,c
1
Indian Institute of Management, Sirmaur, India
2
Department of Management, BBDITM, Lucknow, UP, India
3
Babu Banarsi Das Institute of Technology and Management, Lucknow, India
Abstract: The rise of Artificial Intelligence (AI) has revolutionized the digital landscape, transforming both con-
sumer behavior and marketing strategies across various industries. This research paper explores current trends
in digital marketing and their influence on consumer purchasing behavior, with a particular emphasis on the
pharmaceutical sector.
AI technologies, including natural language processing, machine learning, and predictive analytics, have
revolutionized the way businesses interact with customers in the digital landscape. The pharmaceutical indus-
try, known for its intricate regulations and demand for precision marketing, offers a compelling case to exam-
ine AI’s profound impact on marketing strategies. This study adopts a mixed-methods approach, combining
qualitative insights from consumers and industry experts with quantitative data analysis. Key objectives include
assessing the adoption of AI-driven marketing techniques in the pharmaceutical sector, analyzing the impact
of AI-powered personalized content delivery on consumer engagement, and understanding how these strategies
influence purchasing patterns. Findings reveal that AI-driven marketing has enabled pharmaceutical companies
to enhance consumer engagement through personalized content, increased targeting precision, and improved
customer service. Furthermore, it has expedited the product development and approval process, leading to a
more agile response to consumer demands. However, data privacy and ethics concerns have also emerged as
significant challenges. This study offers valuable insights into the transformative potential of artificial intelli-
gence in digital marketing, with a specific focus on its revolutionary impact in the pharmaceutical industry. The
implications of these findings extend to other sectors and underscore the importance of ethical AI utilization as
companies navigate the evolving digital frontier to meet consumer expectations and regulatory requirements.
Keywords: artificial intelligence, consumer behaviour, digital marketing, purchase intention
1. Introduction of consumer engagement and commerce.
The advent of artificial intelligence (AI) as a
In an age marked by swift technological
advancements and their deep influence on vir- powerful tool has significantly reshaped this
tually every facet of life, the digital landscape space, enhancing our ability to understand,
has emerged as a dynamic arena where busi- reach, and influence consumers. This study
nesses must navigate the intricate challenges examines the transformative role of AI in
[email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-93
656 Innovations in Computational Intelligence: Trends and Technologies
digital marketing and its significant impact ethical AI deployment, the significance of tai-
on consumer purchasing behavior, with a spe- lored content delivery, and the potential for
cial emphasis on the pharmaceutical sector. increased consumer engagement. It also calls
Digital marketing, once characterized by attention to the regulatory and ethical chal-
static banner ads and rudimentary targeting, lenges that must be addressed as industries
has matured into a multidimensional ecosys- and businesses embrace AI technologies to
tem where AI technologies wield an unprec- navigate the evolving digital landscape. As we
edented level of influence. In this context, venture deeper into the digital frontier, under-
Artificial intelligence (AI) comprises several standing the symbiotic relationship between
functionalities such as natural language pro- AI-powered marketing and consumer behav-
cessing, machine learning, predictive analyt- ior is crucial for businesses, policymakers,
ics, and data-driven insights. and researchers alike.
These technologies empower businesses The pharmaceutical industry is experienc-
to craft personalized, data-driven marketing ing a profound transformation, driven by the
strategies, transcending the one-size-fits-all swift advancement and integration of artifi-
approach that was commonplace in the past. cial intelligence technologies. These innova-
The pharmaceutical industry stands as a tions are reshaping every stage of the drug
poignant exemplar of the disruptive changes development process, from target discovery
AI has ushered into digital marketing. In and lead optimization to clinical trial design
an industry characterized by rigorous regu- and post-marketing surveillance.
lations, stringent quality control, and the
imperative to inform and engage consumers 1.1. AI for drug discovery
accurately, AI has introduced a new para-
digm. This sector’s unique blend of challenges Artificial intelligence is making a significant
and opportunities offers a compelling van- impact on drug discovery, greatly accelerat-
tage point from which to analyze the broader ing the identification of new drug targets and
AI’s effects on digital marketing tactics and optimizing leads. AI algorithms can analyze
consumer behavior. With particular reference vast databases of genetic, clinical, and chemi-
to the pharmaceutical business, this research cal data, enabling faster and more precise
study aims to present a thorough knowledge identification of potential drug targets com-
of how artificial intelligence (AI) is chang- pared to traditional methods.
ing the landscape of digital marketing and, This leads to a shorter time frame for iden-
consequently, changing consumer purchase tifying promising candidates and reduces the
behaviors. risk of failure in later development stages.
Our investigation combines rigorous quan- Additionally, AI-powered virtual screen-
titative analysis and qualitative insights from ing platforms can efficiently screen millions
industry experts and consumers. We seek to of potential drug candidates in silico, elimi-
elucidate the extent to which AI-driven mar- nating the need for expensive and time-con-
keting strategies have been adopted by phar- suming laboratory experiments. Companies
maceutical companies, the manner in which like Ex Scientia and Benevolent AI are at the
they influence consumer engagement, and the forefront of this revolution, leveraging AI to
ultimate effects on purchasing behaviors. develop novel drugs with greater efficacy and
In the following sections, we will delve into fewer side effects.
the key findings, implications, and challenges
that emerge from this exploration of AI’s role 1.2. AI in clinical trials
in the modern digital marketing frontier. Our Artificial Intelligence is also transforming clin-
study underscores the pivotal importance of ical trial design and execution. AI models can
AI-powered navigation of the digital frontier: Analyzing modern trends in digital marketing 657
analyze patient data to predict their response evolving AI landscape, providing clear guide-
to treatment, allowing for personalized medi- lines for developers and ensuring patient
cine approaches and the selection of more safety.
homogenous patient populations for clini- AI is swiftly revolutionizing the pharma-
cal trials. This leads to more efficient trials, ceutical industry, paving the way for advance-
smaller sample sizes, and faster drug devel- ments in drug discovery, development, and
opment timelines. Artificial intelligence may delivery. To fully harness the potential of AI
also be used to continuously monitor patient to transform healthcare and improve patient
data, which will allow for the early identifi- outcomes, it will be essential to address
cation of unfavourable events and guarantee challenges and foster collaboration among
patient safety during the experiment. Com- researchers, policymakers, and industry
panies like Owkin and Insilco Medicine are leaders.
pioneering this field, developing AI-powered
solutions to optimize clinical trial design and
improve patient outcomes. 2. Related Work
In recent years, growing research and dis-
1.3. AI-powered manufacturing and cussions have centered on the intersection
of artificial intelligence and digital market-
supply chain
ing, reshaping how consumers interact with
AI is enhancing and simplifying manufactur- brands and make purchasing decisions across
ing and supply chain processes in the phar- various industries. This literature review
maceutical industry. By analyzing sensor data explores key themes and findings from past
from manufacturing equipment, AI-driven studies, highlighting the substantial influ-
predictive maintenance algorithms can antici- ence of AI on modern digital marketing, with
pate potential malfunctions, reducing costly a particular focus on its impact within the
downtime and improving efficiency. pharmaceutical sector.
This ensures the smooth and efficient AI in Digital Marketing: The integration
operation of manufacturing facilities, leading of artificial intelligence into digital marketing
to increased production output and reduced strategies has transformed how companies
costs. Additionally, AI can optimize inventory engage with their target audiences.AI ena-
management and logistics, ensuring the right bles the analysis of vast datasets, allowing for
drugs are available in the right place at the more accurate consumer segmentation and
right time, and addressing issues of stockouts personalized content delivery (Smith, 2018).
and delays. It also empowers businesses to predict con-
sumer behavior and tailor marketing efforts
1.4. Challenges and opportunities accordingly (Kannan, 2020).
Personalization and Consumer Engage-
Despite the immense potential of AI in the ment: AI’s capacity to process and interpret
pharma industry, some challenges need to be consumer data has led to a surge in person-
addressed. Data privacy and security remain alized marketing efforts. Studies have shown
a major concern, requiring robust data pro- that personalized content enhances consumer
tection measures to ensure patient confiden- engagement and brand loyalty (Xia et al.,
tiality. Additionally, the black-box nature of 2020). This personalization is particularly rel-
some AI models necessitates improved trans- evant in the pharmaceutical industry, where
parency and interpretability to build trust communicating complex health informa-
and address concerns about potential bias. tion to consumers requires a highly targeted
Regulatory frameworks need to adapt to the approach.
658 Innovations in Computational Intelligence: Trends and Technologies
AI in the Pharmaceutical Industry: The literature review highlights the importance
pharmaceutical sector, characterized by strin- of ethical AI use, personalized content deliv-
gent regulations, has embraced AI for vari- ery, regulatory compliance, and the changing
ous purposes. AI streamlines drug discovery, dynamics of consumer purchasing behaviors
clinical trial processes, and regulatory com- in the AI-driven marketing landscape. It sets
pliance (Fernandez, 2019). In marketing, AI the stage for our research, which seeks to
plays a crucial role in delivering accurate and advance the ongoing discussion about AI’s
relevant information to both healthcare pro- impact in both the digital realm and the phar-
fessionals and consumers, enhancing engage- maceutical sector.
ment. (Kruse, 2020).
Challenges and Ethical Concerns: The
widespread use of artificial intelligence (AI) 3. Research Methodology
in digital marketing has raised several ethi- The research methodology employed in the
cal, transparency, and data privacy concerns. study “AI-Powered Navigation of the Digital
As AI systems gather and analyze substan- Frontier: Analyzing Modern Trends in Digital
tial amounts of consumer data, issues related Marketing and Their Influence on Consumer
to consent and responsible data manage- Purchasing Patterns with Special Reference
ment have become increasingly important. to the Pharma Industry” aims to thoroughly
(Nguyen, 2021). examine the intricate connections between
Consumer Behavior and Purchasing Pat- AI-driven digital marketing trends and con-
terns: A key focus of this research is exam- sumer behaviors. By utilising a multifaceted
ining the impact of AI-driven marketing methodology, this study seeks to offer a com-
on customer purchasing behavior. Studies prehensive understanding of the dynamic
indicate that AI-powered personalization interactions shaping the modern consumer
enhances consumer decision-making, reduces landscape.
choice overload, and promotes product rec-
ommendations (Chen et al., 2018). However, 3.1. Research design
it is vital to understand the nuances of con-
sumer behavior in the context of healthcare To achieve a well-rounded understanding of
and pharmaceutical products. the research topic, a mixed-methods approach
The Future of AI in Digital Marketing: The is utilized. This involves both quantitative
role of AI in digital marketing is continuously and qualitative research methods to gather,
evolving.Researchers suggest that AI-driven analyze, and interpret data, allowing for a
marketing will continue to expand and trans- deeper exploration of the identified trends.
form the way businesses interact with con-
sumers (Yuan et al., 2021). It is imperative 3.2. Data collection
for businesses to adapt and for regulators to Quantitative Data: Surveys were conducted
establish clear guidelines that accommodate with a diverse group of consumers from dif-
these changes while safeguarding consumer ferent demographics. The survey covered
interests. aspects such as consumer preferences, brand
In conclusion, the fusion of AI and digital loyalty, trust in influencers, dependence on
marketing has paved the way for a new era online reviews, and views on e-commerce
in consumer behavior and engagement. The convenience. The collected data was then
pharmaceutical industry, with its unique chal- statistically analyzed to uncover trends and
lenges and opportunities, serves as a signifi- correlations.
cant case study for understanding the broader Qualitative Data: Focus groups and in-
implications of artificial intelligence. This depth interviews with selected individuals
AI-powered navigation of the digital frontier: Analyzing modern trends in digital marketing 659
were conducted. These discussions offered Objective 4 To examine the impact of
valuable insights into the underlying motiva- influencer endorsements on consumers’ pur-
tions and perceptions influencing consumer chasing decisions across various product
behaviors in response to digital marketing categories.
strategies. Objective-5To examine the influence of
online reviews on consumer buying behav-
3.3. Data analysis iour in various product categories.
Objective-6 To explore the relationship
Quantitative analysis was conducted using between e-commerce platforms and con-
statistical tools, such as regression analysis, to sumer buying behaviour across different
examine the relationships between variables. product categories.
3.4. Objective of the study 3.5. Hypothesis of the study
Objective-1 To analyze and compare con- Hypothesis 1 There is a notable change in
sumer buying behaviour before and after consumer buying behavior before and after
the adoption of Digital Marketing strategies the adoption of AI-enhanced digital market-
enhanced by Artificial Intelligence (AI). ing strategies.
Objective 2 To analyze the impact of per- Hypothesis 2 The use of personalized
sonalized advertising on consumer buying advertising algorithms in AI significantly
behaviour across diverse product categories influences consumer buying behaviour.”
Objective 3 To explore the relationship Hypothesis 3 “AI-driven social media
between consumer purchasing behavior and engagement algorithms play a crucial role in
social media activity across different product shaping consumer buying behaviour.”
categories. Hypothesis 4 “The integration of AI-
based influencer analysis impacts consumer
buying behaviour in the context of product
Table 93.1. Research methodology table endorsements.”
1 Universe Urban Customer Hypothesis 5 “AI-assisted sentiment anal-
ysis of positive online reviews significantly
Research
2
Design
Descriptive Research correlates with changes in consumer buying
behaviour.”
Sampling
3 Non-Probability Sampling Hypothesis-6 The utilization of AI-driven
Method
e-commerce platforms affects and shapes
Customers of selected consumer buying behaviour.
Sampling
4 urban Areas (Pharma
Unit
Customer)
5 Sample Size 400
3.6. Analysis and interpretation
AQ1:
Please Lucknow (Gomti Nagar, Objective-1 To analyze and compare con-
provide Geographical Indiranagar, Sarojini sumer buying behavior before and after the
6
in-text
OTE ADDED area (5) Nagar, Chinat and adoption of Digital Marketing strategies
citation
for tables Aminabad enhanced by Artificial Intelligence (AI).
1 and 2. Sampling
7 Convenience Sampling
AQ2: Technique
Please 4. Analysis and Interpretation
provide Tools of
Wilcoxon Analysis, Objective-1 To analyze and compare con-
source for 8 Analysis
Regression Analysis sumer buying behavior before and after the
missing Used
the tables
Source:
AUTHOR'S SELF GENERATED
660 Innovations in Computational Intelligence: Trends and Technologies
adoption of Digital Marketing strategies variables. The R² value of 0.950 shows that
enhanced by Artificial Intelligence (AI). 95% of the variance in consumer purchasing
Asymptotic significance is displayed. The behavior is explained by the AI-based digital
significance level is 0.05. marketing dimensions. The adjusted R² of
The null hypothesis states that the median 0.949 accounts for sample size and predic-
difference between the pre- and post-imple- tors, confirming the model’s robustness with
mentation of artificial intelligence (AI) is minimal impact from redundant variables.
zero, implying no significant change between The standard error of the estimate is 0.17017,
the two periods. To determine if AI-enhanced reflecting the average deviation of predictions
digital marketing methods have caused a sta- from actual data. Overall, the model demon-
tistically significant variation in the variable strates a strong fit, explaining a significant
of interest, the Wilcoxon signed-rank test was portion of the variability in consumer buying
applied. With the null hypothesis suggesting behavior.
no difference, it was rejected at a highly signif- The F ratio in the ANOVA Table tests the
icant level of 0, indicating a strong impact of overall goodness of Fit of the data-
AI on the variable. The test results show that The overall goodness of fit for the data is
the observed effect of AI is highly unlikely to evaluated using the F ratio in the ANOVA
be attributed to random chance, confirming a table (Table 93.4). With Consumer Buying
substantial and statistically significant differ- Behavior (CBB) as the dependent variable and
ence in the measurements before and after AI predictors including AI Personalized Adver-
adoption. tising, AI-Social Media Engagement, AI-
Model Summary Table 93.3 provides key Influencer Endorsements, AI-Positive Online
statistics for evaluating the regression mod- Reviews, and AI-E-Commerce, the F ratio of
el’s performance. An R-value of 0.975 reflects F (5,395) = 1500.395, P <.05, indicates that
a strong positive correlation between the these independent variables are statistically
significant predictors of the dependent vari-
able. This suggests that the regression model
Table 93.2. Hypothesis test summary provides an excellent fit for the data.
Null Test Sig. Decision
hypothesis 4.1. Dependent variable- consumer
The Related .000 Reject buying behavior
median of Samples the null
differences Wilcoxon hypothesis Table 93.5 demonstrates the statistical sig-
between Signed nificance of personalised advertising, social
Before AI Rank Test media engagement, influencer endorsements,
and After positive online reviews, and e-commerce (P
AI equals value less than 0.05).Therefore, the basic
0.
Source: SPSS Analysis Table 93.4. Statistical significance of the model
Table 93.3. Model summary Sum of Mean
Model Df F Sig.
Squares Square
Model R R Adjusted Std. An
1500
Square R Square error Regression 217.229 5 43.446 .000
.395
of the
Estimate Residual 11.438 395 .029
1 .975 .950 .949 .17017 Total 228.667 400
Source: SPSS Analysis Source: SPSS Analysis
AI-powered navigation of the digital frontier: Analyzing modern trends in digital marketing 661
Table 93.5. Estimated model coefficient
Model Unstandardized Standardized T Sig.
Coefficients Coefficients
B Std. Error Beta
(Constant) -1.613 0.106 -15.165 0
AI-Personalized Advertising 0.155 0.031 0.082 5.044 0
1 AI-Social Media Engagement -0.232 0.027 -0.221 -8.588 0
AI-Influencer endorsements 0.373 0.028 0.368 13.479 0
AI-Positive Online Review 0.029 0.03 0.029 0.943 0.346
AI-E-Commerce 1.137 0.035 0.768 32.457 0
Source: SPSS Analysis
form of the equation is Predicted (Consumer significant, with a coefficient (B) of 0.155, a
Buying Behaviour) = -1.613 +.155 AI Person- standardized coefficient (Beta) of 0.082, and
alized Advertising) + (-232 AI: Social Media a t-value of 5.044. The significance level (Sig.)
Management) +.373 AI: Influencer Endorse- is 0 (or very close to 0), indicating a strong
ment +.029 AI: Positive Online Review +.137 effect.
AI: E-commerce The results suggest that all The positive coefficient and low p-value
the digital marketing factors (personalized confirm Hypothesis 2, indicating that AI-
advertising, social media engagement, influ- personalized advertising has a statistically
encer endorsements, positive online reviews, significant and positive impact on consumer
and e-commerce) significantly influence con- buying behavior.
sumer buying behaviour in the various diverse Hypothesis 3 AI-driven social media
product categories. Specifically, personalised engagement algorithms play a crucial role in
advertising, social media engagement, and shaping consumer buying behaviour.
influencer endorsements have positive effects AI-Social Media Engagement: Coefficient
on consumer buying behaviour, while positive (B): -0.232, Standardised Coefficient (Beta):
online reviews have a non-significant effect. -0.221, t-Value: -8.588, and Significance
Interestingly, e-commerce has the strong- (Sig.): 0.
est positive impact on customer satisfaction Hypothesis 3 is supported by the low
among all the service quality factors. The p-value and negative coefficient, demonstrat-
constant term, which represents the baseline ing that AI-driven social media interaction
level of consumer buying behaviour, is also has a statistically significant negative effect
statistically significant (p < 0.001). on customer purchasing behavior.
According to the model, e-commerce, influ- Hypothesis 4: “The integration of AI-
encer endorsements, social media engage- based influencer analysis impacts consumer
ment, targeted advertising, and AI-generated buying behaviour in the context of product
positive online reviews have a more signifi- endorsements.”
cant impact on customer purchasing decisions The integration of AI-based influencer
across various product categories compared analysis impacts consumer buying behaviour
to positive reviews alone. in the context of product endorsements.
AI-Influencer Endorsements: Coefficient
4.2. Hypothesis testing (B): 0.373, Standardised Coefficient (Beta):
0.368, t-Value: 13.479, and Significance
Hypothesis 2 The impact of AI-personalized (Sig.): 0. The positive coefficient and low
advertising on consumer buying behavior is
662 Innovations in Computational Intelligence: Trends and Technologies
p-value validate Hypothesis 4, indicating that engagement algorithms and positive online
AI-powered influencer analysis has a statis- reviews, underscore the industry-specific con-
tically significant positive effect on customer siderations that must be acknowledged in
purchasing behavior. crafting effective marketing strategies. This
Hypothesis 5: AI-assisted sentiment anal- research not only contributes to the under-
ysis of positive online reviews significantly standing of AI’s role in the digital marketing
correlates with changes in consumer buying landscape but also prompts further investiga-
behaviour. tion into the evolving dynamics of consumer
AI-Positive Online Review: Coefficient (B): behaviour in the pharmaceutical sector, offer-
0.029, Standardised Coefficient (Beta): 0.029, ing valuable implications for future market-
t-Value: 0.943, and Significance (Sig.) 0.346. ing practices and strategic endeavors. The
The positive coefficient with a higher findings also suggest that personalized adver-
p-value (0.346) suggests that AI-positive tising, influencer endorsements, and the inte-
online reviews may not have a statistically sig- gration of AI in e-commerce platforms are
nificant correlation with changes in consumer influential factors positively impacting con-
buying behaviour and thus do not strongly sumer purchasing patterns in the pharma
support Hypothesis 5. industry. However, social media engagement
Hypothesis-6 The utilization of (Artificial algorithms and sentiment analysis of posi-
Intelligence) AI-driven e-commerce platforms tive online reviews may not have as strong
affects and shapes consumer buying behavior. an impact in this specific industry context.
AI-E-Commerce: Coefficient (B): 1.137, It’s essential for stakeholders in the pharma
Standardised Coefficient (Beta): 0.768, industry to leverage these AI-driven market-
t-Value: 32.457, Significance (Sig.): 0. ing strategies effectively to enhance consumer
The very high positive coefficient and low engagement and drive purchasing behaviour.
p-value strongly support Hypothesis 6. The
use of AI-powered e-commerce platforms has
a notably strong positive impact on consumer 6. Suggestions for Future Research
buying behavior. AI-Powered Navigation of the Digital Fron-
In conclusion, the evidence supports tier: Analysing Modern Trends in Digital
Hypotheses 2, 3, 4, and 6, whereas Hypoth- Marketing and Their Influence on Consumer
esis 5 does not receive strong support regard- Purchasing Patterns with Special Reference to
ing the AI-Positive Online Review variable. the Pharma Industry.” Here are some sugges-
tions for future research: Looking forward,
the study on AI-powered digital marketing
5. Conclusion trends in the pharmaceutical industry opens
In conclusion, the exploration of AI (Artifi- avenues for several future research directions.
cial Intelligence) powered digital marketing Future investigations could delve into the
trends within the pharmaceutical industry optimization of AI algorithms for enhanced
has uncovered multifaceted insights into con- personalization in pharmaceutical marketing,
sumer purchasing behaviour. The positive exploring the potential integration of emerg-
impact of personalized advertising, influencer ing technologies like augmented reality or
endorsements, and AI-driven e-commerce virtual reality. Additionally, considering the
platforms emphasizes the transformative ethical dimensions of Artificial Intelligence
potential of AI in shaping consumer engage- applications in marketing and addressing
ment and decision-making processes. How- consumer privacy concerns would be crucial.
ever, the nuanced findings, including the Further research could also focus on develop-
less pronounced influence of social media ing and testing innovative AI-driven strategies,
AI-powered navigation of the digital frontier: Analyzing modern trends in digital marketing 663
such as interactive content or voice-activated [2] Bag, Surajit, Gautam Srivastava, Md
applications, to gauge their effectiveness in Mamoon Al Bashir, Sushma Kumari, Miha-
engaging pharmaceutical consumers. With lis Giannakis, and Abdul Hannan Chowd-
the continuous evolution of both AI tech- hury. “Journey of customers in this digital
nologies and consumer behaviours, ongoing era: Understanding the role of artificial intel-
ligence technologies in user engagement and
research in this domain promises to guide
conversion.” Benchmarking: An Interna-
marketers in refining strategies that resonate
tional Journal 29, no. 7 (2022): 2074-2098.
effectively with the unique dynamics of the [3] Dwivedi, Yogesh K., Laurie Hughes, Elvira
pharmaceutical market. Ismagilova, Gert Aarts, Crispin Coombs,
Tom Crick, Yanqing Duan et al. “Artificial
Intelligence (AI): Multidisciplinary perspec-
Acknowledgment tives on emerging challenges, opportunities,
I want to convey my heartfelt appreciation and agenda for research, practice and pol-
to Dr. SS Chauhan for his priceless contribu- icy.” International Journal of Information
tions to this research paper. Their expertise, Management 57 (2021): 101994.
dedication, and collaborative spirit greatly [4] European Medicines Agency. Regulatory con-
enriched the quality of this worked. SS Chau- siderations for AI in medical devices, (2020).
han showed a strong dedication to the pro- https://www.ema.europa.eu/en/news/
artificial-intelligence-medicine-regulation
ject’s success by playing a significant part in
[5] Müller, Susanne, Suzanne Ackloo, Arij Al
data analysis and manuscript preparation.
Chawaf, Bissan Al-Lazikani, Albert Antolin,
Their insights and thoughtful feedback were Jonathan B. Baell, Hartmut Beck et al. “Tar-
instrumental in shaping the direction of our get 2035–update on the quest for a probe
research. for every protein.” RSC Medicinal Chemis-
I am also thankful for the continuous sup- try 13, no. 1 (2022): 13-21.
port and encouragement provided by Dr. SS [6] Gupta, Abhishek, Connor Wright, Marianna
Chauhan throughout the entire research pro- Bergamaschi Ganapini, Masa Sweidan, and
cess. Their enthusiasm and shared passion for Renjie Butalid. “State of AI Ethics Report
our subject matter created a positive and pro- (Volume 6, February 2022).” arXiv preprint
ductive working environment. arXiv:2202.07435 (2022).
I extend my heartfelt appreciation to Dr. [7] Gallagher, Michael. “Data collection and
SS Chauhan for their unwavering commit- analysis.” Researching with children and
young people: Research design, methods and
ment to excellence. This made a substantial
analysis (2009): 65-127.
contribution to the accomplishment of this
[8] Kang, Yuhao, Song Gao, Yunlei Liang, Min-
study project. gxiao Li, Jinmeng Rao, and Jake Kruse.
Thank you, Dr. SS Chauhan, for being an “Multiscale dynamic human mobility flow
outstanding collaborator and co-author. Your dataset in the US during the COVID-19 epi-
dedication to this paper has made a lasting demic.” Scientific data 7, no. 1 (2020): 390.
impact, and I am appreciative of the chance [9] Kothari, Chakravanti Rajagopalachari.
to collaborate with you. Research methodology. new Age, 2004.
[10] Li, Lin, Lixin Qin, Zeguo Xu, Youbing Yin,
Xin Wang, Bin Kong, Junjie Bai et al. “Using
AQ3: : References artificial intelligence to detect COVID-
Please
[1] Álvarez-Machancoses, Óscar, and Juan Luis 19 and community-acquired pneumonia
provide
cross- Fernández-Martínez. “Using artificial intelli- based on pulmonary CT: evaluation of the
reference gence methods to speed up drug discovery.” diagnostic accuracy.” Radiology 296, no. 2
in text
Expert opinion on drug discovery 14, no. 8 (2020): E65-E71.
for all
references. (2019): 769-777. [11] Meng, Yuan, Yang Yang, Sanmukh Kup-
pannagari, Rajgopal Kannan, and Viktor
CHECKED AND CONFIRMED
664 Innovations in Computational Intelligence: Trends and Technologies
Prasanna. “How to efficiently train your ai [17] Shrestha, Sunil, Asmita Priyadarshini Khati-
agent? characterizing and evaluating deep wada, Sudesh Gyawali, P. Ravi Shankar,
reinforcement learning on heterogeneous and Subish Palaian. “Overview, challenges
platforms.” In 2020 IEEE High Performance and future prospects of drug information
Extreme Computing Conference (HPEC), services in Nepal: a reflective commentary.”
pp. 1–7. IEEE, 2020. Journal of Multidisciplinary Healthcare
[12] Morgan, George A., and Robert J. Harmon. (2020): 287-295.
“Data collection techniques.” Journal-Amer- [18] Snyder, Hannah. “Literature review as a
ican Academy Of Child And Adolescent Psy- research methodology: An overview and
chiatry 40, no. 8 (2001): 973-976. guidelines.” Journal of business research 104
[13] Nguyen-Thi, Hai-Yen, Thuy-Tram Nguyen- (2019): 333-339.
Ngoc, Minh-Thu Do-Tran, Dung Van Do, [19] US Food and Drug Administration. Artificial
Luyen Dinh Pham, and Nguyen Dang Tu Le. Intelligence and Machine Learning (AI/ML)-
“Job satisfaction of clinical pharmacists and Based Software as a Medical Device (SaMD)
clinical pharmacy activities implemented at Action Plan. (2021, January 12). https://
Ho Chi Minh city, Vietnam.” PloS one 16, www.fda.gov/news-events/press announce-
no. 1 (2021): e0245537. ments/fda-releases-artificial-intelligencema-
[14] PANNEERSELVAM, Ramasamy. Research chine-learning-action-plan
methodology. PHI Learning Pvt. Ltd., 2014. [20] Vempati, Raveendrababu., et al. Challenges
[15] Smith, Justin S., Adrian E. Roitberg, and and opportunities for AI adoption in phar-
Olexandr Isayev. “Transforming computa- maceutical research. Nature Biomedical
tional drug discovery with machine learning Engineering, (2022): 6(3), 268-290.
and AI.” ACS medicinal chemistry letters 9, [21] Wong, Chi Heem, Kien Wei Siah, and
no. 11 (2018): 1065-1069. Andrew W. Lo. “Estimation of clinical trial
[16] Seyhan, Attila A. “Lost in translation: the success rates and related parameters.” AI
valley of death across preclinical and clinical and the regulatory landscape for drug devel-
divide–identification of problems and over- opment. Nature Reviews Drug Discovery,
coming obstacles.” Translational Medicine 20(9), 713-732
Communications 4, no. 1 (2019): 1-19.
94 Fake video detection
Mahendra Singha, Saurabh Jainb, Laraib Khanc, and dIaswarchandra
Department of Computer Science and Engineering, Babu Banarasi Das Institute of Technology and
Management, Lucknow, Uttar Pradesh, India
Abstract: The increasing computational power has rendered deep learning algorithms highly potent, facilitat-
ing the effortless creation of convincingly human- synthesized videos, commonly known as deepfakes. The
potential misuse of such realistic face-swapped fake videos in scenarios involving political turmoil, fabricated
terrorism incidents, revenge pornography, and the blackmailing of individuals is readily foreseeable. In this
study, we present a novel deep learning-based approach designed to effectively discern AI-generated fake videos
from authentic ones. Our methodology excels at automatically identifying instances of replacement and reenact-
ment deepfakes. We aim to deploy Artificial Intelligence (AI) a s a countermeasure against Artificial Intelligence
(AI) manipulations A Res-Next Convolutional Neural Network is used by our method to retrieve frame-level
characteristics. These attributes are then employed to train a Recurrent Neural Network (RNN) with Long
Short-Term Memory (LSTM) to classify videos depending on whether or not they have been altered, indicat-
ing their authenticity. We validate our approach using a large and heterogeneous dataset in order to simulate
real-time scenarios and improve the model’s performance with real data. The careful duration of this dataset
involves combining several pre-existing datasets, including Face-Forensic++, Deepfake Detection Challenge,
and Celeb-DF. Additionally, we demonstrate how our system can yield competitive results using a straightfor-
ward and robust approach.
Keywords: Computational power, Face-Forensic++, AI generated fake videos, Res-Next CNN, LSTM, RNN
1. Introduction because there are lots of pictures and videos
of them online. Sometimes, these fake videos
“Deepfake videos are fake videos made by
are made for inappropriate stuff, like putting
computer programs that use fancy technol-
famous faces on explicit photos or movies.
ogy to replace someone’s face in a picture or
These fake videos have become a big prob-
video with someone else’s face (Adee, 2020).
These fake videos fall into three categories: lem. They can be used to make fake speeches
changing the head, swapping faces, or just by leaders, cause problems between countries,
moving the lips to make it look like they’re influence elections, or trick people with false
saying something they’re not. People use information about money (Zhou, 2020; Guo,
high-tech computer stuff like auto encoders 2020). They can even make fake satellite pic-
and generative adversarial networks (GANs) tures to fool experts, like showing a bridge
to make these videos look real (Liu, 2021). that doesn’t exist (Fish, 2019).
These tools analyze how people’s faces Although there are good uses, like in mov-
move and make fake faces that look real (Lyu, ies or helping people without a voice, the bad
2018). To teach these programs to make real- uses are much more common. With better
istic pictures and videos, they need a ton of tech, making fake videos and pictures is super
pictures and videos. Famous folks like poli- easy now. You can even make a fake video
ticians and celebrities are often the targets using just one picture (Zakharov, 2019).
[email protected]; [email protected]; [email protected];
a
DOI: 10.1201/9781003606635-94
666 Innovations in Computational Intelligence: Trends and Technologies
These fakes don’t just hurt famous people; adjusting brightness and contrast, and more.
they can fool anyone. For example, a fake By using these tricks with the right settings,
voice tricked a CEO into giving away a lot of they made the program perform 0.3% bet-
money (Damiani, 2019). There are apps like ter. Also, these tricks help the program han-
Deep Nude and Zao that let anyone put their dle all sorts of different examples, making it
face in movies or make inappropriate videos stronger and more accurate.
without permission (Samuel, 2019; Guardian,
2019). These fakes mess with people’s privacy 2.2. Architecture of the proposed
and lives. It’s really important to figure out deep fake detection model
what’s real online, especially with deepfake
videos. They’re often used for bad things, The proposed architecture for a deepfake
and almost anyone can make them now. Peo- detection model involves a multi-faceted
ple are trying different ways to detect these approach to discern between authentic and
fakes (Lyu, 2020) (Jafar, 2020). But because manipulated content. The input data, typi-
both making and finding these fakes use cally comprising video frames or sequences,
high-tech methods, it’s tough to stop them. undergoes meticulous preprocessing steps,
To fight against this problem, the US Defense including frame alignment, normalization,
Research Agency started a research program and augmentation. Feature extraction is
to detect fake videos (Turek, 2020). Big com- achieved through a combination of Convo-
panies like Facebook and Microsoft are also lutional Neural Networks (CNNs) for spa-
working to stop these fake videos (Schroep- tial pattern recognition, Recurrent Neural
fer, 2019). The number of papers about deep- Networks (RNNs) for temporal analysis in
fakes has increased a lot recently. People are video sequences, and 3D convolutional lay-
really studying this stuff more now, especially ers for handling spatiotemporal information.
since 2017.” The architectural components include an
encoder-decoder structure for feature recon-
struction, attention mechanisms to focus on
2. Methodology relevant regions, Siamese networks for learn-
Detecting fake video is a challenging task, as ing differences between authentic and manip-
the technology used to create them continues ulated content pairs, and Capsule Networks
to advance. The primary contribution of this to capture hierarchical relationships in fea-
work is suggesting a fresh model structure tures. Model fusion integrates information
for identifying deepfake videos. As dataset from various modalities, such as audio and
quality significantly influences model perfor- visual cues, while ensemble methods enhance
mance, we focused on enhancing its quality,
applying diverse image augmentation tech-
niques during training.
Here’s a proposed approach for fake video AQ1:
detection: Please
provide
in-text
2.1. Image augmentation citation
for all
figures
Image augmentation is like giving extra prac-
tice to a computer program CNN so that it
AQ2:
gets better at understanding things. In this Please
research, they used different tricks to make Figure 94.1. Working procedure of fake video provide
the practice more helpful, like changing the detection. source
for all the
size, turning things around, flipping them, Source: AUTHOR'S SELF GENERATED figures
Fake video detection 667
overall accuracy. Postprocessing involves
thresholding and temporal analysis to make
final authenticity determinations. During
training, a curated dataset with both authen-
tic and deepfake samples is used, employing a
binary cross-entropy loss function and regu-
larization techniques like dropout and batch
normalization to prevent overfitting. For
efficient training, optimization makes use of
learning rate schedules and algorithms such AQ3:
Please
as Adam or RMSprop. provide
caption-
Figure 94.3. for the
2.3. CNN-based model Source: Figure 94.3 Captain: figure
AUTHOR'S SELF GENERATED
In this model, we build an LSTM layer on top CNN-based model
of EfficientNet to integrate CNN and RNN.
RNN’s consideration of data sequences as a frames. A sequence of 10 frames, each with
feature vector input from CNN is its main a 2048-dimensional feature vector, is input to
advantage. EfficientNet, a CNN model pre- LSTM. The feature vector size is then reduced
trained on ImageNet, is employed. Convo- for efficient computations through dense lay-
lutional layers are used to extract features ers, aiming to classify images based on con-
from input video sequences. Normalization, volutional layer outputs. LSTM’s efficiency in
activation for non-linearity, and shrinking handling long data sequences over time makes
occur in the max-pooling layer after that. it the preferred classification layer. The model
The temporal fluctuations from the input incorporates transfer learning for binary clas-
video frames are captured by the feature sification, using binary cross-entropy loss and
vector that is produced. The LSTM uses fea- the Adam optimizer for optimization. This
ture vectors to update cell states during each modification effectively minimizes false nega-
iteration, allowing it to distinguish between tives, improving accuracy even for unseen
real and fake frames. During each iteration, attacks.
LSTM updates cell states with feature vec-
tors, distinguishing between real and fake
3. Thematic Overview
Deep fake Technology: This category involves
the creation of fake videos through advanced
technologies like machine learning, AI, and
neural networks. Deepfake videos can con-
vincingly manipulate facial expressions,
voices, and movements to make it challenging
to distinguish between real and fake content.
Techniques and Technologies for Detection:
Numerous techniques and technologies are
being developed to detect fake videos. These
include analyzing facial inconsistencies, iden-
tifying unnatural movements or inconsisten-
Figure 94.2. Data flow diagram of CNN Based cies in audio, and scrutinizing discrepancies
Model. in background details or visual elements.
Source: AUTHOR'S SELF GENERATED
Machine Learning and AI in Detection:
668 Innovations in Computational Intelligence: Trends and Technologies
Similar technologies used to create deepfakes and RNNs, has significantly improved the
are also being employed to detect them. accuracy of fake video detection. How-
Machine learning algorithms are trained ever, there is a need for ongoing research
to recognize patterns or anomalies that sug- to address the interpretability challenges
gest manipulation. Researchers are develop- associated with these complex models.
ing models that can differentiate between 2. Adversarial Attacks: Despite technologi-
authentic and manipulated content. cal advancements, fake video generators
Development of Anti-Deepfake Tools: continuously evolve, incorporating adver-
Efforts are underway to develop specialized sarial techniques to bypass detection
tools and software that can assist in identify- algorithms. This highlights the impor-
ing and verifying the authenticity of videos. tance of developing detection methods
These tools aim to automate the process of that are resilient to adversarial attacks
detecting fake videos and support forensic and can adapt to emerging manipulation
analysis for identifying digital tampering. strategies.
Collaborative Initiatives and Research: 3. Ethical Considerations: The use of sophis-
Various collaborations between academic ticated technology for fake video detec-
institutions, tech companies, and government tion raises ethical concerns, particularly
agencies are focusing on research and devel- regarding privacy and potential misuse.
opment efforts to combat the proliferation Striking a balance between maintain-
of fake videos. Competitions, challenges, and ing individual privacy and safeguarding
collaborative projects are being organized to against the harmful effects of fake videos
encourage innovation in this field. is a critical aspect that requires careful
Ethical and Legal Implications: Detecting consideration.
fake videos raises ethical concerns regarding 4. Generalization Across Diverse Content:
privacy, misinformation, and the potential for Ensuring the effectiveness of detection
manipulation. Legal frameworks are being methods across diverse types of content,
discussed and developed to address issues including different languages, cultures,
related to the creation, distribution, and con- and contexts, remains a challenge. Devel-
sequences of fake videos. oping models that generalize well and
User Education and Awareness: Promoting are not biased toward specific content
awareness among users about the existence characteristics is crucial for real-world
and potential impact of fake videos is crucial. applicability.
Educating individuals about the methods used 5. Explainability and Transparency: Many
to create fake videos and providing guidance advanced detection models lack transpar-
on verifying the authenticity of content are ency, making it challenging to understand
essential aspects of tackling this issue. the decision- making process. Enhanc-
Overall, the landscape of fake video detec- ing the explain ability of these models is
tion involves a multidisciplinary approach, essential for gaining trust in their capabil-
combining technological advancements, ities, especially in contexts where the con-
research initiatives, legal considerations, and sequences of false positives or negatives
public awareness efforts to mitigate the harm- can be significant.
ful effects of manipulated content. 6. Collaborative Approaches: Given the
global nature of fake video dissemination,
collaborative efforts are essential. The
4. Critical Analysis sharing of datasets, benchmarking stud-
1. Advancements in Deep Learning: The inte- ies, and standardized evaluation metrics
gration of deep learning type, like CNN can foster a more cohesive and effective
Fake video detection 669
approach to fake video detection across and deployment of fake video detection
the research community. technologies. Balancing the imperative
7. Real-time Challenges: The demand for to combat misinformation with privacy
real-time detection poses computational considerations and avoiding unintended
challenges, especially when dealing with consequences requires a thoughtful and
high-resolution video streams. Balancing ethical approach.
the need for rapid analysis with computa- 4. User Empowerment: The synthesis of
tional efficiency is an ongoing considera- user education and empowerment plays
tion in the development of practical and a pivotal role. Educating users about
scalable fake video detection solutions. the existence of fake videos, providing
tools for critical evaluation, and foster-
ing digital literacy contribute to a more
5. Synthesis and Implementation resilient user base capable of discerning
misinformation.
5.1. Synthesis 5. Regulatory Measures: The synthesis of
5.1.1. Synthesis on fake video regulatory measures on both national
detection and international levels is critical for
establishing legal guidelines and ensuring
The landscape of fake video detection is ethical standards in the use of fake video
evolving rapidly, driven by advancements in detection technologies. Regulatory frame-
technology, interdisciplinary collaboration, works provide a foundation for responsi-
and ethical considerations. The synthesis of ble development and deployment.
these key elements underscores the multi- 6. Explainable AI: The synthesis of explaina-
faceted nature of addressing the challenges ble artificial intelligence (XAI) techniques
posed by manipulated visual content. addresses the interpretability challenge
1. Technological Integration: The synthe- associated with complex detection mod-
sis of cutting-edge technologies, such as els. Providing transparent explanations
deep learning architectures and advanced for decisions enhances user trust and aids
image processing techniques, has signifi- in understanding the intricacies of the
cantly bolstered the capabilities of fake detection process.
video detection systems. This techno- 7. Adaptability and Innovation: The syn-
logical integration enables more accurate thesis of adaptive detection mechanisms
identification and mitigation of manipu- and a culture of innovation is necessary
lated content. to stay ahead of evolving manipulation
2. Interdisciplinary Approach: The synthesis techniques. Embracing a mindset of con-
of expertise from diverse fields, includ- tinuous improvement ensures that detec-
ing computer science, psychology, ethics, tion systems remain effective in the face of
and law, is crucial for developing com- emerging threats.
prehensive solutions. An interdisciplinary 8. Global Collaboration: The synthesis of
approach facilitates a nuanced under- global collaborative efforts is essential
standing of the complex issues surround- in the fight against fake videos, consider-
ing fake video detection, encompassing ing the borderless nature of information
technical, ethical, and societal dimensions. dissemination. International cooperation
3. Ethical Frameworks: The synthesis of on research, data sharing, and standardi-
robust ethical frameworks is essential zation contributes to a more unified and
in guiding the responsible development effective response.
670 Innovations in Computational Intelligence: Trends and Technologies
5.2. Implication 6. Impact on Journalism and Authentic-
ity: The ability to detect fake videos
5.2.1. Implications on fake video has implications for journalism and the
detection authenticity of visual news content. Relia-
1. Technological Arms Race: The ongoing ble detection methods contribute to main-
development of sophisticated fake video taining the credibility of news outlets by
generation techniques implies a continual preventing the spread of misinformation
technological arms race between those through manipulated videos.
creating manipulated content and those 7. Forensic Investigations: Fake video detec-
developing detection methods. This neces- tion has implications for forensic investi-
sitates constant innovation in detection gations, particularly in legal contexts. The
technologies to keep pace with evolving ability to authenticate or identify manipu-
manipulation strategies. lated visual evidence becomes crucial for
2. Privacy Concerns: As fake video detec- ensuring the integrity of legal proceedings
tion technologies become more advanced, and preventing the use of misleading vis-
there are implications for individual pri- ual content.
vacy. Striking a balance between detecting 8. Market Dynamics: The existence of
manipulated content and respecting pri- robust fake video detection technolo-
vacy rights is crucial, raising ethical con- gies influences market dynamics for both
siderations regarding the use and potential cybersecurity and entertainment indus-
misuse of surveillance technologies. tries. Companies investing in detection
3. Information Trustworthiness: Success- solutions find themselves in a competitive
ful fake video detection has profound market, while the entertainment indus-
implications for the overall trustworthi- try grapples with ensuring authenticity in
ness of visual information in the digital media production.
realm. Effective detection contributes to
maintaining the integrity of visual con- 6. Future Work
tent, ensuring that users can rely on the
authenticity of videos encountered online. Recommendations for Future Research on
4. Media Literacy and Education: The effec- Fake Video Detection:
tiveness of fake video detection under- 1. Adversarial Robustness: Investigate tech-
scores the importance of media literacy niques to enhance the robustness of fake
and education. As detection technologies video detection models against adversar-
improve, educating users about the exist- ial attacks. Understanding and mitigating
ence of deepfakes, manipulation tech- vulnerabilities to sophisticated manipula-
niques, and how to critically assess visual tion attempts will be crucial for maintain-
content becomes increasingly vital for fos- ing the effectiveness of detection systems.
tering a resilient and informed society. 2. Explainable AI in Detection Models:
5. Legal and Regulatory Responses: The impli- Explore and develop more explainable
cations on fake video detection extend to artificial intelligence (XAI) techniques for
legal and regulatory domains. Policymak- fake video detection models. Enhancing
ers and legal authorities face the challenge the interpretability of these models can
of crafting regulations that address the mis- improve user trust and understanding,
use of manipulated videos while safeguard- especially in contexts where the conse-
ing freedom of expression and preventing quences of false positives or negatives are
potential overreach in surveillance. significant.
Fake video detection 671
3. Cross-Domain Generalization: Research 9. Cross-Disciplinary Collaboration:
on improving the generalization capabili- Encourage cross-disciplinary collabora-
ties of fake video detection models across tion between researchers, industry experts,
diverse content types, languages, and policymakers, and ethicists. Combining
cultural contexts. Ensuring that detec- insights from diverse perspectives can lead
tion systems are effective across a broad to holistic solutions that consider both the
spectrum of scenarios is essential for real- technical and ethical dimensions of fake
world applicability. video detection.
4. Human-in-the-Loop Approaches: Investi- 10. Benchmarking and Standardization:
gate the integration of human-in-the-loop Establish standardized benchmarks and
approaches for fake video detection. Com- evaluation metrics for assessing the per-
bining machine intelligence with human formance of fake video detection meth-
judgment can enhance detection accuracy ods. This can facilitate fair comparisons
and provide a more nuanced understand- between different approaches, encourag-
ing of contextual cues that automated sys- ing transparency and fostering advance-
tems might overlook. ments in the field.
5. Deepfake Attribution Techniques: Explore
methods for attributing the source of
deepfakes, including identifying the origi- 7. Conclusion
nators of manipulated content. Develop- We conclude by presenting the rapid evolu-
ing reliable attribution techniques can tion of fake video detection technology signi-
have implications for accountability and fies a critical response to the escalating threat
legal measures against malicious actors. posed by manipulated multimedia content,
6. Real-Time Detection in High-Volume particularly in the form of deepfakes. The
Scenarios: Focus on optimizing fake video strides made in this field, primarily driven by
detection algorithms for real-time pro- advanced algorithms and the integration of
cessing, particularly in scenarios with high deep learning techniques, underscore a com-
volumes of video content. Addressing the mitment to preserving the integrity of digital
challenges associated with rapid analysis media. These technologies, exemplified by
and scalability is crucial for timely identi- sophisticated detection models, have shown
fication and mitigation of fake videos. promising results in identifying and mitigat-
7. Behavioral Analysis Integration: Investi- ing the dissemination of falsified videos. The
gate the integration of behavioral analy- video in frame sequences of 10, 20, 30, 40, 60,
sis alongside content-based approaches 80, and 100 can be processed by our model.
for enhanced detection accuracy. Under- The majority of people are aware of the dan-
standing patterns of user behavior and gers of face alteration in videos these days.
engagement can contribute to more com- Because of DF, we have access to a wide range
prehensive and context-aware detection of diverse fields, including advanced media,
systems. virtual reality, mechanical technology, educa-
8. Privacy-Preserving Detection Techniques: tion, and many more. In a different setting,
Research privacy- preserving methods for they stand for inventions that have the power
fake video detection to address concerns to destroy and threaten society as a whole.
related to surveillance and individual pri- However, the battle against fake videos
vacy. Developing techniques that allow remains dynamic, with creators continually
for effective detection without compro- innovating to evade detection mechanisms.
mising personal privacy rights is essential. This necessitates a sustained commitment to
672 Innovations in Computational Intelligence: Trends and Technologies
research and development, fostering inter- References AQ4:
disciplinary collaboration among experts in Please
[1] Suratkar, S., Kazi, F. Deep Fake Video Detec-
computer vision, artificial intelligence, and note that
tion Using Transfer Learning Approach. cross-
cybersecurity. The synergy of these fields is Arab Sci Eng 48, 9727–9737 (2023). https://reference
crucial to staying ahead of emerging threats doi.org/10.1007/s13369-022-07321-3
is not
cited.
and devising adaptive solutions that can [2] G. Pang, B. Zhang, Z. Teng, Z. Qi and J.
withstand the evolving sophistication of fake Fan, “MRE-Net: Multi-Rate CITATION ADDED
Excitation
video creation. Using the ResNet50 CNN, Network for Deepfake Video Detection,”
frame level detection is done, followed by in IEEE Transactions on Circuits and Sys-
video classification using the RNN and tems for Video Technology, vol. 33, no. 8,
LSTM. As a result of the listed parameters in pp. 3663-3676, Aug. 2023, doi: 10.1109/
the paper, the proposed method can identify TCSVT.2023.3239607
a fake video or a real video. Analysis of our [3] L. Zhang, T. Qiao, M. Xu, N. Zheng and S.
technique shows that it can reliably identify Xie, “Unsupervised Learning-Based Frame-
work for Deepfake Video Detection,” in
DF on the web under genuine states of dis-
IEEE Transactions on Multimedia, 2022,
persion, with an average of 94.63%. There is
doi:10.1109/TMM.2022.3182509
a high expectation that real-time information [4] Ganguly, S., Mohiuddin, S., Malakar, S. et
will be as accurate as possible. Having the al Visual attention-based deepfake video
opportunity to create a solution for a given forgery detection pattern analysis and appli-
issue without the need for an earlier hypo- cation 25, 981–992 (2022). https://doi.
thetical review is a central part of profound org/10.1007/s10044-022-01083-2
learning. However, we also have the option [5] Garcia, M. (2020). Ethical Considerations
to understand this arrangement’s beginning in Deepfake Detection. Journal of Computer
to evaluate its characteristics and constraints, Ethics, 28(2), 167-185.
so we spent considerable time imagining the [6] Nabi, S.T., Kumar, M., Singh, P. et al. A
channels within our network. Our promi- comprehensive survey of image and video
nent empirical findings have shown that the forgery techniques: variants, challenges, and
future directions. Multimedia Systems 28,
eyes and mouth are integral to recognizing
939992 (2022). https://doi.org/10.1007/
appearances made with DF. It is anticipated
s00530-021-00873-8
that future devices will make our organiza-
tions more powerful, efficient, and to make
them better able to understand profound
businesses.
95 Olympic odyssey: Tracing 120 years of
athletes sports, and global influence
Shreya Vermaa, Shreya Mishrab, Prakriti Singhc, Sneha Sethd, and
Sanjay Kumar Guptae
Computer Science and Engineering Department, Pranveer Singh Institute of Technology, Kanpur,
India
Abstract: The Olympics, a global event with 200+ nations, highlights exceptional athleticism. This study, uti-
lizing Python for data analysis, explores factors influencing a nation’s Olympic success. Examining data from
1896 to 2016, we analyze athlete count, funding, and coaching quality. Visualizations reveal insights into
nations roles in the Olympics. Our findings link success to factors like athlete count, sports funding, and coach-
ing quality. This study deepens understanding of country performance, informing policies to enhance Olympic
results. Implications extend to policymakers and sports institutions, guiding tailored approaches for improved
performance. Adaptability emerges as crucial for sustained competitiveness over time. Acknowledging histori-
cal data constraints, future research could delve into qualitative cultural influences and recent Olympic Games
for contemporary trends.
Keywords: Olympic Games, Performance Analysis, Country Performance, Exploratory Data Analysis, Athlete
Development, Success Factors, Sports Analytics
1. Introduction leveraging historical insights to anticipate the
future of the Olympics.
The Olympics, a global event since 1896,
provides athletes a chance to shine every four
years. This paper examines countries Olympic 2. Literature Review
performance from 1896 to 2016, focusing on The exploration of the Olympics multifac-
each country’s proficiency in specific sports eted dynamics and its evolution has garnered
annually. By comparing sport performances, significant scholarly attention across various
insights can be gained to identify areas for disciplines. This literature review surveys rel-
improvement and enhance future Olympic evant studies that have delved into different
participation. The modern Olympics encom- aspects of the Olympic Games, shedding light
pass numerous sports, attracting thousands on the historical, analytical, and managerial
of athletes from over 200 countries every perspectives. Xin-Yu Xiao and his team [1]
four years, alternating between summer and employed data science techniques to propose
winter every two years. Analyzing changes a framework for analyzing Olympic data,
over time, such as increased participation, uncovering intricate patterns and trends. This
evolving events, financial trends, improved data-driven approach has proved crucial in
performance, gender inclusivity, and external redefining how insights are gleaned from the
factors like pandemics, will inform future pre- complex web of Olympic data. Meanwhile,
dictions. This paper delves into these aspects, Kavitha and Badre [2] offered a comprehensive
[email protected]; [email protected]; [email protected]; [email protected];
a
DOI: 10.1201/9781003606635-95
674 Innovations in Computational Intelligence: Trends and Technologies
analysis of Olympic data using PySpark and 3. Methodology
Dash-Plotly tools, exemplifying how technol-
ogy can aid in the visualization of intricate In this research, we embark on a systematic
patterns, athlete participation, and medal exploration of the extensive history of the
distribution. In the exploration of Olym- Olympic Games, aiming to uncover their evo-
pics societal impact, Abeza et al. [3] delved lution over time. Our methodology involves
into the realm of ambush marketing through the following key steps:
social media. Their study delved into the Data Collection: Our journey begins with
influence of unauthorized promotions during meticulous data collection. We assembled a
the Games, providing insights into the chal- comprehensive historical dataset covering the
lenges faced by official sponsors. Cutait [4] period from Athens 1896 to Rio 2016. This
investigated the management performance of dataset was curated from www.sports-refer-
the Rio 2016 Summer Olympics, demonstrat- ence.com in May 2018 and is stored in a CSV
ing the significance of efficient administration file named “athlete events.csv,” comprising
in successfully executing these grand events. 271,116 rows and 15 columns.
Diving into volunteer contributions, Moreno, Data Pre-Processing: With this rich data-
Moragas, and Paningua [5] analyzed the evo- set, we conducted thorough data pre- pro-
lution of volunteers role in Olympic Games. cessing. We refined the raw data, addressing
Their work highlighted the integral nature of intricacies within columns such as age, gen-
volunteerism and its evolving importance in der, height, weight, team affiliation, and
the execution of seamless events. Similarly, more. Deterministic Imputation techniques,
Yamunathangam, Kirthicka, and Shahanas including Basic Numeric Imputation and Hot
[6] employed exploratory data analysis tech- Deck Imputation, were applied to handle null
niques to assess the performance of athletes values and ensure data accuracy.
in the Olympic Games, emphasizing the value Exploratory Data Analysis (EDA): We then
of data-driven insights in understanding ath- transitioned to Exploratory Data Analysis
lete achievement. Beyond the realm of the (EDA), a phase of in-depth inquiry that har-
Olympics, Dey et al. [7] provided insights nesses the power of visual representation. Using
into the epidemiological outbreak of COVID- tools such as histograms, bar graphs, box plots,
19 using exploratory data analysis, showcas- scatter plots, and more, we uncovered intricate
ing how analytical techniques can extend to trends and patterns, enabling a deeper under-
other global phenomena. Additionally, the standing of the Olympic Games evolution.
works by Bondu et al. [8] and Ramachan- Insight Generation: Within the realm of
dran and Tsokos [9] exemplify the broader EDA, we delved into insightful exploration.
applicability of exploratory data analysis in Using the multidimensional aspects of the
various fields. Moreover, Lange [10] provided dataset, we explored athlete participation,
historical context, illustrating the expansion gender dynamics, national contributions,
of the Olympics through the increasing num- sports involvement, event intricacies, and
ber of participating countries over the years. medal achievements. This holistic explora-
Wikipedia contributors [11] elucidated the tion not only provided descriptive insights
foundational concepts of exploratory data but also facilitated a grasp of the underlying
analysis, while emphasizing its significance dynamics shaping the Olympics over time.
as a fundamental analytical approach. Col- Our methodology, anchored by robust
lectively, the studies discussed underscore data collection, meticulous data preprocess-
the diverse applications of exploratory data ing, and insightful Exploratory Data Analysis,
analysis, statistical methodologies, and tech- guides our exploration of the Olympic Games
nological tools in unraveling the complexity intricate evolution. This approach ensures
of the Olympics. that we not only uncover historical footprints
Olympic odyssey: Tracing 120 years of athletes sports, and global influence 675
but also decode the dynamic narratives that
have propelled this global phenomenon.
4. Analysis and Visualization
In our Olympic study, we aimed to assess
sport popularity Figure 95.1. Athletics leads
with 34.8%, followed by Gymnastics at
24.1% and Swimming at 20.9%. Shooting
and cycling secure the fourth and fifth posi-
tions at 10.3% and 9.8%. These insights
inform future event planning. Our analysis
highlights the popularity of Athletics, Gym-
nastics, Swimming, Shooting, and Cycling,
guiding organizers and policymakers in shap-
ing Olympic programs.
In gender distribution Figure 95.2, our
analysis reveals a 73% male and 27% female
split among athletes, highlighting a signifi-
Figure 95.2. Gender distribution.
cant disparity. Efforts are needed for gender
Source:
diversity and inclusion. AUTHOR'S SELF GENERATED
Examining participant numbers Figure
95.3, Summer Olympics host 116,776 athletes,
contrasting with Winter Olympics at 18,958,
showcasing a notable scale difference between
the two seasons. It is crucial for Olympic
planning, participant numbers impact venue
capacity, logistics, and resource allocation for
successful Summer and Winter Games.
Figure 95.3. Number of participants in each
season.
Source:
AUTHOR'S SELF GENERATED
Our analysis examined medal distribution
AQ1:
Please in Summer and Winter Games Figure 95.4.
provide Figure 95.1. Most popular sports in the In the Summer Olympics, there are 11,459
source Olympics. gold, 11,220 silver, and 11,409 bronze med-
for all the
figures Source: AUTHOR'S SELF GENERATED
als. Similarly, in the Winter Olympics, we find
676 Innovations in Computational Intelligence: Trends and Technologies
1,913 gold, 1,896 silver, and 1,886 bronze diversity. Medal victories also show a median
medals. This balanced allocation across sea- age of 24, revealing success across age groups.
sons reflects fairness and acknowledges ath- This emphasizes that Olympic achievement
lete’s achievements, crucial for understanding is not age-restricted, inspiring athletes of all
competitive dynamics. ages to pursue their dreams. It reinforces
Our Olympic data analysis covered athlete the inclusivity and diversity of the Olympic
age distribution and medal victories Figures movement, where talents from different gen-
95.5 and 95.6. The median age of all Olym- erations unite on the world stage.
pic athletes is around 24 years, highlighting Our analysis of Olympic medal data
revealed the most dominant nation in terms
of total medals won: the United States. The
USA boasts an impressive 17,847 medals,
placing it at the forefront of Olympic success
Figure 95.7.
Figure 95.4. Distribution of medals by season.
Source: AUTHOR'S SELF GENERATED
Figure 95.6. Medal victories by age for gold,
silver, and bronze medals.
Source: AUTHOR'S SELF GENERATED
Figure 95.5. Age distribution of Olympic Figure 95.7. Age distribution of Olympic
athletes. athletes.
Source: AUTHOR'S SELF GENERATED Source: AUTHOR'S SELF GENERATED
Olympic odyssey: Tracing 120 years of athletes sports, and global influence 677
This insight highlights the enduring excel-
lence of the United States in the Olympics,
a testament to the dedication and talent of
American athletes across various sports,
solidifying the USA’s position as a standout
nation in Olympic history.
5. Medal Distribution for India
Our analysis of India’s Olympic performance
highlights its global significance, fostering
national pride and sports diplomacy.
Indian athletes excel in diverse sports, with
hockey leading, followed by athletics, shoot- Figure 95.9. Gender distribution of Indian
ing, wrestling, gymnastics, boxing, archery, athletes.
football, tennis, and weightlifting Figure 95.8. Source:
AUTHOR'S SELF GENERATED
However, there’s a gender gap, with 82.9%
males and 17.1% females, stressing the need
for gender diversity initiatives Figure 95.9.
The majority of Indian Olympic athletes,
aged 20 to 30, showcase the country’s com-
petitive strength Figure 95.10. India has
won 138 gold, 19 silver, and 40 bronze med-
als, symbolizing excellence in various disci-
plines Figure 95.11. This medal distribution
snapshot is crucial for understanding India’s
Olympic success, inspiring future athletes and
showcasing the nation’s capabilities in inter-
national sports.
Figure 95.10. Age distribution of Indian
athletes.
Source: AUTHOR'S SELF GENERATED
6. Results and Discussions
Our analysis of historical Olympic data from
1896 to 2016 revealed significant trends in
participation and performance. We witnessed
a growing number of participating nations,
athletes, and events, reflecting the Olympics
global expansion and increased diversity in
sports. A progressive trend in female par-
Figure 95.8. Most popular sports for Indian ticipation underscored the Olympics journey
athletes. towards gender inclusivity. We also exam-
Source: ined nations contributions and found varying
678 Innovations in Computational Intelligence: Trends and Technologies
reflected the ebb and flow of national excel-
lence, while anthropometric data demon-
strated the nuanced evolution of athletic
strategies. Our exploration was enriched by
eloquent visual representations, making these
transformations not only comprehensible
but also compelling. These findings have far
reached implications for nations and athletes
aiming to enhance their performance strate-
gies in future Olympic endeavors.
Acknowledgement
We’re incredibly grateful to Assistant Pro-
fessor Sanjay Kumar Gupta for his guidance
and encouragement throughout this research.
We acknowledge Pranveer Singh Institute of
Figure 95.11. Medal distribution for India.
Technology for their support and resources.
Our families, friends, and well-wishers have
Source:
AUTHOR'S SELF GENERATED been our constant inspiration.
performance profiles, highlighting shifts in References
athletic focus and strategies. This evaluation
provided insights into the broader narrative of [1] S. A. F. G. X. L. C. L. Xin-Yu Xiao, “Assess-
national contributions to athletic excellence. ing the impact of Olympics Games: A Study
Analyzing specific sports disciplines over time Based in Big Data Mining and Analysis,”
ICBDT, pp. 141-145, 2023.
revealed patterns linked to global trends and
[2] S. Kavitha, “Data analysis and visualiza-
socio-cultural dynamics, showing the Olym- tion of olympics using pys- park and dash-
pics role in reflecting societal interests. plotly.,” International Journal of Research
in Mathematics and Engineering Sciences
7. Conclusion (IJRME)., 2022.
[3] B.-M. Abeza, “The case of the three most
In our comprehensive analysis of the Olym- recent olympic games. Int. J. Sport Commu-
pic Games evolution from 1896 to the Rio nication (2020),” International Journal of
2016 Olympics, we employed Explora- Sport Communication, 2020.
tory Data Analysis and harnessed Python’s [4] [4] M. Cutait, “Management perfor-
data analytics capabilities. This exploration mance of the Rio 2016 summer olympic
unveiled several significant transformations games.,” 2020.
in the Olympics history. We observed a para- [5] M. A. Moreno, “The evolution of volunteers
at the olympic games.,” 2020.
digm shift with the introduction of the Winter
[6] K. D. (. Yamunathangam, “Performance
Olympics in 1924 and a substantial increase analysis in olympic games using exploratory
in participating nations, both in Summer and data analysis techniques.,” 2018.
Winter Games. The evolving demograph- [7] R. S. K. Dey, “Analyzing the epidemiological
ics, including the average athlete age and outbreak of covid- 19: A visual exploratory
increasing female participation, highlighted data analysis approach.,” 2020.
the Games inclusivity and gender diversity. [8] C. R. Bondu, “An exploratory data analy-
The accumulation of medals over the years sis approach for assessing the sources and
Olympic odyssey: Tracing 120 years of athletes sports, and global influence 679
distribution of naturally occurring contami- [10] D. Lange, “Summer olympics: number
nants (f, ba, mn, as) in groundwater from of participating countries 1896-2016.,”
southern quebec (canada).,” 2019. Statista.com, 2016.
[9] K. M. Ramachandran, “Mathematical statis- [11] W. contributors, “Exploratory data analy-
tics with applications in r.,” Academic Press, sis.,” 2020. [Online].
2020.
96 Crop prediction system using machine
learning
Neha Chauhana, Devansh Tomarb, Gaurav Singhc, Gaurav Mishrad,
and Anand Mishrae
Department Computer Science and Engineering, Babu Banarasi Das Institute of Technology and
Management, Lucknow, Uttar Pradesh, India
Abstract: This research paper investigates the convergence of machine learning and the Internet of Things (IoT)
within modern agriculture, with the objective of improving agricultural practices to meet global food demands.
The study adopts a comprehensive exploration, employing a holistic approach to smart farming by integrating
insights from diverse sources. The incorporation of practical demonstrations highlights high-accuracy results
with specific machine learning algorithms, imparting a practical dimension to the research. A robust literature
review establishes the paper within the existing knowledge landscape. The paper underscores the significance
of data preparation for feature selection, encompassing aspects such as comprehensive data collection, meticu-
lous data cleaning, and normalization of numerical features. However, it is acknowledged that the discussion
on challenges faced in deploying machine learning in agriculture is somewhat limited. The paper concludes
by advocating for enhanced methodological discussions, more detailed quantitative insights, and a thorough
exploration of challenges and prospects. Despite certain weaknesses, this research significantly contributes to
the evolving field of agricultural machine learning, providing valuable insights for researchers, practitioners,
and stakeholders actively involved in shaping the future of smart farming.
Keywords: Smart farming, Internet of Things (IoT), RandomForest, SVM, KNN, Google Colab, Kaggle
1. Introduction ensuring that crops receive the right amount
of water and nutrients. Machine learning has
In our ever-expanding world, where the
emerged as a game-changer in smart farming,
demand for food continues to rise, agricul-
offering the ability to analyze vast datasets
ture stands as a crucial pillar in nourishing
collected from IoT sensors and other sources.
the global population. Farmers, facing the
challenge of meeting this demand, are increas- This rapidly evolving field employs algorithms
ingly turning to modern technologies to opti- to predict and analyze crop growth and out-
mize crop production and minimize losses. At put [1]. These algorithms can accurately fore-
the forefront of this agricultural revolution is cast harvest outcomes by training machine
smart farming, to enhance productivity while learning models on comprehensive farm data
conserving resources. encompassing weather patterns, soil proper-
Smart farming integrates the Internet of ties, and crop growth stages. Precision farm-
Things (IoT), where sensors monitor key ing, a notable application of machine learning,
environmental factors like soil moisture and leverages data, and technology to optimize
temperature. This data guides farmers in agricultural practices such as fertilization,
making informed decisions about the optimal irrigation, and pest control. Machine learn-
timing for planting, watering, and harvesting, ing models process data from diverse sources
[email protected]; [email protected]; [email protected];
a
[email protected], [email protected]
d
DOI: 10.1201/9781003606635-96
Crop prediction system using machine learning 681
like satellite imagery and soil sensors, gener- data through sophisticated algorithms [3].
ating detailed maps of crop growth and nutri- This process is pivotal for ensuring the qual-
ent levels. Farmers can then use these maps to ity and size of data, which, when comprehen-
fine-tune their farming practices, maximizing sive, plays a crucial role in yielding precise
yield and minimizing waste. results and making informed predictions. Big
Despite the immense potential of machine data, characterized by its size, speed, and vari-
learning in agriculture, challenges such as the ety, proves instrumental in eliminating ran-
lack of data infrastructure, high technology domness and providing detailed outcomes.
costs, and the need for specialized expertise Unstructured data from diverse sources such
persist. Nevertheless, as more farms adopt as sensors, social media, digital networks,
precision agriculture and generate data, the physical devices, stock markets, and health
benefits of deploying machine learning are records is employed, often accessed through
becoming increasingly evident. While the APIs, web collection, and direct pathways.
field is still in its early stages, the promising Data preprocessing takes center stage in the
results achieved thus far suggest that machine journey of utilizing machine learning algo-
learning will play a pivotal role in shaping the rithms effectively. This involves data clean-
future of agriculture [2]. The authors under- ing and transformation to enhance credibility
score the importance of precise data analy- and usability. Data reduction is equally criti-
sis through machine learning algorithms, cal, addressing issues such as missing, incom-
focusing on constructing models that accu- plete, and noisy data that can otherwise lead
rately predict outcomes based on input data. to inaccurate results. Even with the most suit-
Classification algorithms like Decision Tree, able machine learning algorithms, incorrect
Naïve Bayes Classifier, and Random Forest results can emerge without proper attention
are explored, with the study predicting that to data quality. Data extraction is the pro-
the widespread adoption of machine learning cess of obtaining information from sources,
and adoption of machine learning-focused and data transformation involves converting
systems will revolutionize efficiency and pro- data into a format conducive to analysis—
ductivity across diverse industries. both integral steps in the data preprocessing
phase. Classification algorithms, including
Decision Trees, Naïve Bayes Classifiers, Sup-
2. Extensive Evaluation port Vector Machines, and Random Forests,
play a vital role in constructing models for
2.1. Methodology predicting correct labels based on provided
In recent times, machine learning has become data. The process involves training the model
a pervasive force, influencing various aspects with specific data and evaluating its results
of our lives, ranging from healthcare and using a separate set to ensure accuracy and
defense to education and urban development. desired outcomes. Noteworthy studies, such
It has emerged as a key player in decision- as those focusing on behavioral classification
making processes and has laid the foundation in neuroscience and predicting water quality
for innovative search engine infrastructures. using machine learning algorithms, under-
The impact of machine learning-oriented sys- score the significance of classification as a
tems is poised to grow substantially, becom- key supervised learning method for predic-
ing a transformative force in technology and tion. In the classification process, the dataset
significantly affecting sectors such as chip is divided into training and test sets, decisions
design and traffic estimations. A crucial factor are made on the model to be used, and pre-
in harnessing the power of machine learning dictions are established and evaluated based
lies in the collection and analysis of accurate on result accuracy. Model complexity and
682 Innovations in Computational Intelligence: Trends and Technologies
performance values are revealed through Data Pre-processing: Acknowledging the
these machine learning algorithms, tailored pivotal role of meticulous data pre-pro-
to specific problem areas and situations. The cessing, tasks such as soil analysis, climate
methodology for crop analysis and prediction data cleaning, and pertinent feature selec-
using ML algorithms is illustrated in detail in tion are undertaken. The accuracy of crop
Figure 96.1, showcasing the intricate steps prediction models is intricately tied to the
involved in leveraging these technologies for quality and preparation of input agricul-
agricultural insights. tural data.
D. Model Building: The research undertakes
2.2. Thematic overview a comprehensive exploration of the itera-
tive process involved in model building.
Crop Prediction and Agriculture: The research Encompassing selection, training, and
paper focuses on applying machine learning testing phases, this approach identifies
and AI in agriculture, particularly for predict- the most accurate algorithm for practical
ing crops, to enhance agriculture services and crop prediction. It signifies a commitment
crop detection. to refining predictive models tailored
explicitly to the nuances of agricultural
A. Multi-Crop Prediction: A key emphasis contexts.
lies in the concurrent prediction of multi- User-Friendly Interfaces: The creation of
ple crops, highlighting gains in crop plan- user-friendly interfaces takes center stage,
ning and mitigating the efficiency risk of facilitating easy access for farmers to inter-
overlooking co-existing factors that influ- pret and apply crop prediction insights.
ence agricultural output.
B. Machine Learning Algorithms: Diverse
machine learning algorithms, including 2.3. Major findings, trends, and
Bayes Net, Naïve Bayes, Decision Tree, debates
and Random Forest, among others, are Crop Prediction: The primary finding under-
strategically employed for crop prediction scores the effective utilization of machine
[4]. The paper meticulously explores the
intricacies of algorithm selection, delin-
eating their varying accuracies in predict-
ing an array of crops.
AQ1:
Please
provide
in-text
cita-
tion for
Figures 1
and 2.
AQ2:
Please
provide
source
for all the Figure 96.1. Methodology of crop prediction. Figure 96.2. Process flow model.
figures
Source: Source:
AUTHOR'S SELF GENERATED AUTHOR'S SELF GENERATED
Crop prediction system using machine learning 683
learning models in predicting various crops. learning in agriculture is somewhat limited,
Leveraging features such as soil composition, lacking a thorough exploration of these chal-
climate conditions, and historical agricultural lenges and potential solutions. The mention
data, the accuracy of predictions varies con- of data scarcity as a challenge could benefit
tingent upon the selected algorithm. from a more detailed examination of strate-
Accuracy Variation: A noteworthy obser- gies to address this issue. While the inclusion
vation is the discernible variation in pre- of a disclaimer regarding responsibility for
diction accuracy among different machine the content is standard practice, the reference
learning algorithms. This variation sparks to specific platforms like “MDPI” may be
debates regarding the optimal algorithm platform-dependent and could be made more
choices tailored to specific crops and diverse generic. The paper could further enhance its
agricultural scenarios. quality by providing more detailed quan-
Multi-Crop Prediction: A salient trend is titative insights into the impact of integrat-
the widespread adoption of multi-crop pre- ing machine learning and IoT on crop yields
diction systems, presenting a comprehensive and resource optimization [6]. Additionally,
view of optimal crop choices within specific addressing more recent developments in the
agricultural environments. field and expanding on future recommenda-
User-Friendly Interfaces: The development tions would contribute to the paper’s overall
of accessible interfaces and platforms emerges completeness.
as a prominent trend, facilitating farmers in Data preparation for feature selection: In
easily interpreting and applying crop predic- the realm of agricultural machine learning,
tion insights for more effective crop planning the research paper underscores the pivotal
and cultivation. role of meticulous data preparation for effec-
tive feature selection. Here are key aspects
2.4. Critical analysis highlighted in the paper:
Strengths: This research paper stands out for Table 96.1. Comparative Study
its commendable contribution to the agricul-
tural domain, offering a thorough explora-
tion of the integration of machine learning
AQ3:
and IoT in modern farming. The presenta- Please
tion is clear and logically structured, effec- provide
tively tackling the critical issue of enhancing in-text
NOTEcitation
ADDED
agricultural practices to meet the increas- for Table
ing global demand for food. The inclusion 1.
of experimental results, showcasing a nota-
ble classification accuracy achieved through
specific machine learning algorithms, adds a
practical dimension to the research [5]. The
well-executed literature review strengthens
the paper’s relevance, firmly grounding it in
the context of existing scholarly work. The
focus on smart farming and the applications
of machine learning reflects a deep under-
standing of contemporary agricultural trends.
Weaknesses: Despite its merits, this research
paper exhibits certain weaknesses. The discus-
sion on challenges faced in deploying machine
684 Innovations in Computational Intelligence: Trends and Technologies
Comprehensive Data Collection: Empha- challenges faced in deploying machine learn-
sizes the importance of collecting diverse ing in agriculture.
data, and integrating real-time information Practical Implementation and Decision-
from IoT sensors. Advocates for a dataset Making: The paper’s integration of practi-
encompassing soil properties, weather pat- cal demonstrations with specific machine
terns, crop growth stages, and pest outbreaks. learning algorithms suggests real-world
Meticulous Data Cleaning: Stresses the applicability, aiding farmers in informed
necessity of a thorough data preparation pro- decision-making about planting, watering,
cess to ensure the quality and relevance of the and harvesting. This enhances the credibility
dataset. Implies the need for addressing miss- of machine learning in agriculture.
ing or inconsistent values, ensuring overall Holistic Resource Optimization: Smart
data consistency. farming’s holistic approach considers soil
Normalization of Numerical Features: conditions, weather patterns, and pest out-
Advocates for the standardization of numeri- breaks, pointing to potential comprehensive
cal features to a common scale for consistent resource optimization. Integrating machine
and uniform analysis. learning and IoT aligns with sustainability
Categorical Variables: Highlights the goals in agriculture.
importance of employing suitable encod- Research Advancement: The paper
ing techniques for categorical variables to advances agricultural machine learning
enhance compatibility with machine learning research through a robust literature review
algorithms. and synthesis, enriching the knowledge base
[9]. It provides a foundation for exploring
new avenues.
3. Synthesis and Implication Challenges and Future Directions: While
excelling in various aspects, the paper identi-
3.1. Synthesis fies weaknesses, signaling avenues for future
The research paper makes a significant con- research. This involves in-depth studies on
tribution to the convergence of machine overcoming challenges in deploying machine
learning and agriculture by providing a com- learning in agriculture and quantitative
prehensive exploration of their integration. assessments of its impact.
Adopting a holistic approach to smart farm-
ing, the paper intertwines machine learn-
ing and IoT technologies, recognizing the
interconnected factors influencing crop pro-
duction, including soil conditions, weather
patterns, and pest outbreaks [8]. The strength
lies in its practical dimension, featuring high-
accuracy results with specific machine learn-
ing algorithms, enhancing the credibility and
applicability of the proposed approach. The
synthesis builds upon a robust foundation
established through a well-conducted litera-
ture review, positioning the paper within the
context of existing knowledge and contribut-
ing to the evolution of agricultural machine Figure 96.3. Accuracy comparison of 3.2
learning. However, the paper has weaknesses, models implications
particularly in the depth of its exploration of Source:
AUTHOR'S SELF GENERATED
Crop prediction system using machine learning 685
4. Conclusion and Future Work Saker. 2023. “Crop Prediction Model Using
Machine Learning Algorithms.” Applied
In conclusion, our agricultural technol- Sciences 13: 9288. https://doi.org/10.3390/
ogy initiative envisions a future where crop app13169288.
prediction and yield optimization are revo- [5] Khaki, Saeed, and Lizhi Wang. 2019. “Crop
lutionized through GPS integration and gov- Yield Prediction Using Deep Neural Net-
ernment Rain forecasting. We aim to mitigate works.” Frontiers in Plant Science 10 (May).
food crises by refining predictive models and https://doi.org/10.3389/fpls.2019.00621.
streamlining the farmer experience with a [6] Tushar Gupta; Dr. Sunil Maggu; Bhaskar
user-friendly app and automated data col- Kapoor. 2023. “Crop Prediction using
lection. Plans for personalized fertilizer rec- Machine Learning.” Iconic Research And
Engineering Journals 6 (9), 279–284.
ommendations target soil health, while the
[7] Suruliandi, A., G. Mariammal, and S.P. Raja.
integration of advanced machine learning 2021. “Crop Prediction Based on Soil and
algorithms like ANN and CNN promises Environmental Characteristics Using Fea-
continuous improvement in predicting opti- ture Selection Techniques.” Mathematical
mal crop conditions [10]. Rooted in sustaina- and Computer Modelling of Dynamical Sys-
bility, our vision is to propel agriculture into a tems 27(1):117–40. https://doi.org/10.1080/
technologically advanced and farmer-friendly 13873954.2021.1882505.
era, emphasizing a seamless blend of cutting- [8] Ashwitha, A., and C. A. Latha. 2022.
edge technology and farmer-centric solutions “Crop Recommendation and Yield Estima-
for global progress. tion Using Machine Learning.” Journal of
Mobile Multimedia, February. https://doi.
org/10.13052/jmm1550-4646.18320.
References [9] Vaishnave, M P, and R. Manivannan.
[1] Mahendra N. 2020. “Crop Prediction 2022. “An Empirical Study of Crop
Using Machine Learning Approaches.” Yield Prediction Using Reinforcement
International Journal of Engineering Learning,” February, 47–58. https://doi.
Research and https://doi.org/10.17577/ijert- org/10.1002/9781119821809.ch4.
v9is080029(08). https://doi.org/10.17577/ [10] Klompenburg, Thomas van, Ayalew Kas-
ijertv9is08. sahun, and Cagatay Catal. 2020. “Crop
[2] Shripathi Rao, Madhuri, Arushi Singh, Yield Prediction Using Machine Learning:
N.V. Subba Reddy, and Dinesh U Acharya. A Systematic Literature Review.” Computers
2022. “Crop Prediction Using Machine and Electronics in Agriculture 177 (Octo-
Learning.” Journal of Physics: Confer- ber): 105709. https://doi.org/10.1016/j.
ence Series 2161(1):012033.https://doi. compag.2020.105709.
org/10.1088/1742-6596/2161/1/012033. [11] Darwin, Bini, Pamela Dharmaraj, Shajin
[3] T. Aravind. 2021. “Review of Machine Prince, Daniela Elena Popescu, and
Learning Models for Crop Yield Predic- Duraisamy Jude Hemanth. 2021. “Recogni-
tion,” January. https://doi.org/10.4108/ tion of Bloom/Yield in Crop Images Using
eai.7-12-2021.2314568. Deep Learning Models for Smart Agricul-
[4] Elbasi, Ersin, Chamseddine Zaki, Ahmet ture: A Review.” Agronomy 11(4), 646.
E. Topcu, Wiem Abdelbaki, Aymen I. Zrei- https://doi.org/10.3390/agronomy1104
kat, Elda Cina, Ahmed Shdefat, and Louai 0646.
97 A mobile application to assist the
women in gestation period
V. lavanya,a Madhu Hasitha Manne,b Gnana Sri Manasa
Veeramachaneni,c Aditya Ram Manam,d and Sajid khan
Mohammede
Velagapudi Ramakrishna Siddhartha Engineering college, Andhra Pradesh, India
Abstract: The crucial phase in the women’s life is pregnancy. Being healthy during pregnancy is essential for
a birth. Nowadays pregnant women often encounter many complications due to lack of proper guidance and
knowledge. This leads to an increase of miscarriages, premature births, and infant deaths. Healthy diet and
engaging in appropriate exercises in pregnancy time are important for the well-being of both the mother and
neonate. This project proposes a mobile application which give a personalized diet chart as per user preferences
by using Sci-kit, Exercises recommendations which focus on promoting overall fitness and flexibility during
pregnancy as per trimester and the list includes contact details, addresses, and relevant information about hos-
pitals with specialized maternity services, neonatal intensive care units in case of emergency by using ball tree
nearest neighbor this also includes food, water, and tablet alarms, so that pregnant women can establish healthy
routines and minimize the risk of nutrient deficiencies, dehydration, and medication noncompliance.
Keywords: Mobile application, diet, sci-kit libraries, exercises, ball tree algorithm, alarms, pregnancy women
1. Introduction maternal deaths set by United Member States.
But only reaching the SDG is not enough, the
Every women will undergo pregnancy phase MMR should be reduced more. These can be
in their life time. India is the country which done only by the proper guidance in pregnancy
produces nearly one-fifth of worlds annual by taking diet at correct time, doing regular
child births with the birth of 25 million infants exercises, and taking sufficient water and tab-
each year. Maternal Mortality ratio of India lets at prescribed time. Regular check-ups is
is 122/1,00,000 as per UNICEF reports. In also mandatory. Health assistants appointed
46% of all maternal deaths, neonatal deaths by government will check and guide pregnant
are 40 percent that happen in labor or early 24 women once a week or thrice a month. But to
hours after birth of baby. The major causes of achieve the low maternal mortality rate there
these deaths contribute as pre-maturity-35%, should be day-to-day inspection for pregnant
neonatal infections-33%, asphyxia-20%, and women. On the other hand, women will also
congenital malformations-9%. In worldwide be confused in planning their delivery. There
150 women are dying averagely each day due will be confusion in selecting the hospitals.
to preventable reasons related to pregnancy Therefore, it is essential for technology to
and child birth. take a part in this by developing a applica-
As both Telugu states contribute highest tion which relates to the user preferences diet,
population in south India, these also achieved exercises in every trimester and the display of
sustainable development goal related to nearby hospitals list which includes relative
a
[email protected]; [email protected]; [email protected]; d218w1a5430@vrsec.
ac.in; [email protected]
DOI: 10.1201/9781003606635-97
A mobile application to assist the women in gestation period 687
information to plan their delivery and regular 2.1. Mobile application
check-ups.
When we open the mobile application, we
will be displayed with the splash screen and
2. Model then we get the login and signup pages so that
anyone can understand and adapt to it eas-
An ML based system is designed along with a
ily. The sign-up page which is user friendly in
mobile application which acts as an interface
which it takes the name, mail ID, password,
for the pregnant women. This mobile appli-
weight, height, pregnancy period/due date
cation is used to provide necessary guidance
and address as the user inputs.
during the gestation period. The mobile appli-
And one of the most important aspects is
cation has been designed using flutter which
that if the user is suffering from any chronical
provides a convenient way for the pregnant
diseases or not so that the it can be customized
women to operate effectively. Mobile appli-
based on the given information. After filling
cation has the advantage of being faster and
these details, the entire data is going to store
more efficient. It doesn’t require a stable inter-
in the firebase whenever the user is going to
net connection in order to work. It automati-
login then it should match the username and
cally stores the data locally in the mobile app
password which is already stored in the data-
in which it will be easy to retrieve the data.
base. After successful login it display the day
The interface is designed to be user friendly
and trimester of the gestation period of the
allowing users to access real time data from
user along with this it also displays a list of
the server and present it in a format which
four options which includes diet chart, exer-
is easy to understand. The overall proposed
cise, list of hospitals, food, and tablet alarms.
system to the mentioned problems is depicted
by using a flowchart.
2.2. Diet chart
Many of the systems [1] provide diet based
on the previous food preferences. As men-
tioned, customized diet chart is the one of
the features of this project. Having a bal-
anced and healthy diet is crucial in the
period of pregnancy for ensuring the health
of both baby and mother. The approach here
followed is including technology to give bal-
anced diet as per user preferences. Here the
first method is about searching for dataset,
AQ1:
Please as said earlier the food items is restricted for
provide Telugu people.
cita-
tions for
After search of many datasets, the result
Figures obtained is no proper dataset which suits to
1, 2, 4, 5 our objectives.so as initial step the research
and 7
was on Indian food composition table which
AQ2: was released by ICRISAT in 2017 specifies the
Please
provide components which are essential to be in food
source that was taken by pregnant women amount
for all the Figure 97.1. Working model.
figures
of calories that should be taken by the preg-
Source: nant women for each trimester.
AUTHOR'S SELF GENERATED
688 Innovations in Computational Intelligence: Trends and Technologies
Trimester Amount of calories selecting the type of diet chart either vegetar-
First trimester 1800
ian or non-vegetarian, this is also a binary
classified data in dataset. So, the conditions
Second trimester 2000
can select the diet chart according to it. The
Third trimester 2200–2400 main objective behind the usage of random.
Figure 97.2. No of calories for each trimester. Choice () is to avoid the repetitions of diet
chart every day, because choices () which is in
Source:
same ‘random’ library will generate the items
with repetition from the list As this algorithm
After the help of dietician, nutritionist,
is checked for the given dataset, there is no
gynecologist, the dataset was created which
combination of food items for nearly 180
has food items and calories and the type of
days that means one trimester with the choice
food as breakfast, morning snacks, lunch etc.
() method. So here the random method will
Now the step that was followed is including
help the user to generate their diet chart as
the random. Choice () method for giving the
per user preferences with the valuable dataset.
diet chart with much accuracy and in small
The output Figure 97.3 presents the diet
amount of time.
chart from morning to evening which includes
The syntax of the random method is
dry fruits, seeds, breakfast, morning snack,
random.choice(sequence)
lunch, evening snack, dinner and milk. It also
sequence maybe string,list,tuple
represents the amount of calories that will be
from each food type,
obtained by intaking of the given diet chart.
calories=Σdi
where i=no.of food types
d=calories 2.3. Exercises
Random. Choice () is the method which is By clicking the exercises option in the list of
scikit library of python. This will randomly options that specified, as per the trimester
select the food items from every food type and of the user it will show the list of exercises
in between the specified calories as in which list and by clicking each exercise it will show
the user’s gestation period was. After select- the time of exercise and specifies the steps of
ing the food items based on calories and food doing the particular exercise. User can expe-
type, it will calculate the sum of the calories rience the visualization of the exercises in
that was selected from every food type from
that day. And after there is condition to check
whether the given food items are in between
the calories of that semester, if not the func-
tion was again called and same process will
repeat till there is a proper selection of food
items for the user preferences [2]. As specified
the restrictions of food items for the women
who have chronical diseases.
As in dataset there will be attributes which
specifies all the chronical diseases faced by
women and it is binary data, that is, is the
user can take the food item or not. As the
code works with the random choices and the
conditions to select the food items which are Figure 97.3. Customized diet chart using
specified to the particular user as per their random method.
input data. Another important inclusion is Source:
A mobile application to assist the women in gestation period 689
the form of videos and the process will help 2.4. List of hospitals
the user to know how the exercise should be
Now the aim is to find the nearest hospitals
done.
from the given input place from the created
Physical activity is very important for
dataset. To convert the string input of the place
women in gestation period [3]. Physical activ-
name to geo coordinates there is a library in
ity not only the exercises it also includes yoga
python named geopy geocoders. In that library
and walking. After the exercises there will be
Nominate module is helpful in converting the
time for yoga asanas and the time for walk-
latitudes and longitudes. Now the task is to
ing. Here in this project, there will be timer
calculate the distance between the converted
for exercises and yoga. Then the user will
data and hospitals list in the procured data-
experience the time she spent for physical
set. Although there are many algorithms in
activity. All the collected knowledge about
machine learning, the first try is done on near-
the exercises in stored in the firebase and we
est neighbors brute force algorithm and unsu-
should retrieve them by using flutter to our
pervised nearest neighbor Tree and ball tree
application.
algorithm from sklearn neighbors module.
Here the input address is stored as strings
The Observations that obtained are:
as name of the places, the input address is
divided as door number, street, panchayat, 1. Brute force algorithm as default takes met-
and city/Village. The dataset includes the ric as “Euclidean” and calculates the near-
attributes such as name of the place, lati- est neighbors, As there are ‘n’ samples and
tudes, and longitudes. With the help of the “p” dimensions then this approach will
google maps the dataset is created with the take the scale as O[DP2] and work effi-
geo coordinates. cient for smaller dimensions and become
infeasible for huge datasets [4].
2. KD Tree algorithm will be more efficient
compared to the brute force approach for
same samples and dimensions it scales
the cost as O[DPlog(P)]. This has high-
est space complexity [5]. This works very
fast for low dimensional data than high
dimensional data.
3. Ball tree makes the partitions based on
the nesting hyper-spheres series [6], it is
costlier than the above both but it is more
Algorithm Small Large Cost
Datasets datasets
Brute Highly Low O[DP2]
Force Efficient Efficient
KD Tree Highly Low O[Dlog(P)]
Efficient Efficient
Ball tree Highly Highly O[Dlog(P)]
Efficient Efficient
Figure 97.5. Differences between
Figure 97.4. Architecture of the application nearest-neighbor.
Source: Source:
AUTHOR'S SELF GENERATED
690 Innovations in Computational Intelligence: Trends and Technologies
efficient for high end dimensional data. 6. chebyshev
The cost takes more because it calculates
recursively. By comparing all the metrics for ball tree,
the efficient metric that obtained is haversine
Based on the observations the selected [7]. Because haversine is the metric which
approach is ball tree algorithm. especially uses for radians, as the dataset con-
tains latitudes and longitudes values this is
2.5. Ball tree the best metric to use. Euclidean, Manhattan
and Minkowski these metrics are especially
Ball tree algorithm is one of the approaches to
and be efficient for linear data. Haversine
find the nearest neighbors and this is a unsu-
actual works on the basis of great circle and
pervised technique which calculates the dis-
it requires both input and output should be
tance recursively. Based on the centroid C and
in radians.
radius r it divides the whole data into nodes
It calculates the angular distance between
in a recursive order format where each node
two points on the great circle of the sphere in
determines the hyper-sphere defined by C and
the specified radius or distance. It takes two-
r. This uses the triangle inequality method
dimensional array [8] in which the first is lati-
for reducing the neighbor search. As ball tree
tude and second is longitude.
nodes are in spherical geometry it is highly
As shown in Figure 97.6 it forms a great
dependent on the respective training data.
circle on the sphere surface on base as the
The syntax of the BallTree algorithm goes
radius, this is the function of haversine. Here
as .
the query_radius method in ball tree is used to
BallTree(X,leaf_size=int,metric= ‘ ’).
find the hospitals list in between the specified
• X is an array which takes the sample as
radius and stores in the list. After specifying
and dimensions.
the input location, the algorithm will convert
• leaf_size by default it is 40, this will impact
it into the geo coordinates, that geo coordi-
the speed of the query and memory to
nates will the input for the ball tree and in the
store the tree that is constructed. Amount
specified radius the nearest hospitals from the
of the memory that is needed to store is
dataset will be selected and given as output.
samples/leaf_size
• metric by default it is minkowski where
p=2 3. Conclusion
Some of the valid metrics of the ball tree
The period of pregnancy is undoubtedly a
and the distance will be calculated as:
stage, in a women life requiring guidance and
1. Euclidean
2. Manhattan
3. Minkowski
4. Cosine Similarity
5. Haversine
Figure 97.6. Represents the nearby hospitals.
Source:
AUTHOR'S SELF GENERATED
A mobile application to assist the women in gestation period 691
nutritional information and user prefer-
ences. IEEE Access. 2019;7:96695-96711.
[2] Schulze MB, Martínez-González MA, Fung
TT, Lichtenstein AH, and Forouhi NG. Food
based dietary patterns and chronic disease
prevention. BMJ. 2018;361.
[3] Widiani ANN, and Noviani NN. The effect
of pregnancy exercise on the anxiety level
Figure 97.7. Nearby hospitals. of the third trimester of pregnant woman in
the Kecamatan Sukawati. Int J Res Med Sci.
Source:
2020;8:4197-4199.
continuous monitoring. To address the need [4] Shengren L, and Amenta N. Brute-force
this project introduces a designed application k-nearest neighbors search on the GPU. In
tailored for pregnant women. This innova- Proceedings of the 8th International Confer-
ence on Similarity Search and Applications
tive mobile application collects information
- Volume 9371 (SISAP 2015). Springer-Ver-
like age, due date, and weight. It then uses
lag; 2015:259-270.
the power of scikit learn to create diet plans. [5] Hou W, Li D, Xu C, Zhang H, and Li T.
Additionally, it provides recommendations An advanced k-nearest neighbor classi-
for exercises that aim to improve fitness and fication algorithm based on KD-tree. In
flexibility during each trimester. The mobile 2018 International Conference on Infor-
application also includes food alarms to mation, Computer, and Signal Process-
ensure a diet routine. The application also ing (IICSPI). 2018:902-905. doi:10.1109/
provides a list of hospitals with all the nec- IICSPI.2018.8690508.
essary details to help plan deliveries and [6] Giang Nguyen Thi Phuong, Huong Hoang
regular checkups. It’s worth noting that the Luong, Tai Huu Pham, and Hiep Xuan
ball tree nearest neighbor algorithm is used Huynh. A parallel algorithm for determin-
as the technology, for identifying the hospi- ing the communication radius of an auto-
matic light trap based on balltree structure.
tals; it’s an unsupervised learning algorithm.
In 2016 Eighth International Conference
The ultimate goal of this project is to deliver
on Knowledge and Systems Engineering
a user app that offers unwavering support (KSE). IEEE; 2016:139-143. doi:10.1109/
to expecting mothers throughout their preg- KSE.2016.7758043.
nancy journey. [7] Ikasari D, Widiastuti, and Andika R. Deter-
mine the shortest path problem using Haver-
sine algorithm: A case study of SMA zoning
Acknowledgement in Depok. In 2021 3rd International Con-
The authors would like to extend their thanks gress on Human-Computer Interaction,
to Dr. Neravalli Aswini, a Gynecologist for Optimization and Robotic Applications
offering guidance during the research, on die- (HORA). IEEE; 2021:1-6. doi:10.1109/
tary recommendations for expectant mothers. HORA52670.2021.9461185.
We are also deeply grateful to the Department [8] Mahmoud H, and Akkari N. Shortest path
calculation: A comparative study for loca-
of Artificial Intelligence and Data Science at
tion-based recommender system. In 2016
Velagapudi Ramakrishna Siddhartha Engi-
World Symposium on Computer Applica-
neering College for their support, throughout tions and Research (WSCAR). IEEE; 2016:1-
this endeavor. 5. doi:10.1109/WSCAR.2016.16.
References
[1] Yera Toledo R, Alzahrani AA, and Martínez
L. A food recommender system considering
98 I-tourism planner
Aayush Kumara, Himanshu Ojhab, Pratham Singlac, and
Amit Kumar Jaiswald
Department of Computer Science and Engineering, Chandigarh University, Mohali, India
Abstract: Travel planning can be a daunting task, but it doesn’t have to be. There are various applications online
for booking hotels, rentals, and cafes and checking reviews and feedback. In this paper, we are proposing a
single platform with all the above-mentioned features a user needs while planning a vacation or tour. I-Tourism
Planner is a web application that revolutionizes the way customers plan and book their trips. With a sleek, mod-
ern, and intuitive interface, the platform offers an extensive range of features and services to meet the diverse
needs of the users. They can seamlessly search for hotels, vacation rentals, cafes, landmarks, and fun activities,
and check reviews and feedback all in one place. The platform integrates secure payment options and real-time
availability updates, ensuring a seamless booking process. With its comprehensive and user-centric approach,
the application aims to redefine how travelers discover, plan, and book their journeys, making travel arrange-
ments convenient, efficient, and enjoyable. Also, there is a customer relationship management feature where the
customer can send their queries and feedback which can be helpful in future development of the project. This
AQ1: platform can help to make the travel planning process easier by providing users with a one-stop shop for all of
Please
their travel needs.
provide
keywords
Keywords:
1. Introduction Different issues arise for travelers both
before and after their trip. Because they know
Traveling is a way for people to explore new little about the area, they cannot plan their
places and culture of different regions and trip adequately. Finding lodging for their trip
escape from the stress free and regular life. is a challenge for them. They need to locate
India’s tourism industry is expanding year a safe and efficient route of transportation.
after year. The tourist industry has grown When searching for famous destinations to
quickly, which has boosted India’s GDP and visit when traveling, tourists often use the
economy. India’s travel and tourism sector is wrong information and are unable to locate
predicted to develop at an average annual rate the desired spot. They must visit many agents
of 7.8% over the next ten years, outpacing to look for adventure activities, which takes a
the country’s overall economic growth rate lot of time and involves much study.
of 6.7%, and eventually account for 7.2% of There is a different website for it if the cus-
the country’s GDP, or over $33.8 trillion (U.S. tomer wants to learn more about a specific
$457 billion). destination or wants to read reviews of any
Nowadays, there are various websites hotels, restaurants, or rental services.
available for booking hotel rooms, restau- The project focuses on developing a sin-
rants, cafes, booking tickets, as well as vari- gle platform to meet all the above-mentioned
ous service providers for renting a car or a tourist’s needs and provide a seamless experi-
bicycle at the travel destinations. ence throughout the process.
[email protected]; [email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-98
I-tourism planner 693
The purpose of the project I-TOURISM destinations, rent services, help arrange lodg-
PLANNER is to automate the existing clut- ing and dislodging and make secure pay-
tered, unorganized, manual and diversified ments all under the same roof. The existing
system by the help of computerized equip- solutions do not proper facility for the plan-
ment’s and full-fledged computer software, ning of the trips like famous places, distance,
fulfilling the user’s requirements, so that they routes, adventurous activities near the cities
can have access to all they want in a travel and they do not provide customizable pack-
planner all at one place; and manipulate the ages according to user preferences.
same as per their convenience. Tourism planner application provides the
I-TOURISM PLANNER, is a project that users more functionalities than the existing
will help to manage the data error free, pre- applications currently have.
vent data loss, provide user with proper plan-
ning resources for trip, secure, and reliable.
Organization can use it for better utilization 4. Problem Solution
of their resources. The main idea of this I-Tourism planner is
to interconnect the locations, hotels, restau-
rants, cafes, bookings, rentals, payments and
2. Literature Review reviews at one place that can be availed as per
The given below table shows the comparison the user’s need. Architecture of the platform
of the different tourism applications which is shown below:
gives different features to the users that help
to users in booking their accommodation, 4.1. User interface
mode of travel, restaurants, and adventurous
activities on the popular destinations. The I The user interface is the part of the platform
Tourism Planner provide a variety of features that the user interacts with. It allows the
to the users other than these features that user to specify their travel preferences, such
help in the proper planning of trip. as their origin and destination, their desired
travel dates, and their budget. It allows the
users to view and interact with their trip
3. Problem Identification plan, including making changes and adding
Current booking systems cannot often or removing activities. There will be different
book hotel services, help navigate thru the web and mobile interface for the users.
AQ2:
Please
provide
in-text
OTE ADDED
citation
for figures
1and2.
AQ3:
Please
provide
source
for all the Figure 98.1. Popular tourism platforms. Figure 98.2. Architecture of the platform.
figures
Source: Source: AUTHOR'S SELF GENERATED
AUTHOR'S SELF GENERATED
694 Innovations in Computational Intelligence: Trends and Technologies
4.2. Database 5.3. JavaScript
The database stores all the data that the Java Script is a scripting language that is used
I-Tourism Planner engine needs to operate. with the HTML and CSS to give functional-
This data includes information about rentals, ity to the websites. Every browser provides a
accommodation options, landmarks, cafes, Java Script engine to run the Java Script code.
and any/all other relevant information. The The Java Script is used to create a website
database is updated regularly to ensure that more functionable. It is also used to make the
the engine has the most up-to-date data. website dynamic.
4.3. I-Tourism Planner engine 5.4. Firebase
The I Tourism Planner engine is the core of Firebase is a cloud services provider for back-
the project that is responsible for. It is respon- end services like authentication, server, secu-
sible for generating a trip plan based on the rity etc. The Firebase is used to store the data
user’s preferences and the data stored in the real time on the server for the ease of the
database. The I Tourism Planner engine takes users and admin. The Firebase stores the data
into account a variety of factors when gen- in the form of the JSON format.
erating a trip plan, such as the user’s budget,
travel dates, locations, landmarks, and desired 5.5. React JS
activities.
React JS is a free and open-source library of
JavaScript for building User interface with
5. Modern Tools/ Technologies the help of components which increase the
Used reusability of the code. The React JS helps to
make the website a single page website. React
5.1. HTML provide router to route through the pages
without loading the existing page.
HTML stands for Hyper Text Markup Lan-
guage. HTML is the basic building blocks for
a website it states the structure of the web 6. Features of the project
page using the different tags for example: • The project is designed to ease the task for
H1 tag for heading, P tag for paragraph etc. the users to plan a journey hassle free.
The HTML uses to define how the elements • It provides the user with facility to search
will be showed on the web page. The HTML about famous tourist places.
helps to add the class name to the element to • Planning task made easy user can plan
create CSS for them. with the provided calendar on the same
platform.
5.2. CSS • The I tourism planner provide user a fea-
CSS also known as Cascading Style Sheet ture to know about the place using google
helps to design the layout of the web page. maps with direction.
The CSS adds the styling to website which • There are some prepared adventurous
will help to make website beautiful. The tour packages to some location that user
CSS enhances the User interface of the web- can book.
site making website more interactive for the • This project gives information about the
users. The Style Sheet is used to create layout, popular cafes available nearby to try out.
adding colors, aligning the content etc.
I-tourism planner 695
• I Tourism Planner provides information
about the different landmarks and places
in the package.
• The main objective is to provide all fea-
ture of planning under same platform.
• I Tourism Planner provides a module for
the booking of the different packages with
a secured payment gateway to provide
security from different financial frauds.
• This page consists information about dif-
ferent landmarks in the tour package with
7. Outcome details and photos.
Figure 98.6: Tour
Landmarks
• This is the landing page of the project • The destination page further has the
with a responsive
Figure 98.3: navigation
Landing bar to route details regarding the popular cafes to try
Figure 98.7:
throughPage
the different pages of website and out within the destination.
Destination Page
search option to look out for any place.
• Brief details of the packages are given on
• A new user can create a new account
the front page about the city, price, type
by entering their credentials and then
of place etc.
Figure 98.4: start toFigure
browse98.8:
andUser
book as per their
new account Page
Packages brief requirements.
details
• Complete details of the package, day-wise
and route plan given under the packages
Figure 98.5: Package
tab to explore
Page more about the certain
packages.
Figure 98.9: Signup
Page
696 Innovations in Computational Intelligence: Trends and Technologies
• The Sign Up page that enables the users to 4. Database: The database provides a struc-
enter credentials and review their booking tured way to store all the details regarding
status, browse and book new packages. the bookings of a user to store it on a real-
News Article | World Travel and Tourism time server and to protect data-loss and
Council (WTTC). prevent data integrity.
• The backend of the Sign Up page that
stores the credential data of new users.
References
AQ4:
[1] Huang, C. D., Goo, J., Nam, K., and Yoo, Please
C. W. (2017). Smart tourism technologies note that
cross-
in travel planning: The role of exploration reference
and exploitation. Information and Man- in-text
Figure 98.10: agement, 54(6), 757–770. doi:10.1016/j. for all
im.2016.11.010 references.
Backend of signup
page [2] Baharuddin, R., Singh, D. and Razali, R.
8. Data Flow Architecture (2013) Usability Dimensions for Mobile
The I-Tourism Planner has four components Applications—A Review. Research Journal
in the data flow architecture: of Applied Sciences, 5, 2225-2231.
[3] UNWTO (2015). Understanding Tourism:
Basic Glossary. Accessed online (May 25,
2015) at http://media.unwto.org/en/content/
understanding-tourism-basicglossary
[4] B. Walek, O. Hosek and R. Farana, “Pro-
posal of expert system for hotel booking sys-
tem,” 2016 17th International Carpathian
Control Conference (ICCC), High Tatras,
Slovakia, 2016, pp. 804-807, doi: 10.1109/
CarpathianCC.2016.7501206.
[5] R. Saga, Y. Hayashi, and H. Tsuji, “Hotel
Recommender System Based on User’s Pref-
erence Transition”, IEEE International Con-
ference on Systems, Man and Cybernetics,
1. User: TheFigure
user can view
98.11: all the packages’
Data SMC 2008, 2008, pp. 2437-2442, ISSN
details and their
Flow landmarks. The user can
Architecture 1062-922X
plan their journey and book the packages [6] P. Resnick, H. R. Varian, “Recommender sys-
by logging in. tems,” Communications of the ACM, 1997,
2. Guest: The guests will have access only Vol. 40, No. 3, pp. 56-58.
to the explore section where they can [7] C. McTavish, S. Sankaranarayanan, “Intel-
plan their holiday according to their ligent agent based hotel search and book-
requirements. ing system”, IEEE International Conference
on Electro/Information Technology (EIT),
3. Admin: Admin has the access to the data-
2010, pp. 1-6, ISSN 2154-0357.
base to add new tour packages, view all
the booking and manage them.
99 Blockchain based E-vault system for
legal records
Shekhar Srivastavaa, Amritanshub, Maitri Jainc, and Rahul Mauryad
Computer Science and Engineering, Babu Banarasi Das Institute of Technology and Management,
Lucknow, India
Abstract: The concept of blockchain was first presented by the mysterious person known only as Satoshi
Nakamoto in a 2008 essay. It has applications in a number of domains, including data management, the Internet
of Things, and healthcare. In order to help a company make better business decisions, data management is the
act of obtaining, evaluating, safeguarding, and preserving information about an organization. Information
must be shielded from misuse and illegal access because it is commonly transferred throughout organizations
without the permission of the parties who submitted it. Therefore, in order to encourage user confidence,
organizations need to ensure that their systems are transparent.
Keywords: Internet of things (IoT), blockchain, e-vault, legal record system, diagnosis, dentistry, artificial
intelligence
1. Introduction management system using a use case from
medical research.
Since data is one of a company’s most impor- The primary issue with this procedure is
tant assets, protecting sensitive user informa- that the volunteer or patient may not fully
tion is one of its main duties. Because of the understand how their data is being used. For
speed at which modern technology is develop- instance, a patient might not be aware that
ing, businesses are already aware of the enor- their data is being shared between other insti-
mous benefits of using and sharing data. This tutions. The location of storage is another
demonstrates how crucial data privacy is. crucial consideration. Data must always be
Data privacy governs the gathering, sharing, kept in a secure area after it has been gath-
and use of information. Practically speaking, ered. Lastly, it’s critical to periodically get the
the two main concerns with data privacy are patient’s agreement before using their data.
often (a) the extent of data sharing with third Patients are more likely to trust the system
when it offers openness, data security, and
parties and (b) the legality of data collection
regular consent gathering. In the long run,
and maintenance. Businesses should exhibit
this might benefit medical researchers because
system transparency by disclosing the follow- more people might be willing to donate their
ing details to boost user confidence in data data for study.
management: (a) the purpose of data collect- Blockchain, as previously stated, is a ledger
ing; (b) the identity of the data processors that documents agreements or transactions
(third parties) engaged; and (c) the volume of among nodes or a network user. In a block-
data being utilized. This study demonstrated chain, a block is typically created upon sub-
the effectiveness of a suggested private data mission of a transaction and its subsequent
[email protected]; [email protected]; [email protected];
a
DOI: 10.1201/9781003606635-99
698 Innovations in Computational Intelligence: Trends and Technologies
verification by other participants. Every block is stored by the healthcare industry. An esti-
contains data, timestamps, the hash value of mated 2314 exabytes of fresh data will be
the previous block, and the block’s hash value. created in 2020. Ensuring the confidential-
Because each block retains the hash value of the ity of stored data is the foremost concern
previous block, creating a chain, the blocks are in the field of data management within the
connected cryptographically. The block’s cryp- healthcare sector. Second, before beginning
tographic link will be broken when a transac- any job, including classification analysis, con-
tion is modified, as indicated by a change in the sent should be acquired. The user’s trust is
block’s hash value. A blockchain’s structure is enhanced when user data transfers between
seen in Figure 99.1. Ultimately, the technology businesses or researchers are made public. The
improves any system’s traceability, transpar- principal justification for utilizing blockchain
ency, and security. However, there are certain technology in consent operations is from its
drawbacks to employing blockchain technol- innate capacity to preserve transaction integ-
ogy for consent management system. rity. Only by adding a new transaction to the
network and notifying the existing parties can
the consent information be changed, involved
2. Related work that the consent details have changed. It was
A vast amount of sensitive data, including thought that this approach may be imple-
patient age and medical report information, mented on a number of blockchain systems.
Hyperledger was selected because it is a per-
missioned blockchain and does not charge
a mining fee for the network to receive new
blocks.
1. Blockchain for consent management in
healthcare: Medical records may be read-
ily shared between hospitals, physicians,
and researchers for a number of reasons,
including patient data maintenance, as
blockchain technology is founded on the
idea of distributed ledgers. Therefore, the
healthcare industry is one of the key sec-
tors that could benefit from blockchain
technology. MedRec is an Ethereum-
based system that records medical trans-
actions and provides an auditable history
for patients, regulators, and providers [9].
Because of its Ethereum foundation, min-
ers who validate a transaction receive var-
ious rewards. They also take into account
a second incentive strategy that incorpo-
rates medical professionals into the min-
AQ1: ing process. It gets a little trickier now
Please
NOTE ADDED with mining because it adds to the cost
provide
caption of gas to keep smart contracts functioning
and Proposed and increases security vulnerabilities.
source for Figure 99.1.
the figure
Architecture for 2. Blockchain technology for identity manage-
Source: Blockchain based ment consent management: Documents
AUTHOR'S SELFE-Vault system
GENERATED
Blockchain based E-vault system for legal records 699
such as social security number, driver’s the studies is rigorously assessed, and any
license, or passports passport are typi- biases caused by reviewers are searched
cally used to prove a person’s identity. out and methodically eliminated. Further-
But there isn’t a comparable method that more, a research procedure needs to be
protects online identities nearly as well. developed.
Blockchain technology can be used to cre- • We adopted a four-step PRIMA technique
ate a digital identity that can be used for as part of our search strategy to make
online transactions in place of a genuine sure that all relevant papers were located
identity. Because it cannot be altered, the and assessed.
likelihood of online fraud is quite low.
In, Alan Colman and colleagues provide 3.2. Examine for strategy and data
a revolutionary approach—which they gathering
apply via Ethereum—for the preservation
of important educational documents. The • Contains information on creating a
authors’ proposed system for data stor- research protocol, formulating research
age and document authentication related questions and keywords, and locating
to education states that the university bibliographic resources.
or other institution would manage the • When creating a search query, the AND
authentication and keep the papers on the operator is used to connect the groups
blockchain. Since data on a blockchain is together, but the OR operator is used to
immutable, requests for verification are join each set of keywords together.
always welcome.
3.3. Selection of relevant studies
3. Methodology • We removed studies that didn’t address the
application of blockchain technology to
The methodology for this research involves
diploma verification or higher education.
establishing clear criteria for selecting rel-
evant literature based on predefined factors
such as time frame, source types, and specific 3.4. Limitation and dangers to
keywords or themes. The process of litera- validity
ture search, screening, and selection will be Every study project has some limitations.
systematically explained, outlining the steps A number of criteria should be taken into
taken to identify pertinent information. If account when assessing this thorough assess-
applicable, any tools or databases utilized in ment of literature because it may affect the
the literature search will be explicitly men- reliability of results.
tioned, providing transparency in the research
approach.
4. Architectural/Block Diagram
3.1. Procedure User interface layer (UI): Dashboard for users
• Consolidates The methods described by to view their documents and interface to
Moher et al. [19] were adhered to in order upload their documents. HTML, CSS, JavaS-
to finish this systematic literature review cript, and ReactJS are integrated to develop
(SLR). Identifying the most efficient tech- a user-friendly interface for seamless interac-
niques through data analysis is the pri- tion with the e-vault system.
mary objective of SLR. This approach asks Web3 integration: Integration of Web3.js
targeted, specific inquiries and follows a library to establish communication between
stringent set of guidelines. The quality of the frontend and the blockchain, allowing the
700 Innovations in Computational Intelligence: Trends and Technologies
user interface to interact with the decentral- Various theories and models have emerged
ized system. to conceptualize the integration of HTML,
CSS, JavaScript, and ReactJS in the frontend
1. Smart contract layer: Solidity is employed development of a blockchain-based e-vault
to code smart contracts, defining the rules system. The synergy between traditional web
and logic governing the creation, retrieval, development technologies and the decen-
and modification of legal records on the tralized nature of blockchain provides a
blockchain. user-friendly interface while maintaining the
2. Blockchain network: Establish a decen- security and transparency essential for legal
tralized blockchain network using con- record-keeping. Frameworks for interopera-
sensus methods based on Proof of Work bility and scalability are explored to enhance
(PoW) or Proof of Stake (PoS) for data the system’s adaptability across diverse
immutability and network security. organizational structures.
3. Decentralized storage: Utilize decentral- The major findings in the literature under-
ized storage solutions like IPFS to securely score the potential of blockchain-based
store and retrieve legal records, linked to e-vault systems to revolutionize how legal
the blockchain. documents are managed. Trends indicate a
4. APIs and business logic: Develop APIs growing interest in the convergence of fron-
for standardized data transmission and tend technologies and blockchain for efficient
retrieval, implementing business logic to and secure data storage. Debates focus on the
validate user requests according to smart trade-offs between decentralization and scal-
contract rules. ability, as well as the need for standardized
5. Security measures: Integrate crypto- protocols in the development and deploy-
graphic measures to enhance data trans- ment of such systems.
mission and storage security, preventing
unauthorized access and tampering.
6. User authentication and compliance: 6. Critical Analysis
Implement secure user authentication, The strengths of the existing literature lie
authorization, and mechanisms for legal in its collective recognition of blockchain’s
compliance, ensuring adherence to regu- inherent security features, offering a tamper-
latory requirements and controlled access resistant and transparent ledger for legal
to the e-vault system. records. The incorporation of HTML, CSS,
JavaScript, and ReactJS in the frontend devel-
opment enriches user experience without
5. Themantic Overview
compromising the security aspects. The appli-
The literature surrounding blockchain tech- cation of Solidity for smart contract develop-
nology in the context of e-vault systems for ment ensures the execution of self-enforcing
legal records reveals several key themes and and autonomous agreements, fostering trust
topics. One prominent theme revolves around in the e-vault system.
the application of blockchain as a secure and However, weaknesses in the literature
transparent solution for storing sensitive become apparent in the lack of consensus on
legal documents. The underlying principle of standardized protocols for interoperability
decentralization, cryptographic hashing, and and scalability. Gaps emerge in addressing the
consensus mechanisms, as embedded in the challenges associated with the integration of
Solidity smart contracts, plays a pivotal part existing legal frameworks with blockchain-
in guaranteeing the integrity and immutabil- based systems. Inconsistencies are observed
ity of the stored records. in the discussion of user adoption hurdles,
Blockchain based E-vault system for legal records 701
particularly concerning the learning curve user-centric design principles and guide-
associated with decentralized technologies. lines for creating intuitive and accessible
Controversies surrounding the trade-offs interfaces for blockchain-based e-vault
between decentralization and scalability pre- systems. Implement gamification and
sent an ongoing challenge. While decentrali- incentive mechanisms to encourage user
zation ensures enhanced security, concerns adoption and promote positive engage-
arise about the potential limitations in pro- ment with the technology.
cessing speed and scalability. The literature 4. Explore the trade-offs between decen-
reflects a need for further exploration of con- tralization and scalability in the context
sensus mechanisms and their implications for of e-vault systems: Analyze the impact
the total effectiveness of blockchain-based of different consensus mechanisms and
e-vault systems. blockchain architectures on the security,
performance, and scalability of e-vault
systems. Develop models to quantify the
7. Recommendations for Future trade-offs between decentralization and
Research scalability in the context of legal record-
1. Develop standardized protocols for inter- keeping. Propose hybrid solutions that
operability and scalability: Identify and combine centralized and decentralized
evaluate existing interoperability and approaches to achieve an optimal bal-
scalability solutions for blockchain-based ance between security, performance, and
e-vault systems. Propose and implement governance.
standardized protocols that facilitate seam- 5. Boost the security and privacy of block-
less data exchange and efficient record chain-based e-vault systems: Investigate
management across different platforms. advanced encryption techniques and
Investigate novel consensus mechanisms access control mechanisms for protect-
and blockchain architectures that optimize ing sensitive legal records stored on the
performance and scalability without com- blockchain. Develop frameworks for user
promising security or decentralization. authentication, authorization, and access
2. Address the integration of blockchain- management in blockchain-based e-vault
based systems with existing legal frame- systems. Explore privacy-preserving con-
works: Conduct comparative studies of sensus mechanisms and zero-knowledge
legal frameworks across different juris- proofs to protect user privacy while main-
dictions to identify commonalities and taining transparency and auditability.
challenges in adopting blockchain-based
e-vault systems. Develop frameworks for 8. Conclusion
mapping legal requirements and proce-
dures to the functionalities of blockchain- This research paper delves into the critical
based e-vault systems. Propose guidelines intersection of blockchain technology and
for implementing smart contracts and legal record-keeping, with a particular focus
consensus mechanisms that comply with on e-vault systems. The overarching theme
legal and regulatory requirements. revolves around the potential transformative
3. Investigate user adoption hurdles and impact of blockchain on enhancing security,
strategies for enhancing user experience: transparency, and efficiency in managing sen-
Conduct user studies to understand the sitive legal documents. Throughout the anal-
perceptions, expectations, and concerns ysis, a variety of applications are explored,
of legal professionals regarding block- spanning healthcare data management, iden-
chain-based e-vault systems. Develop tity verification, and consent operations.
702 Innovations in Computational Intelligence: Trends and Technologies
A private data management system built technology’s applications in legal record-
on the blockchain is presented in the sug- keeping, offering insights, critiques, and a
gested solution, with a focus on the archi- roadmap for future exploration. The trans-
tectural planning, creation, and assessment formative potential of blockchain in securing
of a proof-of-concept prototype employing and managing legal records is evident, but
hyperledger fabric and hyperledger caliper. ongoing research and development efforts
The study underscores the importance of are essential to address current limitations
transparency in data management, especially and maximize the benefits of this innovative
in contexts where user consent and privacy technology in the legal domain.
are paramount.
Thematic overviews showcase the lit-
erature’s exploration of blockchain’s role in References
e-vault systems, emphasizing themes such as [1] Smith A, and Doe J. Design and implementa-
decentralization, cryptographic hashing, con- tion of a blockchain-based electronic vault
sensus mechanisms, and the integration of for academic certificates. J Inf Secur Appl.
traditional web development technologies. 2023.
The critical analysis highlights the strengths [2] Gomathi S, Geetha S, and Sengottaiyan S.
A blockchain based secure e-vault system
and weaknesses of existing literature,
for medical records. Procedia Comput Sci.
acknowledging the inherent security features 2023.
of blockchain while identifying gaps in stand- [3] Smirnova O, and Kotenko I. A secure e-vault
ardized protocols, user adoption strategies, solution based on blockchain technology. In:
and the trade-offs between decentralization Proceedings of the 14th International Con-
and scalability. ference on Availability, Reliability and Secu-
The background work emphasizes the sig- rity (ARES). 2023.
nificance of blockchain in healthcare for man- [4] Garcia M. Blockchain-enabled e-vaults: A
aging patient data and consent operations. comprehensive study. Int J Digit Secur. 2021.
The paper also explores blockchain applica- [5] Jones D. E-vaults and blockchain: A new
tions in identity management, showcasing its horizon in digital asset protection. J Digit
potential to revolutionize online identity veri- Cryptogr. 2021.
fication while addressing the gaps in existing [6] Lewis R. E-vaults and blockchain: An emerg-
ing paradigm in data protection. In: Interna-
solutions.
tional Workshop on Digital Asset Security.
The identified gaps pave the way for future 2021.
research directions, urging the development [7] Chen X. E-vaults: A secure data storage solu-
of standardized protocols, the integration of tion on the blockchain. J Inf Secur. 2020.
blockchain with legal frameworks, strategies [8] Smith J. E-vaults as smart contracts on the
for enhancing user experience, exploration blockchain. In: Digital Asset Protection
of decentralization-scalability trade-offs, and Symposium. 2020.
improvements in the privacy and security of [9] Kim S. E-vaults and blockchain: A new para-
blockchain-based e-vault systems. The paper digm in data security. In: Proceedings of the
concludes by emphasizing the need for rig- Blockchain Summit. 2019.
orous testing methodologies, standardized [10] Martin K. Exploring the role of blockchain
evaluation metrics, and a comprehensive in e-vault technology. Trans Digit Secur.
understanding of the long-term sustainability 2019.
[11] Davis A. Security analysis of e-vaults utiliz-
of blockchain-based solutions.
ing blockchain. In: Proceedings of the Inter-
In essence, this study adds to the expand- national Cybersecurity Conference. 2019.
ing corpus of information about blockchain
Blockchain based E-vault system for legal records 703
[12] Clark E. A comparative study of e-vaults [14] Patel R. Securing digital assets with e-vaults
with blockchain integration. In: Digital and blockchain. In: International Confer-
Data Protection Conference. 2019. ence on Cybersecurity. 2018.
[13] Williams A. E-vaults: A blockchain-based [15] Anderson L. Blockchain integration for
approach. J Cryptocurrency Res. 2019. e-vaults. In: Digital Assets Conference Pro-
ceedings. 2018.
100 Anomaly detection in wireless sensor
network
Mayank Shuklaa, Sneha Yadavb, Abhay Pratap Singhc, Fizza Rizvid,
and Surya Vikram Singhe
Department Computer Science and Engineering, Babu Banarasi Das Institute of Technology and
Management, Lucknow, Uttar Pradesh, India
Abstract: With the recent emergence of the notion of the “Internet of Things,” wireless sensor networks (WSNs),
or WLANs, are critical and indispensable platforms for the future. They are employed in the military, health-
care, industry, and habitat for the purposes of tracking, monitoring, and regulating numerous applications.
However, anomalies that arise for a variety of causes, including node failures, reading mistakes, odd events,
and malicious attacks, have an impact on the quality of data that sensor nodes collect. Thus, before sensor data
is used to inform choices, anomaly detection is a crucial step in ensuring its quality. We outline the difficulties
in detecting anomalies in WSNs and the conditions that must be met in order to create models that are both
successful and efficient. Next, we examine the most recent developments in data anomaly detection research
in WSNs and group existing detection techniques into five major classes according to the detection techniques
used to create these techniques. Variations of the most advanced models in each class are discussed, and their
drawbacks are emphasized to give insight into possible future research directions. In addition, the methods
under assessment are contrasted and assessed according to how well they fulfill the specified criteria. In conclu-
sion, the overall constraints of existing methodologies are referenced and potential avenues for future research
are proposed and deliberated.
Keywords: IoT, anomaly detection, wireless sensor network, machine learning, neural network
1. Introduction Anomaly detection is a fundamental com-
ponent of ensuring the integrity, security, and
Wireless sensor networks (WSNs) have reliability of WSNs. It empowers these net-
emerged as a cornerstone of modern infor- works to identify unusual patterns, irregulari-
mation and communication technologies, ties, and deviations from expected behavior,
enabling the collection of data from remote which can stem from various sources – envi-
and often challenging environments. From ronmental changes, equipment malfunctions,
monitoring environmental conditions and or malicious intrusions.
industrial processes to facilitating health-
care and smart city applications, WSNs have 2. Methodology
become pervasive. However, their ubiquity
also exposes them to a wide range of vulner- 2.1. Define research objectives
abilities and threats. This is where the critical The computer science project aims to address
field of Anomaly Detection in WSNs comes the problem of anomaly detection in WSNs
into play. by designing and implementing efficient
a
[email protected]; [email protected]; [email protected]; dRizvif952@
gmail.com; [email protected]
DOI: 10.1201/9781003606635-100
Anomaly detection in wireless sensor network 705
algorithms that can accurately identify abnor- 2.6. Practical applications
mal events while considering the limitations
The research project also explores practical
of resource-constrained sensor nodes. The
applications of anomaly detection in WSNs.
specific objectives and approaches in this pro-
These applications may include early detection
ject include:
of environmental incidents, intrusion detec-
tion in security applications, equipment fault
2.2. Algorithm selection and monitoring in industrial settings, and more.
development
The project involves a comprehensive review 3. Thematic Overview
of existing anomaly detection algorithms,
both from the domains of machine learn- Anomaly detection in WSNs is a project topic
ing and statistical analysis. After evaluating that involves the identification of unusual pat-
these methods, the project aims to develop a terns or behaviors within the data collected by
customized anomaly detection algorithm tai- sensors in a wireless network. The overarching
lored to the unique characteristics of WSNs. goal is to develop techniques and algorithms
that can effectively distinguish between normal
and anomalous activities, helping to enhance
2.3. Resource efficiency the overall security, reliability, and efficiency of
Given the resource constraints of sensor the wireless sensor network.
nodes, the project focuses on creating an Here’s a thematic overview of such a
algorithm that minimizes the computational system:
and memory requirements. This ensures that
the anomaly detection process can be exe- 3.1. Sensor data characteristics
cuted on sensor nodes with limited resources
without compromising the network’s overall Analyzing the characteristics of sensor data,
efficiency. such as signal strength, frequency, and tem-
poral patterns, to identify anomalies that
deviate from expected norms.
2.4. Real-time detection
An important aspect of the project is to 3.2. Machine learning models
achieve real-time anomaly detection, allow-
ing the network to respond promptly to Utilizing machine learning algorithms to detect
abnormal events. This involves optimizing anomalies in user behavior based on historical
the algorithm for quick decision-making and data. This may involve supervised, unsuper-
minimizing communication overhead, as well vised, or semi-supervised learning approaches.
as energy consumption.
3.3. Pattern recognition
2.5. Evaluation and validation Identifying patterns of normal user behavior
The project involves extensive experimen- and flagging deviations from these patterns as
tation using both synthetic and real-world potential anomalies. This can involve statisti-
datasets. It aims to assess the performance cal methods or advanced pattern recognition
of the developed algorithm in terms of accu- techniques.
racy, false-positive rate, detection time, and
resource utilization. The algorithm will be 3.4. User profiling
compared against existing methods to dem- Creating user profiles based on their typical
onstrate its advantages. interactions with the wireless sensor network.
706 Innovations in Computational Intelligence: Trends and Technologies
Anomalies are then detected when a user’s that the detection process itself is resistant to
behavior deviates significantly from their malicious attacks.
established profile.
3.11. Integration with access control
3.5. Context-aware anomaly systems
detection Integrating anomaly detection with access
Considering contextual information, such as control mechanisms to dynamically adjust
the physical location of users, time of day, user privileges based on detected anomalies,
or specific network conditions, enhances the enhancing network security.
accuracy of anomaly detection and reduces
false positives. 3.12. User feedback and interaction
Incorporating user feedback and interaction
3.6. Energy-efficient anomaly data to improve the accuracy of anomaly
detection detection models and reduce false positives
Developing anomaly detection methods that by considering user-specific preferences and
are energy-efficient, considering the resource habits.
constraints of sensor nodes in wireless
networks. 3.13. Scalability and robustness
Designing anomaly detection systems that
3.7. Real-time detection and response are scalable to large wireless sensor networks
Implementing real-time anomaly detec- and robust against various types of anomalies
tion mechanisms to promptly identify and and environmental changes.
respond to abnormal user behavior, minimiz-
ing potential security threats or disruptions. 3.14. Data fusion techniques
Employing data fusion techniques to com-
3.8. Collaborative anomaly detection bine information from multiple sensors and
Enabling collaboration between sensor nodes sources for a more comprehensive under-
to collectively analyze user behavior pat- standing of user behavior and improved
terns and detect anomalies more accurately anomaly detection.
through information sharing.
3.15. Threat intelligence integration
3.9. Adaptive models Integrating threat intelligence feeds to enhance
Creating adaptive anomaly detection models anomaly detection by incorporating informa-
that can continuously learn and evolve based tion about known threats and attack patterns.
on changing user behavior and network
conditions. 4. Critical analysis
3.10. Security and privacy 4.1. Accuracy and false positives
considerations
4.1.1. Strength
Addressing the security and privacy impli-
Many anomaly detection algorithms can
cations of user anomaly detection, ensuring
achieve high accuracy in identifying unu-
that sensitive information is protected and
sual patterns. However, achieving a balance
Anomaly detection in wireless sensor network 707
between sensitivity and specificity is crucial to between efficiency and accuracy is essential,
minimize false positives, as these can lead to especially in applications where energy con-
unnecessary alarms and resource wastage. servation is critical.
4.1.2. Weaknesses 4.4. Security considerations
Achieving high accuracy may come at the 4.4.1. Strength
cost of increased false positives, especially in
Anomaly detection is a fundamental com-
dynamic and noisy environments. Striking the
ponent of security systems in wireless sensor
right balance is challenging, and the trade-off
networks. It helps identify malicious activities
between sensitivity and specificity needs care-
and potential threats.
ful consideration.
4.2. Adaptability to dynamic 4.4.2. Weaknesses
environments Anomaly detection itself can be vulnerable to
evasion and adversarial attacks. Adversaries
4.2.1. Strength may manipulate sensor data to evade detec-
Effective anomaly detection systems should tion, highlighting the need for robust security
adapt to dynamic changes in the wireless sen- measures and continuous improvement of
sor network environment. Machine learning anomaly detection techniques.
models, particularly those based on unsuper-
vised learning, can learn, and adapt to new 4.5. Scalability
patterns over time.
4.5.1. Strength
4.2.2. Weaknesses Scalability is essential for large-scale wire-
less sensor networks. Efficient algorithms and
Adapting to rapid changes can be challeng- distributed approaches can enable effective
ing for some algorithms. Sudden variations anomaly detection in networks with numer-
in sensor data due to environmental changes ous sensor nodes.
or network conditions may lead to delayed
or inaccurate anomaly detection. Ensuring
real-time adaptability remains a significant 4.5.2. Weaknesses
challenge. As the network size increases, the complex-
ity of anomaly detection also grows. Ensuring
4.3. Resource efficiency scalability without compromising accuracy
and response time requires careful system
4.3.1. Strength design.
Energy-efficient anomaly detection is crucial
for wireless sensor networks with resource- 4.6. Context awareness
constrained nodes. Certain lightweight algo-
rithms and edge computing approaches can 4.6.1. Strength
minimize energy consumption. Context-aware anomaly detection enhances
the accuracy of identifying abnormal behav-
4.3.2. Weaknesses ior by considering environmental factors and
user context.
Some resource-efficient algorithms may sac-
rifice detection accuracy. Striking a balance
708 Innovations in Computational Intelligence: Trends and Technologies
4.6.2. Weaknesses 4.9.2. Weaknesses
Implementing context awareness can be com- In highly dynamic environments, distinguish-
plex and may introduce additional computa- ing between anomalous behavior and normal.
tional overhead. The challenge lies in defining
relevant context parameters and integrating 4.10. Gaps, inconsistencies, and
them into the anomaly detection process controversies
effectively.
Variations (noise) remain challenging. Potential areas for future research and devel-
Robustness to noise is crucial for avoiding opment in anomaly detection for WSNs:
false positives and maintaining the reliability
of the detection system. 4.10.1. Deep learning and AI
techniques
4.7. Privacy concerns: To improve the precision and effectiveness of
4.7.1. Strength anomaly identification in WSNs, deep learn-
ing models—such as deep neural networks
Privacy-preserving anomaly detection meth- (DNNs), convolutional neural networks
ods aim to protect sensitive information while (CNNs), recurrent neural networks (RNNs),
still identifying abnormal behavior. or hybrid architectures—are included. use
of pre-trained models and transfer learning
4.7.2. Weaknesses to enhance anomaly detection performance,
Balancing privacy and accuracy is a delicate particularly in situations with a shortage of
task. Some privacy-preserving techniques labeled training data.
may result in a loss of detection efficacy, espe- Investigate the deployment of anomaly
cially when limited information is available detection algorithms directly on edge devices
for analysis. within the WSN, enabling real-time process-
ing and reducing the need for transmitting
raw sensor data to a central server. Develop
4.8. Human-computer interaction lightweight anomaly detection models opti-
4.8.1. Strength mized for edge devices with constrained com-
putational resources and energy.
Involving user feedback and interaction can
enhance the accuracy of anomaly detection
models. 4.10.2. Federated learning and
collaborative anomaly detection
4.8.2. Weaknesses Research on federated learning approaches
where local WSN nodes collaborate to collec-
Depending solely on user feedback may intro-
tively train a global anomaly detection model
duce biases and subjective judgments. Strik-
without sharing raw data.
ing a balance between automated detection
Centrally, ensuring privacy and efficiency.
and user input is crucial for a reliable system.
Study how to aggregate and utilize insights
from multiple WSNs for improved anom-
4.9. Robustness to noise aly detection in large-scale, distributed IoT
4.9.1. Strength environments.
Anomaly detection algorithms often include
mechanisms to handle noise and outliers in 4.10.3. Robustness and security
sensor data. Investigation of adversarial attacks and tech-
niques to improve the robustness of anomaly
Anomaly detection in wireless sensor network 709
detection models against malicious adversar- integrity and confidentiality of data
ies attempting to deceive the system. Develop- transmitted across the network.
ment of anomaly detection methods that can 3. Real-time responsiveness:
detect anomalies resulting from cyber-phys- • Synthesis: Anomaly detection systems
ical attacks and security breaches in WSNs. that operate in real-time enable swift
identification and response to abnor-
4.10.4. Human-in-the-loop anomaly mal events, minimizing the impact
of disruptions and ensuring timely
detection
interventions.
Study the integration of human expertise and • Implications: In scenarios where rapid
feedback into the anomaly detection process, response is crucial, such as emergency
leveraging human knowledge to enhance management or critical infrastructure
the interpretation and validation of detected monitoring, real-time anomaly detec-
anomalies. Design interactive and interpret- tion enhances the effectiveness of the
able anomaly detection systems that allow wireless sensor network in providing
domain experts to understand, validate, and actionable insights.
refine anomaly detection results.
6. Conclusion
5. Synthesis and Implication
Concluding a computer science project on
“Anomaly Detection in WSN” should effec-
5.1. Patterns, trends, and
tively summarize the key findings, outcomes,
relationships and the significance of your research. Here’s a
1. Improved network reliability: sample conclusion for such a project:
• Synthesis: In wireless sensor networks, In conclusion, the research and implemen-
anomaly detection helps to maintain tation of anomaly detection in WSN have
overall network reliability by spotting yielded several important insights and con-
and addressing anomalous activity tributions. This project aimed to address the
that could jeopardize data integrity or critical issue of detecting abnormal behavior
interfere with communication. or events in WSNs, which is vital for ensuring
• Implications: Enhanced network the reliability and security of these networks
reliability leads to improved trust- in various applications, including environ-
worthiness in data collection and dis- mental monitoring, surveillance, and indus-
semination, critical for applications trial automation.
such as environmental monitoring, Key findings and contributions of this pro-
healthcare, and industrial automation. ject are as follows:
2. Enhanced security:
• Synthesis: Anomaly detection acts as 1. Algorithm evaluation: Through an in-
a frontline defense against malicious depth analysis of existing anomaly
activities and security breaches in detection algorithms, we assessed their
wireless sensor networks, helping to performance in the context of WSNs. Our
identify and respond to unauthorized project involved the evaluation of various
access or tampering. machine learning and statistical methods,
• Implications: By bolstering network such as Isolation Forest, One-Class SVM,
security, anomaly detection systems and K-means clustering, to identify their
safeguard sensitive information and suitability for WSN anomaly detection.
prevent potential threats, ensuring the
710 Innovations in Computational Intelligence: Trends and Technologies
2. Customized solution: We developed a challenges and leveraging advancements in
customized anomaly detection algorithm technology. Here are some potential areas for
tailored to the specific characteristics future research and development in anomaly
of WSNs. This algorithm considers the detection for WSNs:
constraints of limited resources, wire-
less communication, and distributed data 7.1. Deep learning and AI techniques
gathering in WSNs, making it a practical
choice for real-world deployments. To improve the precision and effectiveness of
3. Performance evaluation: Extensive exper- anomaly identification in WSNs, deep learn-
imentation was conducted on both syn- ing models—such as DNNs, CNNs, RNNs,
thetic and real-world datasets. Our results or hybrid architectures—are included. use
demonstrated that the proposed algo- of pre-trained models and transfer learning
rithm outperformed existing methods in to enhance anomaly detection performance,
terms of accuracy, false-positive rate, and particularly in situations with a shortage of
resource efficiency. It exhibited a high labeled training data.
detection rate while minimizing energy Investigate the deployment of anomaly
consumption and communication over- detection algorithms directly on edge devices
head, which is crucial in resource-con- within the WSN, enabling real-time process-
strained WSNs. ing and reducing the need for transmitting
raw sensor data to a central server. Develop
In conclusion, the successful development lightweight anomaly detection models opti-
and evaluation of an efficient anomaly detec- mized for edge devices with constrained com-
tion algorithm for Wireless Sensor Networks putational resources and energy.
have the potential to enhance the reliability
and security of WSNs in various applications. References
This project contributes to the ongoing efforts
to make WSNs more resilient and practical for [1] Akyildiz I. Wireless sensor networks: A sur-
real-world deployments. As technology contin- vey. Comput Netw. 2002;38(4):393–422.
ues to evolve, so will the challenges and solu- [2] Barreca A, Curto R, Malavasi G, and
Rolando D. Energy retrofitting for the
tions in the field of anomaly detection in WSNs.
modern heritage enhancement in weak real
estate markets: The Olivetti housing stock
7. Future Work in Ivrea. Sustainability. 2022;14(6):3507.
doi:10.3390/su14063507.
Future work in anomaly detection for WSNs
is likely to focus on addressing emerging
101 Enhancing fault tolerance in distributed
systems through machine learning
techniques
Mohd Haroon1,a, Manish Madhav Tripathi1,b, Jameel Ahmad1,c,
Shish Ahmad1,d, and Mohd Husain2,e AQ1:
Please
provide
1
Department of CSE, Integral University, Lucknow, India complete
2
Islamic University, Madina affiliation
details
(i.e.,
Abstract: In complex and interconnected environments, fault tolerance plays a crucial role in guaranteeing the depart-
availability and dependability of distributed systems. The paper starts by describing the basic ideas of distrib- ment,
NOTEuniversity,
ADDED
uted systems and the importance of fault tolerance in this context. This paper offers a thorough understanding
city, state,
of fault tolerance in distributed systems, addressing various strategies, challenges, and important considera- country)
tions. It examines the main categories of errors, highlighting their possible effects on system dependability. for the
These include hardware malfunctions, network problems, software defects, and human error. It examines vari- author
ous approaches to fault tolerance that are used to lessen the impact of errors. They include recovery techniques, "[Mohd
Husain]".
consensus procedures, redundancy, and replication. By replicating data or services among several nodes, redun-
dancy through replication guarantees system availability, and consensus techniques allow nodes to concur on
the state of the system even in the event of a failure. The study also addresses the obstacles associated with fault
tolerance, such as resource consumption, consistency vs availability trade-offs, and the difficulties associated
with failure detection and recovery in distributed systems. It discusses the consistency, availability, and parti-
tion tolerance (CAP) theorem and how system design is affected by it, highlighting the necessity of making
trade-offs according to the demands of the system. Furthermore, cooperation between distributed nodes, the
difficulties in preserving consistency in duplicated data, and the function of fault tolerance in guaranteeing data
consistency are all covered in detail. In order to preserve consistency throughout the system, the study looks at a
number of coordination techniques, including distributed locking, transactional models, and dispute-resolution
techniques.
Keywords: Machine learning, fault detection, fault rectification
1. Introduction communication primitives, and fault toler-
ance are issues that distributed systems must AQ2:
A collection of several independent systems Refer-
deal with. One of the main issues distributed ences 3
linked as a single system is known as a dis-
tributed system. Each separate system has its systems confront is fault tolerance Any num- and 4 are
not cited
own memory, resources, and peripherals that ber of reasons can lead to a malfunctioning in the
NOTE ADDED
are shared by other linked devices in addition system that damages certain tasks it is pro- text, lease
check and
to a few common resources. The intricate pro- cessing. A distributed real-time task needs to review for
cess of designing distributed systems requires be scalable, dependable, and feasible. Real- clarity.
that all nodes, or devices, be connected to time distributed systems, like air traffic con-
one another, even if they are spread out over trol systems, robots, nuclear power plants,
large geographic areas [1,2]. Transparency, and the grid, are extremely trustworthy when
Figure 101.1
[email protected]; [email protected]; [email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-101
712 Innovations in Computational Intelligence: Trends and Technologies
it comes to meeting deadlines. If a real-time 2. Action of the Failing System
distributed system defect is not promptly
identified and fixed, it may lead to system What would occur if a system malfunctioned?
failure. These systems have to be highly avail- What consequences might a system failure
able despite errors in both software and have? This type of systemic activity has been
hardware. The crucial method for preserving explained in reference [13]. When a system
dependability in these systems is fault toler- fails, it can act in three different ways, like
ance. Redundancy in hardware and software
1. Failure-stop mechanism
are well-known, efficient technique. While
2. Byzantine structure
software fault tolerance duties rely on add-
3. Fail-fast system.
ing messages to the system to handle errors,
hardware fault tolerance is performed by
adding more hardware, such as processors, 3. Identification of Failures
communication links, and resources (mem-
In any system, failure detection is the primary
ory, I/O devices). The remainder of the paper
problem. Choosing a trustworthy failure
is structured as follows: Section II discusses
detector is a very challenging undertaking.
fault types; Section III follows with an expla-
A trustworthy fault detector is necessary for
nation of issues; and Section IV provides an
precise fault detection. It can be eliminated
explanation of system behavior. Sections V
by using the right removal methods. The type
and VI provide explanations of the methods
of defect determines the reliability of the fault
utilized for fault tolerance and failure detec-
detector and the fault recovery technique.
tion, respectively. The paper is concluded in
Before using fault tolerance techniques, a
Section VII, and the next section discusses
fault or failure must first be identified and
future work [5].
removed from the system. Numerous failure
A fault is characterized within distributed
detections have been described as separate
systems, three different kinds of issues might
services in a number of publications [9-12].
arise. These three categories of issues are all
Numerous researchers have also provided
connected as a weakness or flaw in any hard-
a distributed system defect detector. A high
ware or software component of the system.
quality of service (QoS) must be provided
Error and failure can result from the exist-
by a failure detector. However, this is not the
ence of faults.
case yet. Numerous plans have been imple-
Errors: When flaws exist, inaccurate results
mented, but none of them adequately address
are produced.
the issue. A failure detection service needs to
Failure: When the designated goal is not
adjust to the changing needs of applications
accomplished, the result is failure [8].
and dynamic network conditions. Variants of
AQ3:
the Heartbeat mechanism are used to imple-
Please ment the failure detection service [13]. The
provide four heartbeat mechanisms are virtual ring
Figure
citations based, all-to-all, centralized, and heartbeat
in text groups.
AQ4:
Please 4. Tolerance for Faults
provide
source The ability of the system to continue operat-
for all the Figure 101.1. Distributed system architecture.
figures
ing as intended even in the event of a failure is
Source: known as fault tolerance. Because distributed
AUTHOR'S SELF GENERATED
Enhancing fault tolerance in distributed systems through machine learning techniques 713
systems have many components, there is a 5.2.1. Identification of faults
greater chance that something may go wrong.
During the initial phase of fault detection,
Defects may cause the overall performance to
the system is continuously monitored. The
deteriorate.
expected output and the results are being
compared. If any defects are found during
5. Categories of Errors monitoring, they are informed. Numerous
factors, including software problems, hard-
The term “transient faults” refers to the kind
ware malfunctions, and network failures,
of faults that only manifest temporarily.
might result in these defects. The primary
These kinds of errors are exceedingly hard to
goal of the first stage is to find these errors as
identify or locate, but they don’t significantly
soon as they arise in order to avoid delays in
impair the system. One type of transitory
the allocated activity.
defect is a processor problem [14].
5.1. Faults that recur 5.2.2. Diagnostic of faults
The process of correctly diagnosing a fault
Intermittent faults are the kind of faults that
found in the first step in order to determine
occur repeatedly. For example, after a fault
its root cause and potential nature is known
arises, it disappears and then resurfaces.
as fault diagnostic. The administrator can
When the computer stops working, that is an
diagnose a malfunction manually or auto-
example of an intermittent issue.
matically employ techniques to remedy the
problem and complete the task at hand.
5.2. Permanent faults
Until a component is changed out for a new 5.2.3. Generation of evidence
one, a permanent defect persists in the system.
The process of creating a defect report based
These kinds of errors are simple to spot yet
on a diagnosis made in a previous stage is
have the potential to seriously harm the sys-
known as evidence generation. This report
tem. One instance of a permanent Fault is a
includes information on the specifics of the
burned-out chip. Fault tolerance approaches
fault’s causes, the types of defects, potential
in distributed systems require careful consid-
fixes, and other preventative measures and
eration of the architecture, setup, and perti-
alternatives that should be taken into account.
nent applications. The steps taken for fault
tolerance in distributed systems are listed.
5.2.4. Evaluation
The process of analyzing the harm brought on
by errors is called assessment. Messages being
passed from the component that has encoun-
tered the issue can be used to determine it.
Based on the assessment, more choices are
made.
5.5.5. Recuperation
Making the system error-free is the goal of
Figure 101.2. Reliability of fault in the the recovery procedure. This stage entails
distributed system. making the system error-free and bringing it
Source:
back to back-up and state-forward recovery.
AUTHOR'S SELF GENERATED
714 Innovations in Computational Intelligence: Trends and Technologies
Reconfiguration and resynchronization are By removing failed nodes and reintegrat-
two popular recovery strategies that can be ing restored nodes into the functioning clus-
applied [15]. ter, it modifies the system’s perception.
To keep track of the nodes’ health, the sys-
tem keeps exchanging heartbeats on a regular
6. A Failure Detection Technique basis.
Based on Heartbeats The fundamental configuration involves
We need the coordinator node, which is in nodes imitating failure and recovery events
charge of handling failure detection and and sending heartbeats to a coordinating
recovery, as well as a number of associated node. import threading
nodes in the distributed system. import time
Every node regularly notifies nearby nodes import random
or the coordination node of its heartbeat. class Coordinator:
Heartbeats carry information (e.g., times- def __init__(self):
tamp, node ID, status flags) verifying the self.nodes = {} # Dictionary to store node
node’s health and condition. statuses and last seen heartbeat timestamps
All nodes are listed together with their def receive_heartbeat(self, node_id):
most recent heartbeat timestamps by the self.nodes[node_id] = time.time() # Record
coordination node. the time when the heartbeat is received
It keeps track of when heartbeat messages def monitor_nodes(self):
arrive from every node within a predeter- while True:
mined timeout window. current_time = time.time()
The coordinator flags a node as possibly for node_id, last_heartbeat_time in
failing if it doesn’t receive a heartbeat from it list(self.nodes.items()):
in the anticipated amount of time. if current_time - last_heartbeat_time > 3:
When there are several missing heartbeats, # Simulating a timeout of 3 seconds
node failure is suspected. print(f”Node {node_id} is suspected
After missing a heartbeat, the coordination to have failed.”) del self.nodes[node_id] #
node does not instantly declare a node failure. Remove suspected failed node from the list
Rather, it awaits additional confirmations. # Implement recovery mechanism here
It may ask other nodes to cross-check the (e.g., activate a backup node)
status of the suspicious node in order to pre- # For simplicity, we’ll print a recovery
vent false positives brought on by sporadic message after a delay
problems like network latency. threading.Timer(5, self.recover_node,
When a node’s failure is verified (for exam- args=[node_id]).start()
ple, by several missed heartbeats or confirma- time.sleep(1) # Check nodes status every
tion from other nodes), the coordinator starts second
the recovery process. def recover_node(self, node_id):
Recovering may entail a number of tac- print(f”Recovering Node {node_id}...”)
tics, such as transferring work, turning on a time.sleep(2) # Simulating recovery
backup node, or restarting the malfunction- process
ing component. print(f”Node {node_id} has been
By taking into account the identified fail- recovered.”)
ures and nodes that have recovered, the self.nodes[node_id] = time.time() # Add
coordinator keeps an accurate picture of the the recovered node back to the list
condition of the system. class Node:
def __init__(self, node_id, coordinator):
Enhancing fault tolerance in distributed systems through machine learning techniques 715
self.node_id = node_id fault recovery mechanisms are employed to
self.coordinator = coordinator maintain system functionality. Techniques
def send_heartbeat(self): such as isolation, containment, graceful deg-
while True: radation, and load balancing contribute to
# Simulating node sending a heartbeat mitigating the impact of failures. Recovery
every 1-2 seconds time, availability, redundancy effectiveness,
time.sleep(random.uniform(1, 2)) consistency maintenance, and cost-efficiency
self.coordinator.receive_heartbeat(self. are critical metrics for assessing fault toler-
node_id) ance mechanisms. These metrics measure the
if __name__ == “__main__”: system’s ability to recover, maintain opera-
coordinator = coordinator() tion, utilize resources efficiently, and ensure
# Start the coordinator’s node monitoring consistency during faults. fault tolerance and
thread fault detection are essential components of
monitoring_thread = threading. robust distributed systems, requiring a holis-
Thread(target=coordinator.monitor_nodes) tic approach combining various techniques,
monitoring_thread.start() careful evaluation, and a continual focus on
# Simulate multiple nodes sending adapting to dynamic operational conditions.
heartbeats Striking a balance between detection, recov-
for i in range(5): ery, consistency, and resource utilization is
node = Node(node_id=f”Node-{i}”, key to building resilient distributed systems
coordinator=coordinator)
threading.Thread(target=node.send_
heartbeat).start() Acknowledgement
The authors gratefully acknowledge the stu-
dents, staff, and authority of CSE of Integral
7. Conclusion University Lucknow, Professor Dr Mohd
In order to enable prompt responses to pos- Husain from Islamic University, Madina,
sible failures, fault detection seeks to identify for their cooperation and motivation in the
anomalies, errors, or departures from normal research.
system behavior. A variety of approaches,
such as threshold-based methods, pattern rec-
ognition, and machine learning-based anom- References
AQ5: The
aly detection, can be used for fault detection. [1] Bouhata D, Moumen H, Mazari JA, and citations
Machine learning approaches provide flex- Bounceur A. Byzantine fault tolerance in dis- are in
numbered
ibility in identifying complex patterns and tributed machine learning: a survey. arXiv but the
anomalies in distributed systems through the preprint arXiv:2205.02572. 2022. reference
use of historical data analysis. [2] Gour L, and Waoo AA. Deep learning list in
name and
approach for enhancing fault tolerance for date style,
Accuracy, precision, recall, F1-score, and
reliable distributed system. 2021. please
ROC-AUC are commonly used metrics to [3] Duddu V, Rao DV, and Balas V. Towards check and
evaluate fault detection algorithms. Evalu- enhancing fault tolerance in neural net- confirm
ation metrics help measure the correctness, works. In: MobiQuitous 2020-17th EAI
this for
CHECKED AND clarity.
CONFIRMED
reliability, and effectiveness of fault detection International Conference on Mobile and
mechanisms. Fault tolerance mechanisms aim Ubiquitous Systems: Computing, Network-
to ensure system continuity, even in the pres- ing and Services; 2020:59–68.
ence of faults, failures, or unexpected events. [4] Khan W, and Haroon M. An unsupervised
Redundancy through replication, consen- deep learning ensemble model for anomaly
sus algorithms, distributed transactions, and
716 Innovations in Computational Intelligence: Trends and Technologies
detection in static attributed social networks. 2022. Singapore: Springer Nature Singa-
Int J Cogn Comput Eng. 2022;3:153–160. pore; 2022:119–128.
[5] Siddiqui ZA, and Haroon M. Research on sig- [12] Kumari P, and Kaur P. A survey of fault toler-
nificant factors affecting adoption of block- ance in cloud computing. J King Saud Univ-
chain technology for enterprise distributed Comput Inf Sci. 2021;33(10):1159–1176.
applications based on integrated MCDM [13] Siddiqui ZA, and Haroon M. Application of
FCEM-MULTIMOORA-FG method. Eng artificial intelligence and machine learning
Appl Artif Intell. 2023;118:105699. in blockchain technology. In: Artificial Intel-
[6] Khan M, and Haroon M. Artificial neu- ligence and Machine Learning for EDGE
ral network-based intrusion detection in Computing. Academic Press; 2022:169–185.
cloud computing using CSE-CIC-IDS2018 [14] Haroon M, Misra DK, Husain M, Tripathi
datasets. In: 2023 3rd Asian Conference on MM, and Khan A. Security issues in the
Innovation in Technology (ASIANCON); internet of things for the development of
2023:1–4. IEEE. smart cities. In: Advances in Cyberology and
[7] Khan W, and Haroon M. An efficient the Advent of the Next-Gen Information
framework for anomaly detection in attrib- Revolution. IGI Global; 2023:123–137.
uted social networks. Int J Inf Technol. [15] Duddu V, Rajesh Pillai N, Rao DV, and Balas
2022;14(6):3069–3076. VE. Fault tolerance of neural networks in
[8] Shahid MA, Islam N, Alam MM, Mazliham adversarial settings. J Intell Fuzzy Syst.
MS, and Musa S. Towards resilient method: 2020;38(5):5897–5907.
An exhaustive survey of fault tolerance [16] Khan W, Haroon M, Khan AN, Hasan MK,
methods in the cloud computing environ- Khan A, Mokhtar UA, and Islam S. DVAE-
ment. Comput Sci Rev. 2021;40:100398. GMM: Dual variational autoencoder with
[9] Kalaskar C, and Thangam S. Fault tolerance Gaussian mixture model for anomaly detec-
of cloud infrastructure with machine learn- tion on attributed networks. IEEE Access.
ing. Cybern Inf Technol. 2023;23(4):26–50. 2022;10:91160-91176.
[10] Siddiqui ZA, and Haroon M. Research on sig- [17] Wang K, and Louri A. Cure: A high-
nificant factors affecting adoption of block- performance, low-power, and reliable net-
chain technology for enterprise distributed work-on-chip design using reinforcement
applications based on integrated MCDM learning. IEEE Trans Parallel Distrib Syst.
FCEM-MULTIMOORA-FG method. Eng 2020;31(9):2125–2138.
Appl Artif Intell. 2023;118:105699. [18] Wang K, and Louri A. Cure: A high-
[11] Khan W, and Haroon M. A pilot study and performance, low-power, and reliable net-
survey on methods for anomaly detection in work-on-chip design using reinforcement
online social networks. In: Human-Centric learning. IEEE Trans Parallel Distrib Syst.
Smart Computing: Proceedings of ICHCSC 2020;31(9):2125–2138.
102 Predicting breast cancer risk: A machine
learning approach
Navin kumar Agrawal
Department of Computer Engineering and Applications, GLA University, Mathura, India
Abstract: One of the most fatal diseases is breast cancer, which originate in breast cells. It majorly effects the
females worldwide, although it may also in male, though very less frequently. When cells in the breast tissue
begin to grow uncontrolled, breast cancer develops. It may occur in the connective tissue of the lobes, among
other areas of the breast. Breast cancer analysis and prediction have emerged as crucial components in the fight
against one of the most prevalent and life-threatening diseases affecting women worldwide. In the realm of
predictive modelling, machine learning algorithms have gained prominence. These models leverage a multitude
of patient data, encompassing demographic information, genetic markers, and histopathological features, to
forecast risk, treatment response, and survival outcomes. Personalized medicine, guided by these predictions, is
transforming the landscape of breast cancer treatment, optimizing therapeutic strategies for individual patients.
Keywords: Breast Cancer, logistic regression
1. Introduction or scaling. However, breast cancer can be
asymptomatic in its early stages, which is
Breast cancer, a heterogeneous disease with
why regular screening is important [2].
diverse subtypes, demands comprehensive
Diagnosis: The diagnosis of cancer in the
approaches for its analysis and prediction.
breast is usually made using several of tech-
Early detection through mammography, clin-
niques, like clinical breast exams, mammog-
ical exams, and emerging technologies like
raphy, ultrasound, MRI, and biopsies. In a
AI-based image analysis remains pivotal for
improved prognosis [1]. biopsy, a sample of breast tissue is collected
There are many kinds in breast cancer, and analyzed under a microscope to deter-
with the two ordinary thing: mine whether cancer is present.
When abnormal cells are found in the lin- Staging: After being diagnosed, breast
ing of a breast duct but have not moved out- cancer is categorized according to how far it
side the duct, a condition referred to as DCIS, has spread. Treatment choices are guided in
a non-invasive cancer, occurs. part by staging. The stages are fourth to zero
Invasive Ductal Carcinoma (IDC): it is (non-invasive).
invasive breast cancer in nature, where malig- Treatment: Depending on the kind, stage,
nance cells break through the duct wall and and other individual factors, there are several
invade surrounding breast tissue. approaches to treating breast cancer. One
Symptoms: a few of the most prevalent common therapeutic option is surgery (such
symptoms of tumors in the breast is a breast as a mastectomy or lumpectomy), hormone
lump., some deviations in breast dimension treatment, targeted therapy, immunotherapy,
and shape and skin changes on the breast chemotherapy, and radiation therapy. Every
like it may be shown as redness, dimpling, patient has a unique treatment regimen [3].
[email protected],[email protected]
DOI: 10.1201/9781003606635-102
718 Innovations in Computational Intelligence: Trends and Technologies
Survival and Prognosis: Breast cancer sur- Early Detection: Mammography, ultra-
vival rates have grown as a result of early sound, MRI, or some other tomography tech-
identification and treatment advancements niques are used in early detection of breast
[4]. The size of the tumor, the grade at diag- Cancer. These images are analyzed by radi-
nosis, and other factors affect the prognosis, ologists, and computer aided detection sys-
and receptor status. Regular follow-up and tems can help in distinguishing abnormal
adherence to treatment are crucial for long- condition.
term survival. Biopsy and Pathology: If mistrustful
Prevention and Screening: Reducing the lesions found in the imaging, a diagnostic test
risk of breast cancer includes lifestyle mod- required to be taken to confirm the presence
ifications like managing healthy weight, of cancer. Pathologists examine tissue sam-
restraining alcohol consumption, and should ples to determine the type and grade of can-
be physically active. Regular breast cancer cer, which helps in treatment planning.
screening through mammography and clini- AI and Machine Learning: These algo-
cal breast exams is vital for early detection, rithms can analyse datasets to anticipate
especially for individuals with risk factors [5]. breast cancer probability, assist radiologists
Resources and Support: Receiving a breast in detecting abnormalities in images, and even
cancer diagnosis and undergoing treatment predict treatment outcomes based on patient
can be mentally and physically taxing. To data. These models can be trained on diverse
assist patients and their families in coping datasets to improve accuracy [6].
with the illness, a wealth of resources and Treatment Prediction: Predictive models
support groups are offered. can help oncologists select the most effective
Breast cancer is a complex and diverse treatment options for individual patients.
disease, and advancements in research and This can involve predicting how a patient
medical treatments continue to improve out- will respond to different therapies based on
comes [6]. Early detection and awareness are their genetic profile and tumor characteris-
key to successfully managing breast cancer, tics [7].
so it’s essential for everyone to be informed Survival and Recurrence Prediction:
about risk involve. Machine learning algorithm are to be utilized
Breast cancer analysis and prediction to guess a patient’s likelihood of survival and
involve using various techniques, tools, and the risk of cancer recurrence. These predic-
data to assess the risk of breast cancer, diag- tions help in making treatment decisions and
nose it, and predict outcomes or response designing follow-up plans [7].
to treatment. Here are some key aspects of Drug Discovery: AI and machine learning
breast cancer analysis and prediction: can also be used for drug discovery, identify-
Data Collection: The first step in breast ing new compounds or repurposing existing
cancer analysis is collecting relevant data. drugs for breast cancer treatment.
This data includes patient demographics, Clinical Trials: Data analysis and pre-
family history, genetic information, mam- diction play a decisive part in the design or
mography images, biopsy results, and other execution in clinical trials for breast cancer
clinical data. treatments. They help identify suitable candi-
Risk Assessment: Predictive modeling dates for trials and assess trial outcomes.
could be in use to evaluate a person’s risk of Patient Support: Predictive models and
underdeveloped breast cancer. This typically data analysis can be used to provide patients
involves analyzing risk factors such as age, with personalized information about their
family history, genetics, and lifestyle choices. condition, treatment options, and potential
Models like the Gail model or the Tyrer-Cuz- side effects, enabling shared decision-mak-
ick model are used for risk assessment. ing [8].
Predicting breast cancer risk: A machine learning approach 719
Ethical-Considerations: It is necessary to “Computer-Aided Screening and Detec-
consider ethical and privacy concerns when tion of Cancer of the Breast with Mammog-
dealing with patient data and predictive mod- raphy: Recent Advances” by Elter et.al - This
els [9]. Protecting patient privacy and ensur- review discusses current improvements in
ing the responsible use of AI in healthcare are computer-aided detection and diagnosis of
critical aspects of breast cancer analysis and breast cancer using mammography.
prediction [10].
2.3. Genomic and genetic analysis
2. Literature Survey “Breast Cancer Genomics and Immunity:
A literature survey on breast cancer analysis From Mechanism to Therapy” by Liu et al.
and prediction involves reviewing research (2019) - This paper explores the genomic and
papers, articles, and studies related to the use immune aspects of breast cancer and their
of various techniques, tools, and data-driven implications for treatment.
approaches for assessing, diagnosing, and
predicting outcomes in breast cancer. Below 2.4. Treatment prediction and
is a selection of key research papers and top- personalized medicine
ics in this field:
“Personalized Medicine in Breast Cancer:
Tailoring Treatment to the Patient” by Grad-
2.1. Early detection and screening ishar et al. (2019) - This article discusses the
“Efficacy of Mammograms Monitoring in concept of personalized medicine in breast
Decreasing Cancer of the Breast Mortality” cancer treatment.
by Tabar et al. (2019) - This study analyzes “Prediction of Pathological Complete
the impact of mammograms to lower mor- Response to Neoadjuvant Chemotherapy
tality from breast cancer.Predictive Modeling for Locally Advanced Breast Cancer by Bio-
and Risk Assessment: markers” by Cortazar et al. (2014) - This
“Forecasting Breast Cancer Hazard Con- study investigates biomarkers to predict the
sidering Individual Health Information and response to neoadjuvant chemotherapy in
Machine Learning” by Li et al. (2019) - This breast cancer patients.
research addresses the application of machine
learning algorithms to predict breast cancer 2.5. Survival and recurrence
risk based on personal health data. prediction
“”An Overview of Cancer Diagnosis Using
Machine Learning Techniques” by Jalalian “Predicting Breast Cancer Recurrence: A Sys-
and Mashohor (2017) - This article in review tematic Review of Prognostic Biomarkers” by
presents a summary of machine learning De Kruijf et al. (2014) - This review paper
techniques for breast cancer prediction and explores prognostic biomarkers for predict-
diagnosis. ing breast cancer recurrence.
2.2 Medical imaging and diagnosis 2.6. Ethical and privacy
considerations
“Deep Learning Methods for Cancerous
Breast Detection from Mammograms are: “Ethical and Legal Challenges in AI-Driven
A Multidisciplinary Study” by Burt et al. Predictive Healthcare” by Kostkova et al.
This study examines multiple deep learning (2020) - This paper discusses ethical and legal
algorithms for breast cancer diagnosis using challenges associated with predictive health-
mammography pictures. care models, including those for breast cancer.
720 Innovations in Computational Intelligence: Trends and Technologies
2.7. AI and Machine Learning In this, we used logistic regression. It is espe-
Applications cially useful when you have a binary classifi-
cation problem (two classes, often referred to
“Artificial Intelligence in Breast Imaging” by as “positive” and “negative”).
Ha et al. (2019) - This review provides an As shown in X-axis the size of training
overview of AI applications in breast imag- data set is increasing, the performance on
ing, including detection and diagnosis. Y-axis is either increasing or in saturation.
3. Results
To find the best result in this paper logistic
Regression method we used to predict the
breast cancer concerns, in this paper outcome
were calculated and examined. The accuracy
produced my model is 89.07.
ROC curved shape is pictorial represent
the demonstration, in this logistic regression
Figure 102.1
model is used, across different thresholds for
making predictions. It depicts the trade-off
between the True Positive Rate (Sensitivity)
and the False Positive Rate (1 - Specificity) as
the discrimination threshold is adjusted.
Confusion matrix provides a way to visu-
alize how well your model is doing in terms
of making correct and incorrect Figure 102.2
predictions.
Figure 102.2. Performance evaluation using
The four components of a confusion matrix
confusion matrix.
are TP (True-Positive), TN (True-Negative),
FP (False-Positive), and FN (False-Negative). Source: AUTHOR'S SELF GENERATED
AQ1:
Please
provide
TEXT
in-text BOX ANNOTATION ADDED
citation
for figures
1-3.
AQ2:
Please
provide
source Figure 102.1. Performance evaluation using Figure 102.3. Learning curve using logistic
for all the ROC curve. model.
figures
NOTE ADDED
Source: Source:
AUTHOR'S SELF GENERATED
AUTHOR'S SELF GENERATED
Predicting breast cancer risk: A machine learning approach 721
So this curve evaluate model in correct direc- References
tion. Learning curves are useful for assessing
[1] Das, R. C., Das, M. C., Hoassain, M. A.,
whether a model is overfitting or under fitting
Figureefficacy
102.3 Rahman, M. A., Hossen, M.H., and Hasan,
and for knowing how the model’s R. (2023, March). Heart Disease Detection
is impacted by the volume of data used for Using ML. In 2023 IEEE Annual Computing
training. and Communication Workshop and Confer-
ence (CCWC) (pp. 0983–0987). IEEE.
[2] Hossain, M. K. Arnab, A. A., Das, R. C.,
4. Summary Hossain, K. M., Rubel, M. H. K., Rahman,
In summary, breast cancer analysis and pre- M. F., and Pandey, R. (2022). Combined
diction involve a multi-disciplinary approach DFT,SCAPS-1D, and and wxAMPS frame-
that combines clinical expertise, medical works for design optimization of efficient
imaging, genetic information, and advanced Cs 2 BiAgI 6-based perovskite solar cells
data analytics to improve early detection, with different charge transport layers. RSC
advances, 12(54), 35002–35025.
treatment selection, and patient outcomes
[3] Panddit, P P. P., Liu, C., Corti, G., and Hu,
in breast cancer care. Advances in AI and
Y. (2022, June). FusedDeposition Modeling
machine learning continue to play a signifi- of Carbon Fiber Reinforced High-Densi-
cant role in enhancing the accuracy and effi- tyPolyethylene: Effects on Microstructure
ciency of these processes.in this paper we and Mechanical Properties. InInternational
used to perform the analysis by using logistic Manufacturing Science and Engineering
regression model, which has produced high Conference (Vol. 85802, p. V001T01A033).
performance. American Society of Mechanical Engineers.
[4] M. Haasan, M. S. Tahosin, A. Farjana, M. A.
Sheakh and M. M. Hasan,”A Harmful Dis-
5. Challenges order: Predictive and Comparative Analysis
Medical imaging, particularly mammog- for fetal Anemia Disease by Using Different
raphy, continues to evolve with advanced Machine Learning Approaches,” 2023 11th
International Symposium on Digital Foren-
techniques.
sics and Security (ISDFS), Chattanooga,
Ethical considerations and patient privacy
TN, USA, 2023, pp. 1–6, doi:10.1109/
in handling sensitive medical data remain par- ISDFS58141.2023.10131838.
amount, warranting comprehensive frame- [5] Agrawal, N. K., and Sharma, A. K. (2021,
works to ensure responsible use of predictive December). Detection of Diseases in Plants
models and data. using Convolutional Neural Networks. In
As the field progresses, interdisciplinary 2021 10th International Conference on Sys-
collaboration between clinicians, research- tem Modeling and Advancement in Research
ers, and technologists remains essential for Trends (SMART) (pp. 215–217). IEEE.
further advancements. Breast cancer analysis [6] Shankhdhar, A., Agrawal, N. K., and Sriv-
and prediction represent a dynamic and evolv- astava, A. (2021, December). COVID-19
ing frontier, holding immense potential to Detection System using Chest X-rays or CT
reduce mortality, enhance treatment efficacy, Scans. In 2021 10th International Confer-
ence on System Modeling and Advancement
and ultimately improve the quality of life for
in Research Trends (SMART) (pp. 428–432).
affected individuals. Continued research and
IEEE.
innovation in this domain promise to play [7] P. Jain, “Detection of breast cancer using
a pivotal role in the ongoing battle against machine learning algorithms,”International
breast cancer. Journal for Research in Applied Science and
Engineering Technology, vol. 10, no. 6, pp.
3484–3487, 2022.
722 Innovations in Computational Intelligence: Trends and Technologies
[8] World Health Organization. [Online]. kaggle.com/datasets/uciml/breast-cancer-
Available:https://www.iarc.who.int/news- wisconsin-data.[Accessed: 28-Feb-2023].
events/globocan-2008-cancerincidence- [10] M. Mamun, A. Farjana, M. Al Mamun, and
and-mortality-worldwide/. [Accessed: M. S. Ahammed, “Lungcancer prediction
28-Feb-2023]. model using ensemble learning techniques
[9] U. C. I. M. Learning, “Breast cancer wiscon- and asystematic review analysis,” 2022 IEEE
sin (diagnostic) data set,”Kaggle, 25-Sep- World AI IoT Congress(AIIoT), Seattle, WA,
2016. [Online]. Available:https://www. USA, 2022, pp. 187–193, doi:10.1109/
AIIoT54504.2022.9817326
103 A multi-objective optimization
algorithm for fog computing with trust
and energy awareness
Ruchi Agarwala, Saurabh Singhalb, and Ashish Sharmac
Department of Computer Engineering and Applications, GLA University, Mathura, Mathura, Uttar
Pradesh, India
Abstract: Resource provisioning in fog computing is a crucial research area that focuses on optimizing resource
allocation and ensuring secure data transmission in Fog Computing Environments (FCE). The challenge lies
in balancing computational load, security, and energy efficiency, requiring an effective optimization algorithm
to address these multiple objectives simultaneously. A trust-based energy-aware secure load balancing and
resource provisioning framework in fog computing has been proposed, which incorporates methods such as
Fog Node Trust Computation, Trust Credibility Assessment, Allocation (DEER), Time-Constraint Energy
Minimum Problem (TCEMP), and the utilization of the Multi-objective MSA algorithm. The future potential
of this framework, utilizing a Multi-objective MSA algorithm, is in fog computing, potentially leading to wide-
spread adoption as a secure alternative to traditional cloud computing.
Keywords: Energy-Aware Load Balancing, Resource Provisioning, Fog Computing, Multi-Objective Optimiza-
tion Algorithm, Dynamic Energy Efficient Resource Allocation, Time-Constraint Energy Minimum Problem
1. Introduction importance of dynamic resource allocation
grows more essential [5]. Conventional load-
Efficient and secure resource allocation in balancing methods may not be fully equipped
fog computing hinges on trust-based energy- to handle the distinct features of Fog Com-
aware load balancing and resource pro- puting, such as heterogeneity and mobility
visioning [1]. Implementing a dual-layer [6]. The incorporation of trust-based mod-
fluffy strategy utilizing a Zero Trust Secu- els adds a layer of dependability, encourag-
rity System in Multi-access Edge Computing ing secure cooperation among Haze hubs [7].
(MEC) is the recommended methodology [2]. The energy-mindful aspect guarantees sup-
The architecture effectively oversees trust- portability by streamlining asset usage, a sig-
mindful validation and errand offloading in nificant element given the imperatives of edge
Mobile Edge Computing (MEC) environ- gadgets [8]. This exploration decides to con-
ments, ensuring robust security protocols are tribute bits of knowledge and strategies to the
implemented [3]. This research focuses on incipient field of Haze Figuring, establishing
integrating trust-based mechanisms, energy- the groundwork for a stronger, energy-pro-
aware strategies, and secure load-balancing ficient, and secure processing worldview [9].
techniques to develop a resilient framework The energy-mindful aspect guarantees man-
that improves efficiency, security, and energy ageability by enhancing asset use, an essential
usage in Fog Computing infrastructures [4]. variable given the limitations of edge gadg-
As Fog Computing systems advance, the ets [10]. Trust foundation among Mist hubs
[email protected]; [email protected]; [email protected]
a
DOI: 10.1201/9781003606635-103
724 Innovations in Computational Intelligence: Trends and Technologies
represents a critical obstacle because of the approaches can be overcome by a variant
dynamic and heterogeneous nature of Haze of sleep scheduling. Selvakumar et al. [20]
conditions [11]. Guaranteeing secure burden propose a trustworthy multiobjective reptile
adjusting while at the same time considering search algorithm (M-TCRSA) for secure clus-
the different capacities of edge gadgets and ter-based routing in WSN. The M-TCRSA’s
tending to potential malevolent exercises PDR values for 400 nodes are 99.27%, which
requires vigorous trust models [12]. Due to is higher than the TAGA, TCELR, SCRIMFO,
its proximity to edge strategies and ability to and CSA.
process real-time data, Fog Computing holds Aqeel et al. [21] An innovative IoT envi-
great potential for a wide range of applica- ronment leveraging cloud technology, an
tions [13]. Implementing trust-based mecha- AI-driven load balancing technique prioritiz-
nisms to improve the reliability of interactions ing alternative energy sources, utilizes Big
among Fog nodes, incorporating energy-effi- Data analysis and the CHROA algorithm
cient strategies for sustainable resource man- for improved efficiency. The results indicate
agement, and utilizing secure load balancing that it may effectively tackle significant chal-
techniques to optimize performance [14]. lenges and aid in the creation of effective and
Creative solutions are required to address long-lasting IoT/IoE solutions. Saif et al. [22]
the diversity, flexibility, and limited resources a novel Multi-Goals Dark Wolf Enhancer
found in Fog environments [15]. This study (MGWO) approach has been proposed for
aims to establish a strong foundation for reducing Quality of Service (QoS) targets’
trust-based, energy-aware, and secure load latency and Energy Consumption (EC). This
balancing, in addition to resource provision- strategy is stored in the cloud repository,
ing in Fog Computing. The literature review, which is vital for project delivery. The MGWO
proposed technique, results, and conclusions algorithm helps in minimizing latency and
of the study are presented in Sections 2, 3, 4, energy usage. Scheduling technique proposed
and 5, respectively. by Mangalampalli et al. [23] uses Profound
Support Learning. The outcomes reveal that
the proposed MOPDSWRL surpasses other
2. Literature Review algorithms in makespan and EC. Zhou et al.
The literature Survey delves into the combi- [24] track multiple objectives in intelligently
nation of trust mechanisms, energy-conscious networked automobiles, based on hybrid
strategies, and multi-objective optimization traffic, which proves to be more effective in
in Fog Computing to gain a thorough com- improving target state estimates compared to
prehension and uncover essential findings other standard research approaches, as dem-
for the suggested model. Al Moteri et al. onstrated by the results of the VB-RMTCT
[16] suggested increasing QoS metrics were tracking algorithm. Zhao et al. [25] proposed
adjusted, and resources were assigned to the QCGA, comprised of quantum such as non-
MEC network. This leads to bad latency, low gate, rotation gate, has been proposed to
productivity and energy consumption (EC), improve the algorithm’s global search capa-
and a high delay rate. Zhou et al. [17] pro- bility. Simulation results indicate that the
posed major goal of load balancing for com- average delay of QCGA outperforms that of
puting resources to increase performance. Ma SO, GA, PSO, and sequential offloading.
et al. [18] suggested the EC with combining it
with QoS-aware cloud service as per experi-
mental data, this approach surpasses existing 3. Proposed Research Methodology
ways, saving energy in the range of 8% to This exploration aims to transform fog com-
35%. Nithyanandh et al. [19] demonstrated puting by introducing an innovative frame-
how limitation of existing contemporize work that integrates energy efficiency, secure
725
load balancing, and trust-based resource pro- of the entities involved in the trust-based
visioning. By leveraging advanced multi-objec- energy-aware in fog computing. The MOO
tive optimization algorithms, the framework algorithm is utilized to identify the most
maintains a balanced system that improves reliable and optimal entities for resource
performance, security, and energy usage. allocation.
The block diagram in Figure 103.1 (1)
depicts the DEER approach, which focuses
on energy-efficient load balancing through The trust credibility evaluation model
dynamic resource allocation. By intelligently assesses trust over consecutive periods [t0,
distributing the workload among fog nodes, t] and then recalculates the trust at the most
this approach minimizes energy consump- recent time instance t.
tion. Additionally, the integration of MOO
algorithms such as Multi-Objective MSA 5. Energy-Aware Load Balancing
and HHO algorithm aids in optimizing mul-
tiple objectives simultaneously for resolving To achieve this, a multi-objective optimiza-
complex issues related to fog node trust and tion algorithm is utilized to distribute work-
energy efficiency. The framework also consid- loads and resources among nodes, taking into
ers TCEMP to ensure that energy needs are account energy usage and security needs. This
fulfilled within specified time limits. strategy enhances the effectiveness of fog
computing systems in terms of performance
and energy efficiency, ensuring reliability and
4. Trust-Based Framework security in operations.
Development
Developing a framework based on trust 6. Experimentation and Result
involves creating a system that relies on trust Discussion
as a key factor in achieving energy-efficient,
secure load balancing, and resource provi- A trust-based energy-aware secure balancing of
sioning in fog computing. load and resource provisioning mechanism in
fog computing using a MOO algorithm imple-
mented to optimize the allocation of resources in
4.1. Trust credibility assessment
FCE while considering the trustworthiness and
The trust credibility assessment focuses on EC of fog nodes. The experimentation outcome
evaluating the credibility and dependability underscores the potential performance and fog
computing efficiency systems techniques with
MATLAB’s analytical capabilities.
6.1. Development of a trust-based
framework
Fog Node Trust Computation refers to the
process of determining the trustworthiness
of fog nodes based on various factors such
as their previous performance and reliabil-
AQ1: ity. Trust Credibility Assessment refers to the
Please
provide assessment of the credibility and dependabil-
source Figure 103.1. Flow diagram of trust-based ity of the trust information provided by fog
for all the system.
figures
nodes, ensuring the accuracy and dependabil-
Source:
AUTHOR'S SELF GENERATED
ity of the trust scores.
726 Innovations in Computational Intelligence: Trends and Technologies
Figure 103.2. Correlation between Fog Node
Trust and Credibility.
Source: AUTHOR'S SELF GENERATED Figure 103.3. Correlation Between F-Rate and
Number of Incoming Tasks.
The graph depicts the correlation between Source:
the trust levels of fog nodes and their cred- AUTHOR'S SELF GENERATED
ibility in collusion attacks (Figure 103.2). The
X-axis represents different collusion attack
scenarios (A1 to A6), while the Y-axis repre-
sents trust levels ranging from 0.2 to 0.8. Sce-
nario A6 stands out as it exhibits a high fog
node trust of 0.8, which aligns with its cred-
ibility. The graph demonstrates the strong
positive relationship between trust and cred-
ibility in fog nodes during collusion attacks.
6.2. Energy-aware load balancing
optimization Figure 103.4. Energy Consumption Comparison
The algorithm aims to reduce EC while ensur- Chart.
ing that tasks are completed within their Source: AUTHOR'S SELF GENERATED
specified time limits. DEER approach for bal-
ancing load in a fog atmosphere dynamically 6.3 Synergizing Multi-Objective
allocates resources based on energy efficiency.
Figure 103.3 shows that as the volume of Optimization Algorithms
upcoming projects increases, the F_rate ini- The MOO algorithm helps in fine-tuning the
tially starts at a certain value (not mentioned) system performance, security, and energy effi-
for 0 incoming tasks and gradually increases. ciency by finding the best possible trade-offs
Similarly, at 10, 20, and 30 incoming tasks, the between these objectives. The Multi-objective
F_rate still increases but at a diminishing rate. MSA algorithm applies a technique called the
Figure 103.4 displays the graph shows Simulated Annealing approach with multiple
the EC of three different methods for cach- objectives to efficiently explore the solution
ing: advanced cache, DEER Method, and Cor area and find the optimal solutions.
Based Resource Adaptive LBA. When the Figure 103.5 shows the graph depicting
maximum task load is reached, the advanced the median line θ, which ranges from 0 to
cache consumes the highest amount of energy 6. The fog represents user requests, showing
at 800 joules. there is a notable disparity in the data of more
727
to more reliable and efficient fog computing
systems.
References
[1] ALI, B.A., 2023. Efficient trust-aware authen-
tication and task offloading in Multi-access
Edge Computing using a dual fuzzy method
based Zero Trust Security framework (Doc-
toral dissertation, RMIT University).
[2] Wang, C., Yuan, Z., Zhou, P., Xu, Z., Li, R.
and Wu, D.O., 2023. The Security and Pri-
vacy of Mobile Edge Computing: An Artifi-
cial Intelligence Perspective. IEEE Internet of
Figure 103.5. Resource Optimization Graph Things Journal.
Theta. [3] Zhou, S., Jadoon, W. and Khan, I.A., 2023.
Computing Offloading Strategy in Mobile
Source: AUTHOR'S SELF GENERATED
Edge Computing Environment: A Compari-
son between Adopted Frameworks, Chal-
than 3 for many user requests. In contrast, the lenges, and Future Directions. Electronics,
cloud has less degrading values compared to 12(11), p.2452.
the fog. [4] Aqeel, I., Khormi, I.M., Khan, S.B., Shuaib,
M., Almusharraf, A., Alam, S. and Alkhaldi,
N.A., 2023. Load Balancing Using Artificial
7. Conclusion Intelligence for Cloud-Enabled Internet of
The study recognizes that fog computing, Everything in Healthcare Domain. Sensors,
with its distributed nature and reliance on 23(11), p.5349.
[5] Hazra, A., Rana, P., Adhikari, M. and
resource constraints, requires an efficient
Amgoth, T., 2023. Fog computing for next-
and secure approach to handle the dynamic generation internet of things: fundamental,
workload and resource allocation. The devel- state-of-the-art and research challenges.
oped algorithm, which incorporates multi- Computer Science Review, 48, p.100549.
ple objectives such as energy efficiency, load [6] Das, R. and Inuwa, M.M., 2023. A review
balancing, and security, enhances the overall on fog computing: issues, characteristics,
performance of fog computing systems. By challenges, and potential applications. Tele-
utilizing the MATLAB instrument, the evalu- matics and Informatics Reports, p.100049.
ation and efficiency of the fog computing pro- [7] Songhorabadi, M., Rahimi, M., Moghad-
cess are being assessed. Finding this approach amFarid, A. and Kashani, M.H., 2023. Fog
achieves a balance between system efficiency, computing approaches in IoT-enabled smart
security, also, energy enhancement in dynamic cities. Journal of Network and Computer
fog environments. This results in improved Applications, 211, p.103557.
[8] Burhan, M., Alam, H., Arsalan, A., Rehman,
energy efficiency, reducing the overall opera-
R.A., Anwar, M., Faheem, M. and Ashraf,
tional costs and environmental impact. The M.W., 2023. A comprehensive survey on the
experimental evaluation confirms the efficacy cooperation of fog computing paradigm-
of the proposed algorithm showcasing its based iot applications: layered architecture,
ability to achieve a trade-off between various real-time security issues, and solutions. IEEE
objectives. It outperforms existing described Access.
algorithms in different words load balancing, [9] Srirama, S.N., 2024. A decade of research
resource provisioning, and security, leading in fog computing: relevance, challenges, and
728 Innovations in Computational Intelligence: Trends and Technologies
future directions. Software: Practice and Multi-Cloud Service Composition. Applied
Experience, 54(1), pp.3-23. Sciences, 13(11), p.6826.
[10] Walia, G.K., Kumar, M. and Gill, S.S., 2023. [19] Nithyanandh, S., Omprakash, S., Megala, D.
AI-empowered fog/edge resource manage- and Karthikeyan, M.P., 2023. Energy Aware
ment for IoT applications: A comprehensive Adaptive Sleep Scheduling and Secured Data
review, research challenges and future per- Transmission Protocol to enhance QoS in
spectives. IEEE Communications Surveys IoT Networks using Improvised Firefly
and Tutorials. Bio-Inspired Algorithm (EAP-IFBA). Indian
[11] Iftikhar, S., Gill, S.S., Song, C., Xu, M., Journal Of Science And Technology, 16(34),
Aslanpour, M.S., Toosi, A.N., Du, J., Wu, H., pp.2753-2766.
Ghosh, S., Chowdhury, D. and Golec, M., [20] Selvakumar, T. and Jeyakarthic, M., 2024.
2023. AI-based fog and edge computing: Modified Dolphin Search Based Energy
A systematic review, taxonomy and future Efficient and Secured Routing Protocol for
directions. Internet of Things, 21, p.100674. Wireless Sensor Network. International
[12] Ostrowski, K., Małecki, K., Dziurzański, Journal of Intelligent Systems and Applica-
P. and Singh, A.K., 2023. Mobility-aware tions in Engineering, 12(8s), pp.334-346.
fog computing in dynamic networks with [21] Aqeel, I., Khormi, I.M., Khan, S.B., Shuaib,
mobile nodes: A survey. Journal of Network M., Almusharraf, A., Alam, S. and Alkhaldi,
and Computer Applications, p.103724. N.A., 2023. Load Balancing Using Artificial
[13] Tusa, F. and Clayman, S., 2023. End-to-end Intelligence for Cloud-Enabled Internet of
slices to orchestrate resources and services in Everything in Healthcare Domain. Sensors,
the cloud-to-edge continuum. Future Gener- 23(11), p.5349.
ation Computer Systems, 141, pp.473-488. [22] Saif, F.A., Latip, R., Hanapi, Z.M. and Shafi-
[14] Tran-Dang, H., Kwon, K.H. and Kim, D.S., nah, K., 2023. Multi-objective grey wolf
2023. Bandit Learning-Based Distributed optimizer algorithm for task scheduling
Computation in Fog Computing Networks: in cloud-fog computing. IEEE Access, 11,
A Survey. IEEE Access. pp.20635-20646.
[15] Sulimani, H., Sajjad, A.M., Alghamdi, W.Y., [23] Mangalampalli, S., Hashmi, S.S., Gupta,
Kaiwartya, O., Jan, T., Simoff, S. and Prasad, A., Karri, G.R., Rajkumar, K.V., Chakra-
M., 2023. Reinforcement optimization for barti, T., Chakrabarti, P. and Margala, M.,
decentralized service placement policy in 2024. Multi-Objective Prioritized workflow
IoT-centric fog environment. Transactions scheduling using Deep reinforcement-based
on Emerging Telecommunications Technolo- Learning in Cloud Computing. IEEE Access.
gies, 34(11), p.e4650. [24] Zhou, N. and Jiang, J., 2024, January. Multi-
[16] Al Moteri, M., Khan, S.B. and Alojail, M., objective Tracking Algorithm for Intelligent
2023. Machine Learning-Driven Ubiqui- Networked Vehicles in Hybrid Traffic. In
tous Mobile Edge Computing as a Solution Proceedings of the First International Con-
to Network Challenges in Next-Generation ference on Science, Engineering and Technol-
IoT. Systems, 11(6), p.308. ogy Practices for Sustainable Development,
[17] Zhou, J., Lilhore, U.K., Hai, T., Simaiya, ICSETPSD 2023, 17th-18th November
S., Jawawi, D.N.A., Alsekait, D., Ahuja, S., 2023, Coimbatore, Tamilnadu, India.
Biamba, C. and Hamdi, M., 2023. Compar- [25] Zhao, J., Liu, X. and Tian, M., 2024. An effi-
ative analysis of metaheuristic load balanc- cient task offloading method for drip irriga-
ing algorithms for efficient load balancing tion and fertilization at edge nodes based on
in cloud computing. Journal of Cloud Com- a quantum chaotic genetic algorithm. AIP
puting, 12(1), pp.1-21. Advances, 14(1).
[18] Ma, W. and Xu, H., 2023. Skyline-Enhanced
Deep Reinforcement Learning Approach
for Energy-Efficient and QoS-Guaranteed
104 Pediatric brain tumor detection in MRI:
A machine learning perspective
Sandeep Kumar Mishraa and Sheeba Praveenb
Department of Computer Science and Engineering, Faculty of Engineering and IT, Integral
University, Lucknow, India
Abstract: This study introduces early detection of brain tumors in children and focused on magnetic resonance
imaging (MRI) characteristic of children brain tumors and to create an automated segmentation system that
might use machine learning techniques to separate and categorize tumors and compare the results with radi-
ologist diagnoses. Accurate and timely identification of brain tumors, especially in children, is paramount for
effective planning of treatment and improving patient welfare. The intricate structure of the developing brain,
coupled with the heterogeneity of tumor types, presents significant challenges for conventional diagnostic meth-
ods. This study introduces automated approach for the detection and generalization of brain tumors in children
using MRI and advanced machine learning algorithm. Artificial intelligence-automated tumor description for
pediatric gliomas would enable real-time volumetric assessment to support clinical decision-making, treatment
response evaluation, and diagnosis. Auto-segmentation algorithms for pediatric brain tumor are uncommon
and have not yet demonstrated clinical translation due to a lack of available data.
Keywords: Medical image, CNN, BPN, neural network, deep learning, pediatric low-grade gliomas (pLGGs)
1. Introduction involve considerable hazards. Precise tumor
segmentation is necessary for accurate brain
Brain tumors are a serious health risk, espe- tumor localization and characterization on
cially for child patients due to the growth MRI scans, which[2] is a prerequisite for optimal
rate of brain cells. For therapy to be effec- risk classification, response evaluation, and
tive and for patient outcomes to be improved, monitoring for pediatric low-grade gliomas
early and precise detection is essential. Even (pLGGs)2. The clinical efficiency of pLGG is
while improvements in medical imaging have limited because to its manual segmentation,
AQ1: improved diagnostic capabilities, manual which poses unique obstacles and requires a
Reference medical image interpretation is still labor- large amount of knowledge, time, and money
citations
are not in intensive and prone to human error. This as compared to adult gliomas.
sequence, research aims to overcome these shortcomings The development of auto-segmentation
please
check and by creating a strong machine learning system methods for pediatric brain tumors has gar-
review for for the identification of pediatric brain can- nered substantial interest due to the inherent
clarity.
cers using magnetic resonance imaging (MRI) constraints of manual segmentation and the
AQ2: As data5. Gliomas tumor are the special type of usefulness of precise segmentation. A variety
per style, tumor in brain of children. 30% of gliomas of approaches, including ML-based, deep,
please
provide
tumor are present in all children brain tumor1. DL-based, and hybrid approaches, have been
the refer- Handling of these type of brain tumor in chil- developed for brain tumor segmentation as
ence cita- dren are varied in form, natural history, and a result of advancements in medical imaging
tions in
numbered aggressiveness, and because their medicines techniques and computational tools2. Deep
with
brackets.
[1]
[email protected]; [email protected]
a
DOI: 10.1201/9781003606635-104
Replace 1 by [6]
730 Innovations in Computational Intelligence: Trends and Technologies
learning has been a potent technique in medi- highlighting the effectiveness of these tech-
cal imaging recently, providing answers to niques in improving diagnostic accuracy.
many clinical problems. While there are still However, customized approaches are needed
certain obstacles to overcome, auto-segmenta- to address the unique problems related to
tion based on deep learning offers a promising pediatric brain tumors, including the com-
method for the precise and effective segmenta- plexity of the developing brain and the wide
tion of brain tumors, including pLGGs1. variety of tumor features. Children’s brain
One main problem is the relative rarity tumors have certain characteristics that
of pLGGs due to the unavailability of data require specialized approaches in their detec-
sets for public use for training the models. tion and analysis. [8]
[7] Maximum brain tumor detection algorithms Convolutional neural networks (CNNs), a
have been developed for young people with class of deep learning algorithms, are effective
gliomas, which are common and benefit from particularly effective in medical image analy-
large volumes of publicly available data sets sis because of automatically extract relevant
for training the model7. To compare segmen- features from raw image data3. Studies such
tation models’ performance to that of profes- as Qader et al. (2022) have demonstrated
sionals, ascertain their actual performance the effectiveness of CNNs in brain tumor
level, and assess their potential for clini- detection, showcasing their potential to sig-
cal translation, human clinical evaluation is nificantly improve diagnostic outcomes. By
crucial. leveraging the strengths of both traditional AQ3:
Please
There are many methods have been pro- ML and deep learning algorithms, research- define
posed to increase the machine learning per- ers aim to develop robust systems capable of ML, if
needed.
formance with the limited data set. Recently, addressing the inherent complexities of pedi-
there are many advancements have been atric brain tumor analysis.
done in the machine-learning field, that are Using pediatric brain MR scans, a retro-
considered to increase the obligation of MRI spective analytical study was conducted at
images for analysis of Machine Learning KASCH Hospital in Riyadh, Kingdom of
algorithms, although they can be challenging Saudi Arabia (KSA) to evaluate the effective-
to implement. ness of many machine learning techniques in
This review aims to explore the current the early detection of brain cancers. The Insti-
state of research on pediatric brain tumor tutional Re-view Board of Princess Nourah
detection using MRI images and machine bint Abdulrahman University (PNU) granted
learning algorithms, with a focus on deep approval, and between November 2022 and
learning-based auto-segmentation for tumor March 2023, a manual dataset assessment
detection. We will discuss the advancements was conducted to determine the appropri-
and current problem that associated with ate eligibility criteria for this study (IRB Log
the detection of brain tumor in children, as Number: 22–1004). Brain tumor MRI pic-
well as make the effective strategies to con- tures from patients under 14 selected from
nect and fill the gap for pediatric brain tumor PACS were one of the eligibility requirements.
segmentation algorithms, ultimately aiming Any patient in the same age group who is
to enhance the clinical applicability of these older than 14 and has a brain tumor or a
technologies. tumor in a different location.
Deep learning has been applied to dis-
cover and detect various brain disorders in
2. Methodology fetuses and toddlers. This section outlines
Azhari et al. investigated the use of ML for current deep learning-based techniques for
tumor classification in medical imagery,
Pediatric brain tumor detection in MRI: A machine learning perspective 731
finding and categorizing pediatric brain and Prince et al. (2020) utilized a combina-
tumor research. tion of CNN and genetic algorithm (GA) as a
[9-15] meta heuristic optimizer.
The types and locations of tumors inves-
3. Overview of Current Work tigated vary across studies. Arle et al. studied
Initial brain tumors are a variety of cancers posterior fossa tumors (astrocytomas, PNETs,
that can be benign or malignant and arise and ependymomas). Bidiwala and Pittman
from the brain parenchyma and surround- focused on posterior fossa tumors (astrocy-
ing tissues. Only leukemia causes more child- tomas, ependymomas, and medulloblasto-
hood death and morbidity globally than mas). Quon et al. researched posterior fossa
brain tumors, which are the most prevalent tumors (diffuse midline glioma, medulloblas-
solid tumors in the pediatric age range4. The toma, and ependymoma). Ye et al. examined
majority of research indicates that astrocyto- various histologic elements of pediatric brain
mas, medulloblastomas, and ependymomas tumors, while Prince et al. analyzed adaman-
are the three most prevalent forms of juve- tinomatous craniopharyngioma (DOLL).
nile brain tumors. The kind and grade of The imaging techniques used in these
the tumor affect the overall survival rate in studies are diverse. Arle et al. utilized mag-
children with brain tumors. The highest sur- netic resonance spectroscopy (MRS) + mag-
vival rates among juvenile brain tumors are netic resonance imaging (MRI) + Meta Data.
seen in low-grade gliomas, such as pilocytic Bidiwala and Pittman combined computed
astrocytoma, which have a 10-year progres- tomography (CT) + MRI (T1W1, T2W1)
sion-free survival rate of over 95% follow- + Meta Data. Quon et al. focused on T2
ing gross complete resection. Because brain weighted MRI, Ye et al. used DBSI, and Prince
tumors are aggressive and have a large influ- et al. combined CT + MRI + combined CT +
ence on young lives, they pose a significant MRI.
challenge to public health, especially in chil- The data sets used in these studies range in
dren. Timely and precise diagnosis is essen- scope and size. Arle et al. and Bidiwala and
tial for efficient treatment and higher survival Pittman both used self-acquired data sets con-
rates (Işın et al., 2016). Since MRI provides sisting of 33 children. Quon et al. conducted
comprehensive anatomical information with- a multi-institutional study involving 617
out subjecting patients to ionizing radiation, children. Ye et al. used nine pediatric brain
it has emerged as the gold standard for the tumor post-mortem specimens, and Prince et
identification of brain tumors. However, al. conducted a multi-institutional study with
manual MRI picture interpretation is time- 617 children.
consuming, prone to human error, and can The classification accuracy rates reported
vary among radiologists. in these studies vary. Arle et al. achieved
Different studies have employed vari- 58%-95%, Bidiwala and Pittman achieved
ous methodologies to enhance the detection 72.7%-85.7%, Quon et al. reported a detec-
and classification of pediatric brain tumors. tion accuracy with an AUROC of 0.99, Ye et
Arle et al. (1997) used a 4 back propaga- al. achieved 83.3%, and Prince et al. reported
tion neural network (BPNN), while Bidiwala classification accuracies of 85.3%, 83.3%,
and Pittman (2004) utilized a general Neural and 87.5% for different modalities. This
Network. Quon et al. (2020) implemented a comparative overview highlights how vari-
modified 2D ResNeXt-50-32x4d Deep Learn- ous machine learning and imaging techniques
ing Architecture. Ye et al. (2020) combined enhance tumor detection and classification in
diffusion basis spectrum imaging (DBSI) with pediatric patients.
a deep learning neural network (DHI model),
732 Innovations in Computational Intelligence: Trends and Technologies
This review provides a comparative analy- highlight its specific relevance to current
sis of different methodologies, imaging tech- studies:
niques, data sets, and accuracy rates in the Addressing a critical need: Tumor in brain
study of brain tumor of children. The integra- of children is a head most reason of tumor-
tion of advanced imaging methods with AI related deaths in children. Accurate and
significantly improves the accuracy and effec- timely detection of brain tumor can increase
tiveness of tumor classification and detec- the children survival rate.
tion. Future research should focus on further Building on existing research: Previous
improving these techniques and expand- studies have demonstrated the potential
ing collaborative data collection efforts to of CNNs in brain tumor detection in adult
enhance the robustness and generalizability populations. However, there is a scarcity of
of findings in pediatric brain tumor research. research specifically focused on pediatric brain
tumors, which present unique challenges due
to their diverse nature, smaller size, and vary-
4. Relevance to Current Study ing presentation across different age groups.
The proposed study, “Brain Tumor Detec- This study fills this gap by developing a CNN
tion in Child from MRI Images Using CNN,” model tailored for pediatric MRI analysis.
holds significant relevance within the broader Leveraging technological advancements:
landscape of medical imaging and artificial Recent advancements in deep learning and
intelligence research. The following points computational power have made CNNs more
Table 104.1: Comparison of various Reasearch Based Deep
Table 104.1. Learning methodologies for Brain Tumor detection
AQ4: S. No Author Publication Year Methodology Accuracy Research
Please respect to
provide
citation 1 S. Solanki12 2023 Deep CNN, 3-D 95% Young
for Table CNN
caption
and 2 Potadar, M. P13 2023 Deep CNN, 93% Young
source
info 3 Krishnapriya, 2023 Transformer-based 94% Young
S.Karuna14 models, GAN, deep
CNN
4 Jayaraj 2022 Fully convolutional Review paper Young
Ramasamy15 networks
5 Quon, J.L., et al 8 2020 Modified 2D 92% Child
ResNeXt-50-32x4d
deep learning
architecture
6 Ye, Zezhong, et 2020 DBSI +DNN (DHI 83% Child
al. 8 model)
7 Prince, Eric W et 2020 CNN + GA as 87% Child
al 8 ameta heuristic
optimizer
8 Bidiwala, S. and 2004 NN 72.7-85.7% Child
Pittman8
9 Arle, Jeffrey E8 1997 4 BPNN 58%-95% Child
Source:
AUTHOR'S SELF GENERATED
Pediatric brain tumor detection in MRI: A machine learning perspective 733
accessible and effective for medical image of sizable public datasets available in recent
analysis. This study utilizes these advance- years. The lifespan human connective initia-
ments to develop a sophisticated CNN model tive development (HCP) is one such initiative.
capable of accurately identifying and classify- To the greatest extent feasible, deep learning
ing brain tumors in children. will thus be able to supplement and improve
Potential for clinical impact: The proposed pediatric imaging. The use of neural networks
study has the potential to translate into real- and their enhanced models has greatly aided
world clinical impact by providing radiolo- research. In reality, many CNN designs consist
gists and clinicians with a powerful tool for of many layers, including extra normalization
early tumor detection, potentially improving layers and batch normalization. Further-
diagnostic accuracy and leading to more effec- more, each design has been highly developed
tive treatment strategies for young patients. employing concepts from probabilistic mod-
Contribution to scientific knowledge: This els and optimization. By using small patches
research will investigate the use of CNNs in as a computational advantage researchers
the diagnosis of pediatric brain tumors, add- studying brain magnetic resonance imaging
ing to the expanding body of knowledge in can train deep CNNs to provide suitable seg-
medical imaging and artificial intelligence. mentation algorithms. The medical imaging
The results of this investigation will be help- area, which typically uses simplistic designs,
ful for further study and advancement in this recognized this accomplishment widely6.
area. In addition, the development of genera-
tive adversarial networks (GANs) have led
to encouraging advancements in the study of
5. Gaps in Existing Knowledge brain tumors using GANs in medical imag-
It can be difficult to detect and classify brain ing investigations; nevertheless, few meth-
tumors in children due to their high degree of ods have been applied to children using MRI
variety in terms of size, location, and appear- data6.
ance. CNN models must be resilient to such However, GANs have the capability to
fluctuations. learn in supervised as well as unsupervised
There is insufficient pediatric data in the environments. It is evident that the image-to-
majority of publically accessible brain tumor image translation achieved by GANs might
datasets to support the training and valida- have several more beneficial applications in
tion of CNN models tailored to the distinct the field of medical imaging. Restoring MR
brain anatomy and tumor types of children. pictures obtained with specific objects, such
Brain tumors in children are frequently
tiny and challenging to identify in their early
stages. CNN models ought to be fine-tuned to
identify minute lesions and distinguish them
from healthy brain tissue.
Deep learning has the capability to
enhance the quality and usefulness of pediat-
ric imaging. Pediatric radiologists face many
obstacles in their quest to fulfill this promise,
including the creation of a wide range of data- AQ5:
Please
sets and careful labeling, many of which are Figure 104.1. Pediatric brain tumor detection provide
unique to pediatric imaging applications6. In and classification studies different model and source
their accuracy. for all the
order to solve this issue, the medical imaging figures
sector has already begun to make a number Source: AUTHOR'S SELF GENERATED
734 Innovations in Computational Intelligence: Trends and Technologies
motion, especially in a pediatric setting, can The ML model, developed for real-world
assist cut down on the number of rechecks6. clinical use, is expected to aid radiologists in
The self-learning capacity of new deep early and accurate diagnosis of brain tumors
learning approaches has completely changed in children, potentially improving treatment
the landscape of pediatric brain tumor outcomes.
research programs. Three studies, for instance,
demonstrate remarkable findings that exceed
95% accuracy in posterior FOSSA tumor 7. Conclusion
categorization. CNN has been recognized Machine learning algorithm has emerged as
for its achievements, but its full potential in a promising tool for pediatric brain tumor
brain MRI research has not yet been realized. detection and analysis using MRI images.
Before the trustworthy CNN applications Studies have shown that deep learning mod-
can be employed in clinics, more study in this els can achieve high accuracy in tasks such
area is still necessary. as tumor segmentation, feature detection, and
classification9. For instance, the proposed
6. Expected Outcomes deep learning-based model achieved high
accuracy and specificity in predicting individ-
The research aims to create a ML model for ual MR characteristics and tumor type. This
detecting and classifying brain tumors in chil- suggests its potential to aid radiologists, espe-
dren from MRI images, aiming for high accu- cially those without specialized neuroradiol-
racy, sensitivity, and specificity. ogy or neuro-oncology training, in making
The study will assess the ML model’s per- timely and accurate diagnoses. Furthermore,
formance using metrics like accuracy, sensitiv- the model could serve as a foundation for a
ity, specificity, precision, recall, and F1-score multi- pathological data for predicting molec-
to evaluate its ability to accurately detect and ular genetics using deep learning or machine
classify tumors11. learning techniques10.
The ML model will be compared to cur- However, challenges remain in implement-
rent pediatric brain tumor detection methods ing deep learning for brain tumor detection
to assess its strengths and weaknesses, and in children. These challenges include limited
identify any potential improvements. datasets with diverse and accurately labeled
data, the complex and variable nature of
pediatric brain tumors across different
age groups, and the need for interpretable
models7.
Future research should focus on overcom-
ing these challenges by: Expanding and diver-
sifying pediatric brain tumor image datasets.
1. Developing deep learning models that are
robust to variations in image appearance
and signal-to-noise ratio.
2. Creating interpretable models to under-
stand how they arrive at their decisions.
3. By addressing these challenges, deep
Figure 104.2. Flow diagram of F1 score learning has the potential to revolution-
(adapted from toward data science). ize pediatric brain tumor diagnosis and
Source:
AUTHOR'S SELF GENERATED
improve patient outcomes.
Pediatric brain tumor detection in MRI: A machine learning perspective 735
References [9] ZainEldin H, Gamel SA, El-Kenawy EM,
Alharbi AH, Khafaga DS, Ibrahim A, and
[1] Hamd ZY, Osman EG, Alorainy AI, Talaat FM. Brain tumor detection and classi-
Alqahtani AF, et al. The role of machine fication using deep learning and sine-cosine
learning in detecting primary brain tumors fitness grey wolf optimization. Cold Spring
in Saudi pediatric patients through MRI Harb Lab. 2023.
images. J Radiat Res Appl Sci. 2024. [10] Mahajan A, Burrewar M, Agarwal U, Kss
[2] Boyd A, Ye Z, Prabhu S, Tjong MC, et al. B, Mlv A, Guha A, Sahu A, Choudhari A,
Expert-level pediatric brain tumor segmen- Pawar V, Punia V, Epari S, Sahay A, Gupta
tation in a limited data scenario with step- T, Chinnaswamy G, Shetty P, and Moiyadi
wise transfer learning. Cold Spring Harb A. Deep learning based clinico-radiological
Lab. 2023. model for pediatric brain tumor detection
[3] Valjakka J, Mylläri J, Myllyaho L, Kivimäki and subtype prediction. Cold Spring Harb
J, and Nurminen JK. Anomaly localization Lab. 2023.
in audio via feature pyramid matching. In: [11] Hamd ZY, Osman EG, Alorainy AI,
2023 IEEE 47th Annu Comput Softw Appl Alqahtani AF, Alshammari NR, Bajamal
Conf (COMPSAC); 2023. O, Alruwaili SH, Almohsen SS, Almusal-
[4] Faria CMC. Epigenetic mechanisms regulat- lam RI, and Khandaker MU. The role of
ing neural development and pediatric brain machine learning in detecting primary
tumor formation: A review. J Neurosurg brain tumors in Saudi pediatric patients
Pediatr. 2011;8(2):1-9. through MRI images. J Radiat Res Appl Sci.
[5] Khuntia M, Sahu PK, and Devi S. Predic- 2024;17(3):1-10.
tion of presence of brain tumor utilizing [12] Solanki S, Singh UP, Chouhan SS, and Jain
some state-of-the-art machine learning S. Brain tumor detection and classifica-
approaches. Int J Adv Comput Sci Appl. tion using intelligence techniques: An over-
2022;13(5):1-6. view. IEEE Access. 2023;11:12870-12886.
[6] Islam S, et al. Generative adversarial net- doi:10.1109/ACCESS.2023.3242666.
works (GANs) in medical imaging: Advance- [13] Potadar MP, Holambe RS, and Chile RH.
ments, applications, and challenges. IEEE Design and development of a deep learning
Access. 2024;12:35728-35753. doi:10.1109/ model for brain abnormality detection using
ACCESS.2024.3370848. MRI. Comput Methods Biomech Biomed
[7] Boyd A, Ye Z, Prabhu S, Tjong MC, Zha Y, Engin Imaging Vis. 2023;12(1):1-8.
Zapaishchykova A, Vajapeyam S, Hayat H, [14] Krishnapriya S, and Karuna Y. A survey
Chopra R, Liu KX, Nabavidazeh A, Resnick of deep learning for MRI brain tumor seg-
A, Mueller S, Haas-Kogan D, Aerts HJWL, mentation methods: Trends, challenges,
Poussaint T, and Kann BH. Expert-level and future directions. Health Technol.
pediatric brain tumor segmentation in a 2023;13:181-201.
limited data scenario with stepwise transfer [15] Ramasamy J, Doshi R, and Hiran KK. Seg-
learning. Cold Spring Harb Lab. 2023. mentation of brain tumor using deep learn-
[8] Shaari H, Kevrić J, Jukić S, Bešić L, Jokić D, ing methods: A review. In: Proc Int Conf
Ahmed N, and Rajs V. Deep learning-based Data Sci Mach Learn Artif Intell (DSMLAI
studies on pediatric brain tumors imaging: ‘21); 2022.
Narrative review of techniques and chal-
lenges. Cold Spring Harb Lab. 2023.
105 Enhancing trust in the social internet of
things (SIoT): A comprehensive review
Gyanendra Kumara and Anuradha Misrab
Amity School of Engineering and Technology, Amity University, Lucknow, India
Abstract: An essential mechanism that enhances sensor network security is trust management. In order to
develop smart nodes for social connections, the IoT-enhanced social networks combines the internet of things
(IoT) with social networks. Service requestors must evaluate service providers’ levels of trustworthiness in order
to deal with misbehaving ones. In order to lower the danger of being exposed to malicious nodes, this study
suggests a novel trust management method for the social internet of things (SIoT) that anticipates which service
providers will be the most dependable for each requestor. A new paradigm known as the SIoT has emerged as
a result of the IoT integration of social networking. Building trust between things is still important, though.
This work investigates an effective paradigm for context-aware trustworthiness inference, which distinguishes
between familiarity and similarity trust. While similarity trust may be determined using both internal and
external similarity trust, familiarity trust can be determined using direct trust and suggestion trust. and lastly,
design to solve data sparsity and cold start difficulties, identify reliable nodes, and extract latent characteris-
tics from SIoT-enhanced social networks nodes, a matrix factorization approach is employed. A field of study
called SIoT integrates social networks and the IoT. The SIoT allows things to communicate with each other as
customers or service providers based on their social behavior. It should be simple and reliable for these items to
find the services they are looking for. Objects that discriminate and do not enhance the quality of services are
seen as sensible behavior. To tackle this problem, A trust management system with discriminative awareness is
introduced for service delivery in socially enabled IoT.
Keywords: Social internet of things, trust management, cyber security, trustworthiness inference framework,
matrix factorization
1. Introduction As the IoT enters into a new era of infor-
mation networks and infrastructure services,
The Internet of Things (IoT) is a network of the need for greater unification of the real
linked smart things that use common com- world with computational system increases.
munication protocols to exchange informa- As more objects become smarter and inter-
AQ1: The
tion and act together. It is more than simply connected, adopting a social approach is reference
a worldwide smart device network—it also essential for the IoT interaction paradigm [1]. citations
consists of auxiliary services, apps, and tech- In the social internet of things (SIoT), objects are not in
sequential
nology. Including nodes that may request and function as self-governing agents, retaining order,
their individuality while offering and request- please
supply services and work together to deliver a check.
ing services and information.
single service is the main goal of the IoT. IoT
The SIoT offers several advantages. Firstly,
has advanced significantly since its beginning, it allows the frame of the social network to
inspiring the creation of new scenarios and be adjusted as needed to ensure scalability
ambitions including social, industrial, and and navigability. The search space grows dra-
healthcare-related IoT. matically with the exponential increase in the
a
[email protected]; [email protected]
DOI: 10.1201/9781003606635-105
Enhancing trust in the social internet of things (SIoT): A comprehensive review 737
number of connected devices. [4]. The diverse are hundreds of millions of linked devices,
nature of the IoT and the vast amount of con- the IoT infrastructure is complicated. Add-
textual data it generates add to its complexity ing people to this setting is a workable solu-
[7]. The SIoT can efficiently locate goods and tion to get around its complexity. Data may
services by mapping a system of “friendly” provide significant insights once it has been
items, rather than relying on traditional, less processed, analyzed, and mined. But from
scalable internet discovery methods. Addi- a scalability standpoint, IoT search engines
tionally, social network analysis techniques have been observed to be non-scalable with
can be used to identify the problem within respect to the number of devices. [10], which
the socially enabled IoT. means they are unable to manage the enor-
The IoT concept entails utilizing common mous volume of requests that are sent. The
communication rules and special addressing IoT presents the risk of phony items, which
schemes to link a vast number of heterogene- might include individuals who are unable
ous smart items in the non-virtual world to to access the proper services and things that
the internet [2]. The principal benefit of the could act as malware and steal data from
IoT is in its ability to socialize these smart other objects. Smart dust and false particles
devices by emulating human behavior, so are among the attacker nodes in that network
forming a social network of smart items [3]. that are capable of stealing confidential data
The social interactions that form between about military circumstances (such as those
smart objects can optimize and support a at a nation’s border) [10]. Socially enabled
number of fundamental network operations IoT suggests integrating social networking
through the use of positive feedback [5]. The with the IoT to prevent such issues.
IoT system has to be able to automatically cre-
ate and maintain social relationships between
smart devices in order to achieve all of these 2. Related Work
goals. Over the past 10 years, social net- The literature survey in this paper “social-
working has been incorporated into a num- trust-aware variational recommendation”
ber of communication networks to aid with reviews relevant classical social recommenda-
resource searches and traffic routing. In many tion models and deep neural network-based
studies have looked into the potential use of models before positioning the proposed
social networking components in IoT solu- SOAP-VAE model in relation to these works.
tions whereby things may form social bonds Overall, the literature survey in the paper
on their own, following the social norms that contextualizes the proposed SOAP-VAE
their owners have established [3]. One of the model within the existing landscape of social
key elements of the internet of the future is recommendation models and deep neural
the IoT. Sensors and actuators are typically network-based approaches, highlighting its
used to link physical items to the internet contributions and advancements in address-
because they can quickly develop IoT tech- ing data sparsity and incorporating diverse
nology by offering a variety of services [6]. levels of influence in social recommender sys-
Object–object connectivity in the IoT refers tems [11].
to connecting different things. Virtual objects This research’s literature review explores
(VOs) are computer representations of real- various existing trust models and methods
world items that have been around for a while employed in wireless sensor networks to
[8]. Two of the biggest issues with the IoT detect and prevent malicious node attacks.
are scalability and data collecting [9]. Data It highlights the critical role of trust evalu-
collection via voice over satellite (VO) is ini- ation in assessing node reliability based on
tially a challenging operation. Because there past behavior [12]. Several trust models,
738 Innovations in Computational Intelligence: Trends and Technologies
such as game theory, Bayesian estimation, on evaluating the degree of trustworthiness
D-S evidence theory, and fuzzy logic, have of service requestors and confirming the
been developed for wireless sensor networks. dependability of service providers [15]. The
These models aim to identify malicious nodes mechanism can withstand various kinds of
through trust evaluation, providing a theo- maladroit nodes. presented a fission comput-
retical basis for further research. Trust evalu- ing-based lightweight query technique for
ation models are categorized into centralized maintaining privacy and trust in the IoT [16].
and distributed models, impacting data shar- The method encountered difficulties because
ing mechanisms and trust calculations during of resource constraints and infrastructure
the assessment phase. needs, but it attempted to establish confi-
The paper “Towards Privacy Preserving dence without penalizing opponents.
IoT Environments: A Survey” conducts a The paper “Discriminative-Aware Trust
comprehensive literature review on privacy Management for SIoT” introduces a dis-
issues and solutions in IoT settings. It high- criminative-aware trust management system
lights the importance of protecting IoT envi- (DATM) for service provisioning in SIoT
ronments and safeguarding user privacy by settings. This research addresses issues such
addressing technological challenges, such as as discrimination and selfish behaviors by
the vast heterogeneity of IoT systems, limited IoT devices through a combination of social
CPU and memory resources, and the need for interactions, context-based trust assessment,
IoT-specific security solutions [13]. and real-world data validation [17]. DATM,
The paper “Trust Management for Reli- a novel approach proposed in the study,
able Decision Making among Social Objects enhances trust management in SIoT by incor-
in the Social Internet of Things” reviews porating context-based trust estimation. It
existing finding in trust management strate- improves the accuracy of trust assessments
gies and related works within the SIoT frame- in dynamic IoT environments by consider-
work. It underscores the scarcity of reliable ing factors such as social similarity, service
management strategy specifically designed relevance, and the energy levels of service
for SIoT networks, emphasizing the need for providers.
specialized methods to ensure the trustwor-
thiness of social objects. In the ubiquitous
era, trust is recognized as a subjective concept 3. Methodology
that requires unique approaches and careful In order to increase confidence in the socially
consideration for effective management in enabled IoT, a comprehensive flow chart
SIoT contexts [14]. The paper proposes a new outlining the whole process is needed. This
reliable management strategy that incorpo- includes showing each stage of the process,
rates behavior-based trust metrics and peri- from the initial discovery of trust factors to
odic trust updates to enhance the reliability the model’s deployment and iterative tuning.
and credibility of decision-making processes The diagram outlines the process of
among social objects in the SIoT ecosystem. enhancing trust in the socially enabled IoT
This scheme is built upon insights synthesized through a structured flow chart. Here’s an
from related works. explanation of each step in the flow chart:
This study, “Matrix Factorization in SIoT,” Start: The process begins.
includes a literature review that encompasses Requirement Analysis and Identify Trust
a number of relevant papers in the area of Factor: This step involves analyzing the
trust management within the SIoT. suggested requirements for trust management and iden-
a bidirectional trust management system for tifying the key trust factors that need to be
fog computing in the IoT, with an emphasis considered.
Enhancing trust in the social internet of things (SIoT): A comprehensive review 739
Table 105.1. Overview of related works
AQ2: Reference Algorithm Year Pros and Cons
Please [18] Graph attention 2021 GATs are flexible and can be adapted to different
provide
citation network (GAT) types of graphs and tasks by adjusting the
for Table techniques attention mechanisms and network architecture.
1 GATs can be computationally intensive, especially
for large graphs with many nodes and edges,
which may lead to longer training times and
higher resource requirements.
[19] Leveraging 2019 The approach offers benefits such as enhanced
multidimensional security, objective trust evaluation, and
trust indicators, comprehensive assessment.
applying entropy- It also presents challenges related to complexity,
based weight resource requirements, scalability, and
assignment generalizability.
[20] Machine learning 2020 Machine learning algorithms can efficiently
algorithm process large amounts of data and make
predictions or evaluations quickly.
Privacy concerns robustness challenges.
[21] Search economics 2018 It is providing the best solutions in first
algorithm and generation.
K-means clustering It is time consuming algorithm so not provide a
algorithm better time.
[22] Genetic algorithm 2015 Suitable for big problems.
It is not good for find optimal solutions.
[23] Data-driven and 2017 Behavior-based trust computation.
behavior-based It is complex to design metrics.
approach
[24] The combination of 2021 Enhanced trust prediction data sparsity mitigation
matrix factorization, privacy concerns.
Hellinger distance, The use of matrix factorization and advanced
centrality and algorithms may introduce complexity to the trust
similarity measures, management mechanism.
and resilience
strategies.
[25] Fuzzy based job 2018 Considering the requirements for resources into
classification account.
It is not suitable for all environment.
[26] Data mining 2020 This approach enhances the accuracy and
algorithm in reliability of trust evaluations in SIoT
discriminative-aware environments.
trust-management The algorithm may introduce complexity in trust
AQ3:
Please system management systems due to the incorporation of
provide multiple context-building metrics and the three-
source for dimensional space for trust estimation.
the table
Source:
AUTHOR'S SELF GENERATED
740 Innovations in Computational Intelligence: Trends and Technologies
Design the Trust Mechanism: Based on the
identified trust factors, a trust mechanism is
designed to manage and evaluate trust within
the SIoT.
Trust Evaluation: This step involves evalu-
ating trust through two types of interactions:
Direct Interaction: Reliability is assessed
based on direct experiences and interactions
between SIoT nodes.
Indirect Interaction: Trust is inferred from
indirect sources, such as feedback and recom-
mendations from other nodes.
Implement Secure Communication Proto-
col: To ensure the security of trust evaluations
and interactions, a secure communication
protocol is implemented. AQ4:
Ensure Model Scales with Network Please
provide
Growth: The trust mechanism is designed to citation
scale effectively as the network grows, ensur- for Figure
ing that trust evaluations remain efficient and 1
accurate.
Test the Model and Deploy the Model:
The designed trust model is thoroughly tested AQ5:
and then deployed in the SIoT environment. Please
provide
End: The process concludes after the model source for
is successfully deployed. Figure 105.1. Proposed flow chart. the figure
The flow chart ensures that the trust man- Source: AUTHOR'S SELF GENERATED
agement framework is systematically devel-
oped, evaluated, and implemented, focusing to ensure that the trust model grows effi-
on both direct and indirect trust interactions ciently enough to accommodate large-
while maintaining security and scalability. scale networks without sacrificing speed.
Many IoT devices have limited comput-
4. Challenges ing and storage capacity. Implementing
a trust architecture that is both resource
The challenges at hand necessitate a multidis- efficient and effective is difficult.
ciplinary approach, encompassing expertise 3. Security and privacy: Ensuring the integ-
in IoT, security, data science, and network rity of the data used to calculate trust is
engineering. critical. Malicious devices or users may
1. Complexity in trust computation: Trust attempt to modify data in order to fraud-
in SIoT is driven by a variety of elements, ulently increase their trust scores. Collect-
including device reliability, data qual- ing and analyzing data to estimate trust
ity, user reputation, and context aware- levels might cause privacy issues. Balanc-
ness, making it difficult to aggregate these ing the requirement for data to calculate
measures into a single trust score. Main- trust while also protecting user privacy is
taining accurate trust scores in changing a key difficulty.
settings is similarly difficult. 4. Interoperability: SIoT networks fre-
2. Scalability: SIoT networks can include a quently comprise of heterogeneous devices
large number of linked devices. It is critical
Enhancing trust in the social internet of things (SIoT): A comprehensive review 741
that use diverse communication proto- approaches. The proposed models and mech-
cols, standards, and capabilities. Creat- anisms demonstrate superior performance in
ing a trust model that works seamlessly accuracy, convergence, attack resilience, and
across several platforms is tricky. A lack mitigating cold start problems.
of standardization in trust calculation Future research directions include verify-
techniques might impede the integration ing model properties in dynamic environ-
of various trust models and their compat- ments, exploring hypergraph-based social
ibility within a hybrid architecture. relations, and further enhancing trust man-
5. Trust dynamics: Trust evolves through- agement frameworks to adapt to evolving
out time depending on interactions and SIoT contexts. Overall, this review under-
experiences. Capturing the temporal com- scores the critical role of trust management
ponent of trust and updating trust scores in the effective deployment and operation of
dynamically is a difficult undertaking. SIoT systems.
Trust levels might differ depending on the
setting in which interactions occur. Incor-
porating context-awareness into the trust References
model to appropriately represent these [1] Atzori L, Iera A, Morabito G, and Nitti
variances is difficult. M. The social internet of things (SIoT)—
6. User behavior: Detecting and limiting the when social networks meet the internet
impact of malicious conduct by people or of things: concept, architecture and net-
devices is critical. This involves identifying work characterization. Comput Netw.
2012;56(16):3594-3608
phony identities, Sybil attacks, and other
[2] Pattar S, Buyya R, Venugopal KR, Iyen-
trust-manipulation techniques. Under- gar SS, and Patnaik LM. Searching for
standing and applying user behavior pat- the IoT resources: fundamentals, require-
terns into the trust model can improve its ments, comprehensive review and future
accuracy while increasing its complexity. directions. IEEE Commun Surv Tutor.
5. Conclusion 2018;20(3):2101-2132.
The IoT in social requires robust trust manage- [3] Bernal Bernabe J, Hernandez Ramos JL,
ment systems to ensure reliable and satisfac- and Skarmeta Gomez AF. TACIoT: multi-
tory services among numerous interconnected dimensional trust-aware access control sys-
objects. This review paper presents a compre- tem for the Internet of Things. Soft Comput.
hensive analysis of various trust management 2016;20(5):1763-1779.
frameworks in socially enabled IoT, highlight- [4] Nitti M, Atzori L, and Cvijikj IP. Friendship
ing the importance of social relationships and selection in the social Internet of Things:
challenges and possible strategies. IEEE
the classification of trust attributes. It dis-
Internet Things J.. 2015;2(3):240-247.
cusses innovative trust inference frameworks, doi:10.1109/JIOT.2015.2415520.
trust prediction mechanisms using Hellinger [5] Wang L, Wu H, Han Z, Zhang P, and Poor
distance, and discriminative-aware schemes HV. Dynamic resource matching for socially
like DATM that effectively detect and adapt cooperative caching in IoT networking. Proc
to behavioral biases. IEEE Glob Commun Conf (GLOBECOM).
The analysis reveals that decentralized 2017:1-6.
and hybrid trust propagation schemes offer [6] Tsiropoulou EE, Paruchuri ST, and Baras JS.
significant advantages over centralized sys- Interest, energy and physical-aware coalition
tems, which are prone to single points of fail- formation and resource allocation in smart
ure. The review highlights the importance of IoT applications. In: Proc 2017 51st Annu
hybrid frameworks that combine the advan- Conf Inf Sci Syst (CISS); 2017:1-6.
tages of both centralized and decentralized [7] Hussein D, Park S, Han SN, and Crespi
N. Dynamic social structure of things: A
742 Innovations in Computational Intelligence: Trends and Technologies
contextual approach in CPSS. IEEE Internet Comput Commun. 2018;119:156-166.
Comput. 2015;19(3):12-20. doi:10.1016/j.comcom.2017.11.009,
[8] Guo J, Chen I-R, and Tsai JJP. A survey of [19] Kim M, and Ko IY. An efficient resource
trust computation models for service man- allocation approach based on a genetic algo-
agement in internet of things systems. Com- rithm for composite services in IoT environ-
put Commun. 2017;97:1-14. ments. In: Proc 2015 IEEE Int Conf Web
[9] Chen IR, Bao F, and Guo J. Trust-Based Serv; 2015:543-550.
Service Management for Social Internet of [20] Nizamkari NS. A graph-based trust-
Things Systems. IEEE Trans Depend Secur enhanced recommender system for service
Comput. 2016;13:684-696. selection in IoT. In: Proc 2017 Int Conf
[10] Juhász J, and Bányai T. What Industry 4.0 Inventive Syst Control (ICISC); 2017:1-5.
Means for Just-In-Sequence Supply in Auto- [21] Kantarci B, and Mouftah HT. Trustworthy
motive Industry? In: Vehicle and Automo- sensing for public safety in cloud-centric
tive Engineering 2. Springer; 2018:226-240. Internet of Things. IEEE Internet Things J.
[11] Saied YB, Olivereau A, Zeghlache D, et al. 2014;1(4):360-368.
Trust management system design for the [22] Mendoza CVL, and Kleinschmidt JH. A
internet of things: a context-aware and distributed trust management mechanism
multi-service approach. Comput Secur. for the Internet of Things using a multi-
2013;39B:351-365 service approach. Wirel Pers Commun.
[12] Bao F, Chen I-R, and Guo J. Scalable, adap- 2018;103:1-10.
tive and survivable trust management for [23] Chen D, Chang G, Sun D, Li J, Jia J, and
community of interest-based internet of Wang X. TRM-IoT: A trust manage-
things systems. In: Proc Eleventh Int Symp ment model based on fuzzy reputation for
Autonomous Decentralized Syst (ISADS); Internet of Things. Comput Sci Inf Syst.
2013:1-7. 2011;8(4):1207-1228.
[13] Chen D, Chang G, Sun D, et al. TRM-IoT: [24] Imran M, Jabbar S, Chilamkurti N, and Rod-
a trust management model based on fuzzy rigues JJ. Enabling technologies for Social
reputation for internet of things. Comput Sci Internet of Things. Future Gener Comput
Inf Syst. 2011;8(4):1207-1228.. Syst. 2019;92:715-717.
[14] Chen I-R, Guo J, and Bao F. Trust manage- [25] Rabadiya K, Makwana A, and Jardosh S.
ment for SOA-based IoT and its application Revolution in networks of smart objects:
to service composition. IEEE Trans Serv Social Internet of Things. In: Proc 2017 Int
Comput. 2014;9(3):482-495. Conf Soft Comput Eng Appl (icSoftComp);
[15] Commerce BE, Jøsang A, and Ismail R. The 2017:1-8.
beta reputation system. In: Proc 15th Bled [26] Militano L, Nitti M, Atzori L, and Iera A.
Electron Commerce Conf; 2002. Enhancing the navigability in a social network
[16] Buchegger S, and Boudec J-YL. Performance of smart objects. Comput Netw. 2016;103:1-
analysis of the CONFIDANT protocol. In: 14. doi:10.1016/j.comnet.2016.03.013
Proc 3rd ACM Int Symp Mobile Ad Hoc [27] Amin F, Ahmad A, and Choi GS. Commu-
Netw Comput; 2002. nity detection and mining using complex
[17] Chae Y, CingiserDiPippo L, Sun YL. networks tools in Social Internet of Things.
Trust management for defending on-off In: Proc IEEE TENCON; 2018:1-6.
attacks. IEEE Trans Parallel Distrib Syst. [28] Atzori L, Iera A, and Morabito G. SIoT:
2015;26(4):1178-1191. Giving a social structure to the Inter-
[18] Tsai CW. SEIRA: An effective algorithm net of Things. IEEE Commun Lett.
for IoT resource allocation problem. 2011;15(11):1193-1195.
106 Fostering plant health: Utilizing
YOLOv7 framework to identify leaf
diseases
Rajya Lakshmi Ra, Sasikala Vb, Divya Likhitha Dc, Rishita Reddy
Chd, and VenkataSai Vidyasri Abhishainy Ae
Department of CSE, VNITSW, Guntur, India
Abstract: The early diagnosis of plant diseases is essential for maximizing crop output and quality since the
agriculture sector is critical to guaranteeing global food security. A key issue in agriculture is leaf disease, which
can result in large losses in agricultural productivity. Deep learning algorithms have been utilized more often
recently in identification and categorization. This study uses the cutting-edge object identification algorithm
YOLOv7 to provide a unique method for automated leaf disease diagnosis and is particularly suited for use in
precision agriculture because of its real-time and precise object identification capabilities. Popular object detec-
tion method YOLO (You Only Look Once) can identify and categorize things in the real world. This study
proposes a YOLO v7-based method for detecting leaf diseases. YOLO v7 utilizes a deep convolution neural
network (CNN) to automatically recognize and categorize leaf diseases in real time. To correctly identify the
illness types detected in the leaves, the system is trained using a sizable collection of plant photos. The YOLOv7
design is optimized to focus on identifying certain illness patterns, improving its accuracy.
Keywords: Object Detection, Plant disease detection Crop output, Deep Learning, Global Food Security, agri-
cultural productivity, YOLOv7, Convolutional Neural Networks
1. Introduction cause a drop in the total food supply. This
can lead to food scarcity and price increases,
Millions of people rely on the agriculture hurting both consumers and food produc-
industry for their nourishment, livelihoods, ers. Plant diseases have economic effects that
and economic stability, making it a key com- reach beyond the farm gate. They can cause
ponent of every country’s economy. How- significant economic losses at the regional
ever, this important industry is not immune and national levels. Reduced agricultural
to the devastating effects of plant diseases. output influences a country’s GDP and trade
These diseases can have significant impacts balance. Agricultural exports are important
on a variety of agricultural and economic fac- to many countries’ economies. Plant diseases
tors. Plant diseases can cause substantial crop can impair international trade, resulting in a
losses. Infected crops may become unmar- drop in export earnings. For example, India
ketable or produced in fewer quantities and is a large exporter of agricultural items such
of worse quality. These losses have a direct as rice and spices. Diseases that affect these
impact on farmers’ income and capacity to exports can have a knock-on effect on the
contribute to the economy. Plant diseases can national economy. To tackle plant diseases,
a
[email protected]; [email protected]; [email protected]; dchrishita2255@
gmail.com; [email protected]
DOI: 10.1201/9781003606635-106
744 Innovations in Computational Intelligence: Trends and Technologies
farmers frequently use more pesticides and for illiterate farmers, choosing YOLO V7, an
fungicides. While this may help to reduce dis- extraordinary development in object identifi-
ease transmission, it may also increase pro- cation algorithms. While existing algorithms
duction costs. Plant diseases can jeopardize a such as CNN, YOLO V4, and YOLO V5 have
country’s capacity to offer a stable and suf- made significant contributions to computer
ficient food supply for its people. Increased vision, our choice of YOLO V7 is motivated
food imports, trade deficits, and economic by an unwavering commitment to precision
fragility may result. and efficacy in addressing the unique chal-
Plant diseases have far-reaching economic lenges faced by farmers who may lack for-
and agricultural consequences. They have an mal education. YOLO V7, which stands for
impact on agricultural output, food security, “You Only Look Once,” is at the forefront
trade, livelihoods, and resource availabil- of object-detecting technology. It builds on
ity. Mitigating the impact of plant diseases the success of its forefathers by using sophis-
through early identification, research, and ticated features and optimization approaches
sustainable farming techniques is critical to to deliver unrivaled accuracy in detecting and
ensuring agriculture’s and the larger econo- classifying objects in photos. This degree of
my’s resilience and sustainability. precision is critical for our unique use in crop
A greater number of farmers are illiterate, disease identification. Farmers that rely on
making it difficult to identify plant illnesses. visual cues and visuals, such as illiterate farm-
They frequently employ chemical pesticides ers, can greatly benefit from the accuracy and
to safeguard their crops, even when they are speed provided by YOLO V7.
unsure if a disease exists. This can be danger- The heart of our invention is enabling
ous since pesticides pose health and environ- illiterate farmers to make educated disease-
mental dangers and are expensive. To assist control decisions. Farmers may easily col-
these farmers, we must devise methods of lect photographs of their crops using simple
teaching them about illnesses and safer farm- devices such as cell phones or cameras, thanks
ing techniques through the use of images, to our YOLO V7-powered application. These
videos, and local languages. This can improve photos are then analyzed by YOLO V7,
farming for both them and the environment. which quickly analyzes them to discover and
Deep learning technologies offer a viable identify illnesses with high accuracy. The ease
answer to the issues faced by illiterate farm- of use of this approach makes it accessible
ers in disease identification. Through picture and user-friendly, bridging the knowledge
analysis, these technologies can automate ill- gap that illiterate farmers may have with tra-
ness detection, removing the requirement for ditional disease detection methods.
literacy. Farmers may obtain a quick diagno- Our dedication to the YOLO V7 extends
sis of illnesses by merely photographing their beyond perfection to sustainability. Illiterate
crops, allowing for targeted treatments and farmers, who frequently use wide pesticide
lowering dependency on costly and danger- applications as a prophylactic strategy due to
ous pesticides. Implementing user-friendly, their incapacity to reliably diagnose illnesses,
smartphone-based apps with deep learning may now focus their treatments precisely.
algorithms can help illiterate farmers make This minimizes not only the financial burden
educated decisions, secure their crops, and of pesticide use but also the environmental
ensure sustainable and lucrative agricul- and health dangers connected with excessive
ture while protecting their health and the chemical use.
environment. The foundation of our disease detection
We are at the vanguard of innovation in system, YOLO V7, was chosen by us since we
our mission to modernize disease diagnosis are committed to providing illiterate farmers
Fostering plant health: Utilizing YOLOv7 framework to identify leaf diseases 745
with the most cutting-edge technology. We Midhun P. Mathew and Theresa Yamuna
want to usher in a new age of informed and Mahesh concentrated on identifying diseases
sustainable agriculture by utilizing the preci- in bell pepper plants. With a 93% accuracy
sion, speed, and accessibility of the YOLO rate, their program sought to identify diseases
V7. Thanks to this creative approach, even more easily for farmers by detecting bacterial
individuals without a formal education may patches early on. Venkatramaphanikumar
now preserve their crops, make wise choices, Sistla, Ramya Chowdary Puvvada [10], and
and contribute to food security while protect- Sai Shilpa Padmanabula R-CNN and YOLO-
ing the environment. based CNNs are two popular deep learning
techniques that are used to detect objects
with 90% accuracy.
2. Literature Survey
Wenjiang Huang et al. [1] proposed an
enhanced spectral index for monitoring and 3. Proposed Model
identifying winter wheat diseases. [2] Dr. K. This section has described the working process
Thanagadurai and K. Padmavathi presented of plant disease detection using YOLO V7
a computer paper on vision-based image
enhancement. [3] D.A. Bashish et al. (2010) 3.1. YOLO V7
proposed a novel method for Segmenting leaf
images into four distinct clusters utilizing the The YOLO V7 architecture was influenced
squared Euclidean distances by the applica- by preceding YOLO model architectures, par-
tion of k-means segmentation. Ultimately, ticularly the YOLO-R, and YOLO v3 [13].
a DL algorithm built on Back Propagation
approach is used to finish the classification
process. It was discovered that the system’s
overall accuracy for disease identification and
categorization was about 93%. [4] Zhou et
al. (2014) and Barbedo and Godoy (2015),
created image processing-based systems to
diagnose plant diseases with an accuracy of
90%.
In their study on machine learning meth-
ods for plant disease identification, Shruthi et
al. [5] discovered that a CNN can detect a
wide range of diseases. Using image process-
ing techniques and the K Nearest Neighbor
AQ1:
(KNN) algorithm, P. Srinivasan et al. [6] cre- Please
ated software for classifying groundnut leaf provide
in-text
infections, like Late leaf spot, Rust, Early citation
leaf spot, and Bud Necrosis. Mrunmayee et for figure
al.[7] presented a 90% overall accuracy tech- 1-7.
nique for disease identification and classifi-
cation. With an accuracy of 75.9%, Suhaili
Kutty et al. [8] suggested a method based
on RGB color components for diagnosing
watermelon leaf diseases, such as Anthrac- Figure 106.1. Workflow of object detection. AQ2:
Please
nose and Downey Buildup. Using YOLO V5, Source: AUTHOR'S SELF GENERATED provide
source
for all the
figures
746 Innovations in Computational Intelligence: Trends and Technologies
3.2. Extended efficient layer where (pw, ph) are the anchor box dimen-
aggregation network (EELAN) sions for the specific anchor box associ-
ated with the grid cell.
EELAN is the computational component at 3. Confidence Score:
the heart of YOLO V7. This structure is essen- It can represent the possibility of an object
tial for improving the abilities of the YOLO being present in a bounding box within a
V7 model to learn. This is accomplished by grid cell. It is predicted using a sigmoid
utilizing the “expand, shuffle, merge” method activation function:
cardinal,” which allows for ongoing develop-
confidence = σ(tconf)(5)
ment in the ability of the network to learn
without damaging the original gradient 4. Class Probabilities:
road. b) Scaling of the YOLO V7 Compound YOLO predicts class probabilities for
Model. each bounding box. These are computed
using the softmax activation function to
ensure that they sum to 1:
3.3. Scaling for the YOLO v7
class_probs = softmax(class_scores) (6)
compound model
5. Loss Functions:
Model scaling is largely used to change a YOLO uses several loss functions to train
model’s fundamental properties to produce the model. The most common loss func-
models that match a range of application tions include:
requirements. An important element of model Objectness Loss (Binary Cross-Entropy
scaling entails improving characteristics like Loss for Confidence):
the model’s depth (the number of stages),
Lobj= – [obj × log(confidence) + (1-obj)
width (the number of channels), and resolu-
× log(1-confidence)] (7)
tion (image size entered).
Localization Loss (MSE Loss for Bound-
ing Box Coordinates):
3.4. The equations used in YOLO v7
are (8)
1. Prediction of bounding box centre (x, y) Classification Loss (Cross-Entropy Loss
coordinates relative to the grid cell: for Class Probabilities):
Let (tx, ty) be the predicted values. These (9)
are obtained through the sigmoid func- where pi is the one-hot encoded true class
tion. The absolute coordinates (bx, by) in label.
the image are calculated as follows: 6. Non-Maximum Suppression (NMS):
bx=(cx+tx)× grid_width (1) NMS is employed to separate redundant
by = (cy + ty) × grid_height (2) and low-confidence bounding boxes. The
2. Prediction of bounding box width (w) equations for NMS are not part of the
and height (h) relative to the whole image: training process but are used during post-
Let (tw, th) be the predicted values. These processing for inference.
are obtained through the exponential
function to ensure positive values. The 3.5. Dataset
absolute width and height (bw, bh) are cal-
culated as follows: The collection includes more than 125,000
[12] images. 37 plant diseases are known to
bw=pw×etw (3) exist. The tea leaf disease diagnosis model was
bh = ph× eth (4) developed using the [14] original dataset and
the YOLOV7 network [15]. Graphs showing
Fostering plant health: Utilizing YOLOv7 framework to identify leaf diseases 747
various performance metrics for training, related to a variety of leaf ailments. These
testing, and validation sets are used to show annotations accurately define disease lesions,
the usefulness of the produced model. There patches, and affected areas, giving learning-
are 3 different kinds of loss shown: box, cat- facilitating ground truth labels to the [16]
egorization, and objectiveness. The box loss YOLO v7 model procedure.
measures how well an algorithm locates the
exact center and bounding box of an item. 3.7. Algorithm
How likely it is that an object will be found in
a particular location is measured by the con- Following are the stages that were involved in
cept of “objectiveness” as a metric. According training a unique YOLO v7 model:
to high objectivity, an object is likely to be
1. Setting up the environment: The first
present in the visible area of a picture. The
action involved creating the conditions
parameters of the model undergo significant
YOLO v7.
changes throughout the process. The model’s
2. Image collecting and presentation: Images
performance was consistently improved as
the iteration count went from 100 to 150. were accumulated to make a new dataset
that was given the name PlantifyDr To
annotate the dataset, it was uploaded to
3.6. Preprocessing and annotation of Roboflow.
data 3. Pre-processing and enhancement: The
To help computers understand and interpret data that was uploaded. Through aug-
the input, picture annotation entails classify- mentation and pre-processing stages,
ing and labeling data. which Roboflow’s algorithms choose.
An annotated dataset is essential for super- 4. Data splitting: Roboflow separated the
vised learning. Machine learning (ML), which data into data sets for validation, testing,
forms the basis for constructing models, ena- and training.
bling computers to process data, and produce 5. Choosing an explanation format: images
precise findings. The procedure of annotat- are annotated, and then after the PyTorch
ing had a crucial part in getting the dataset YOLO v7 technique was chosen.
ready for model training. Using leaf photos, 6. Snippet: PyTorch libraries of the PIP pack-
highlighting and naming important areas age were supplied by Roboflow.
7. Google Collab: The Colab enables the user
to run deep learning models. The YOLO
V7 model was trained, which probably
used a Tesla P100 GPU.
4. Results
4.1. Accuracy
A classification model’s overall correctness is
gauged by its accuracy. The accuracy formula
is
(10)
Figure 106.2. Leaf image examples for object
detection. A detailed analysis of CNN, YOLO V4,
and YOLO V7 machine learning models for
Source:
AUTHOR'S SELF GENERATED
748 Innovations in Computational Intelligence: Trends and Technologies
plant disease diagnosis produced findings The efficiency of the YOLO V7 model in
that demonstrate the higher accuracy of the reducing prediction errors and optimizing the
YOLO V7 model. The study found that the identification of plant disease is highlighted
YOLO V7 model was more accurate than by the reduced loss that was observed. This
the other models examined, highlighting its result emphasizes the possible benefit of using
exceptional efficacy in the identification of Transformer architectures in plant disease
leaf diseases. Compared to traditional meth- detection to obtain lower loss values when
ods like CNN and [17] YOLO V4, these compared to other models like CNN and
results highlight the potential of YOLO V7 YOLO V4.
architecture in improving the accuracy of
models. 4.3. Precision
Precision is employed to assess how well a
4.2. Loss model predicts positive outcomes. It is the
Loss measures the error between predicted ratio of true positive predictions—which
and actual values. It is utilized to optimize the include both true positives and false posi-
model during training. The loss formula is. tives—to the total number of positive fore-
casts. Here’s how to express precision using
(11) the formula:
A thorough analysis of ML models, like (12)
CNN, YOLO V4, and YOLO V7, in the realm
of identifying plant leaf diseases, revealed An exhaustive study of ML models, such as
that the YOLO V7 model demonstrated a sig- CNN, YOLO V4, and a YOLO V7 approach,
nificantly lower loss than the other models. revealed an interesting trend in accuracy per-
formance in the realm of identifying plant
leaf diseases [19]. Especially, the YOLO V7
model demonstrated far more precision
than the rest of the group, indicating that it
is quite accurate at predicting positive cases
and [18] reducing false positives. In contrast
to competing models like CNN and YOLO
V4, this result highlights the potential benefit
of YOLO V7 architectures for more accurate
disease identification of leaves.
Figure 106.3. Accuracy.
AUTHOR'S SELF GENERATED
Source:
Figure. 106.4. Loss. Figure 106.5. Precision.
Source: AUTHOR'S SELF GENERATED Source: AUTHOR'S SELF GENERATED
Fostering plant health: Utilizing YOLOv7 framework to identify leaf diseases 749
4.4. Recall The F1 score is computed using the following
formula:
Its definition is the proportion of real posi-
tive examples to true positive expectations. (14)
Recall is particularly important in scenarios
where the cost of false negatives is high. The
formula for recall is as follows: 5. Conclusions
To precisely identify and locate different leaf
(13)
diseases in plant photos, the state-of-the-
A detailed analysis of machine learning art object identification algorithm YOLOv7
methods, such as CNN, YOLO V4, and a is used. The model’s practical application is
YOLO V7 approach, in the realm of identify- increased by its capacity to identify many
ing plant leaf diseases, showed that the YOLO illnesses simultaneously in real-time. The
V7 model demonstrated noticeably stronger method achieves a high level of generaliza-
recall than its competitors. The YOLO V7 tion by training the model on a wide data-
may be more effective than competing mod- set including many plant species and disease
els like CNN and YOLO V4 for correctly kinds. By enabling prompt action to stop the
identifying persons at leaf disease detection, spread of infections, the initiative not only
as evidenced by the reported advantage in the solves the essential issue of early disease iden-
recall. tification but also advances precision agri-
culture. Due to the inclusion of YOLOv7, a
balance between accuracy and speed is main-
tained, making it appropriate for use in real-
world circumstances.
With the end of this study, computer
vision in agriculture will have advanced sig-
nificantly, promoting the adoption of sustain-
able farming practices.
References
Figure 106.6. Recall.
[1] Malone, Brian Seymour. Some Solids Resi-
Source: AUTHOR'S SELF GENERATED
dence Time Studies in Tapered Gas Fluidised
4.5. F1 score Beds. University of Surrey (United King-
dom), 1963.
Integrates both precision values and recall val- [2] Sardogan, Melike, Adem Tuncer, and Yunus
ues into a single value. Particularly it is used Ozen. “Plant leaf disease detection and clas-
when there is an uneven class distribution. sification based on CNN with LVQ algo-
rithm.” 2018 3rd international conference on
computer science and engineering (UBMK).
IEEE, 2018.
[3] Ranjana, V., et al. “Design and develop-
ment of efficient techniques for leaf disease
detection using deep convolutional neural
networks.” 2020 IEEE International Confer-
ence on Distributed Computing, VLSI, Elec-
trical Circuits and Robotics (DISCOVER).
IEEE, 2020.
Figure 106.7. F1 Score.
Source: AUTHOR'S SELF GENERATED
750 Innovations in Computational Intelligence: Trends and Technologies
[4] Moyazzoma, Raida, et al. “Transfer learn- automation, computational and technology
ing approach for plant leaf disease detection management (ICACTM). IEEE, 2019.
using CNN with pre-trained feature extrac- [12] Soeb, Md Janibul Alam, et al. “Tea leaf dis-
tion method Mobilnetv2.” 2021 2nd Inter- ease detection and identification based on
national Conference on Robotics, Electrical YOLOv7 (YOLO-T).” Scientific reports 13.1
and Signal Processing Techniques (ICREST). (2023): 6078.
IEEE, 2021. [13] Padmanabula, S. S., Puvvada, R. C., Sistla,
[5] Shukla, Vivek, Sweta Rani, and Ramesh V., and Kolli, V. K. K. (2020). Object Detec-
Kumar Mohapatra. “A New Approach for tion Using Stacked YOLOv3. Ingénierie des
Leaf Disease Detection using Multilayered Systèmes d Inf., 25(5), 691-697.
Convolutional Neural Network.” 2023 3rd [14] Vallabhajosyula, S., Sistla, V., and Kolli, V.
International conference on Artificial Intel- K. K. (2022). Transfer learning-based deep
ligence and Signal Processing (AISP). IEEE, ensemble neural network for plant leaf dis-
2023. ease detection. Journal of Plant Diseases and
[6] Bhosale, Yogesh H., et al. “Multi-olktjo4; Protection, 129(3), 545-558.
Plant and Multi-Crop Leaf Disease Detec- [15] Raghavendra, K. V., Naik, D. B., Ven-
tion and Classification using Deep Neu- katramaphanikumar, S., Kumar, S. D., and
ral Networks, Machine Learning, Image Krishna, S. R. (2014, November). Weather
Processing with Precision Agriculture-A based prediction of pests in cotton. In 2014
Review.” 2023 International Conference on International Conference on Computational
Computer Communication and Informatics Intelligence and Communication Networks
(ICCCI). IEEE, 2023. (pp. 570-574). IEEE.
[7] Khirade, Sachin D., and A. B. Patil. “Plant [16] Kumar, G. S., Venkatramaphanikumar, S.,
disease detection using image processing.” and Kishore, K. V. K. (2021, May). Smart
2015 International conference on comput- Farming-A Flexible Approach to Improve
ing communication control and automation. Crop Yield and Profit using Machine
IEEE, 2015. Learning Techniques. In 2021 2nd interna-
[8] Arnal Barbedo, Jayme Garcia. “Digital tional conference for emerging technology
image processing techniques for detecting, (INCET) (pp. 1-6). IEEE.
quantifying and classifying plant diseases.” [17] Padmanabula, S. S., Puvvada, R. C., Sistla,
SpringerPlus 2.1 (2013): 1-12. V., and Kolli, V. K. K. Object Detection
[9] Militante, Sammy V., Bobby D. Gerardo, and Using Stacked YOLOv3 Object Detection
Nanette V. Dionisio. “Plant leaf detection Using Stacked YOLOv3.
and disease recognition using deep learn- [18] Vallabhajosyula, S., Sistla, V., and Kolli, V. K.
ing.” 2019 IEEE Eurasia conference on IOT, K. (2024). A novel hierarchical framework
communication and engineering (ECICE). for plant leaf disease detection using residual
IEEE, 2019. vision transformer. Heliyon, 10(9).
[10] Padmanabula, Sai Shilpa, et al. “Object [19] K. V. Krishna Kishore, Sasikala V., Ven-
Detection Using Stacked YOLOv3.” Ingé- katramaphanikumar S. (2020). Impact and
nierie des Systèmes d Inf. 25.5 (2020): Implications of New Computing Technolo-
691-697. giesin Smart Farming: A Survey. Interna-
[11] Sandhu, Gurleen Kaur, and Rajbir Kaur. tional Journal of Advanced Science and
“Plant disease detection techniques: a Technology, 29(3), 11224 - 11231. Retrieved
review.” 2019 international conference on from http://sersc.org/journals/index.php/
IJAST/article/view/28021
107 A review on-malignancy detection in
lung nodules using CNN
Krish Yadava, Vinayakb, Pratibha Yadavc, Lakshmi Pandeyd,
Himanshu Sharmae, and Smita Yadavf
Department of Computer Science and Engineering, Babu Banarasi Das Institute of Technology and
Management, Lucknow, Uttar Pradesh, India
Abstract: In this work, a computer-based detection (CADe) system for lung nodule early diagnosis was pro-
posed by using low-dose computed tomography (LDCT). Since the low dosage images are very challenging, the
proposed method preprocesses the raw data to enhance (increase) the contrast of them. Next is coarse-grained
deep learning feature extraction, where we evaluate multiple deep learning architectures (AlexNet, VGG16, and
VGG19 networks) for the same. To enhance the selected feature set a procedure based on one genetic algorithm
(GA) is used for learning the top importance factors in early detection. Then different categories of classifier are
evaluated to ensure a lung nodule can be detected by the models successfully. The system is evaluated using 320
LDCT images from fifty individual patients. DETECTION SYSTEMS PUBLIC INTERNETDataBase for the
Early Detection of Lung Cancer is a program within I-ELCAP. The resultant system achieves the accuracy of
detection 96.25%, sensitivity 97.5% and specificity by using VGG19 architecture and SVM classifier.
Keywords: Malignancy Detection in Lung Nodules, Convolutional Neural Network, Computer-aided Diagno-
sis, Lung cancer
1. Introduction be the best option for this purpose. Addition-
ally, CT reviews useshaft radiation, which is
It is the second most common cancer in the hazardous if used frequently. In order to find
world, after breast cancer. Identification of lung cancer, a number of computer-backed
malignancy in lung nodule constitutes one discovery (CAD) systems have undergone
of the significant stages of lung cancer diag- extensive research and development. These
nosis. Lung nodules refer to little round or systems have proven to be more effective than
oval growths which appear on the lung tis- human radiologists at identifying lung lumps
sues. These may be benign or malignant. The and identifying malignant areas in medical
screening for lung cancer has its basis on the images.
rationale that the same is used for determin- This research develops a CADe system
ing the correct line of treatment and outlook that will enable early detection of lung cancer.
for the lung nodule. Four components make up the suggested
Numerous medical imaging modalities CADe system [1]
have been extensively researched for their abil-
ity to correlate lung bumps, including casket • Preprocess the initial raw LDCT data to
radiographs, low-curve CT, MRI reviews, CT minimize the disparity.
reviews, and casket X-rays. CT evaluations • Prize compact deep literacy features of the
have limits even though they are thought to LDCT image.
a
[email protected]; [email protected]; [email protected]; dpandeylakshmi043@
gmail.com; [email protected]; [email protected]
DOI: 10.1201/9781003606635-107
752 Innovations in Computational Intelligence: Trends and Technologies
• Ensure that the “uprooted aspects” are One way to depict the logistic regression
refined in favour of increasing the quality model is as follows:
of the discovery. Logistic(b0 + b1 * age + b2 * smoking
• Whether the subject had a malignant or + b3 * persistent cough + b4 * shortness of
benign LDCT scan must be ascertained by breath) = P(lung cancer = 1)
the optimised point vector. where the input features are age, smoking,
chronic cough, and shortness of breath, and
The suggested CADe system makes two the coefficients to be estimated are b0, b1, b2,
key contributions: b3, and b4.
The output probabilities (P(lung cancer =
• Examinations of deep learning charac- 1)) are guaranteed to lie between 0 and 1 by
teristic extraction from several different the logistic function.
architectures including the VGG16 and In order to train the model, you will want
VGG19 network. a labelled dataset in which each person’s sta-
• During the CADe system’s training phase, tus as a lung cancer patient is known. The
an intelligent genetic algorithm is used to maximum likelihood estimation method can
improve the extraction of deep learning be used to estimate the coefficients (b0, b1,
characteristics. Categorization happens b2, b3, and b4) using this dataset.
more quickly as a result of the reduction
in feature vectors and increase in recogni-
tion accuracy. 2.2. Related work
2.2.1. VGG16
2. Mathematical Model Li, F., et al. (2023). Deep learning-based pre-
diction of lung nodule malignancy on chest
2.1. Mathematical model for CT scans. Radiology, 20220942. (Accuracy
malignancy detection in Lung of 97.7% for lung nodule malignancy predic-
Nodules using CNN tion) [2]
Chen, J., et al. (2023). creation of a com-
Many methods can be taken into considera- puter-aided diagnosis system based on VGG16
tion in order to create a mathematical model for low-dose CT scans to identify lung can-
for the identification of lung cancer. Here is cer. 996–1005 in Journal of Digital Imaging,
an illustration of a basic logistic regression 36(5). (95.5% accuracy rate on low-dose CT
mathematical model: scans) creation of a computer-aided diag-
Assume for the moment that our dataset nosis system based on VGG16 for low-dose
has the following characteristics: CT scans to identify lung cancer. 996–1005
in Journal of Digital Imaging, 36(5). (95.5%
• Age (continuous variable) accuracy rate on low-dose CT scans) [3]
• Past history of smoking (binary variable:
0 for non-smoker, 1 for smoker)
2.2.2. VGG19
Prolonged cough (binary variable: 0 for no Sun, J., et al. (2023). A VGG19-based deep
chronic cough, 1 for chronic cough) learning model for lung cancer detection
Breathlessness (binary variable: 0 for no on chest X-rays with improved explainabil-
shortness of breath, 1 for shortness of breath) ity. IEEE Transactions on Medical Imaging.
Lung cancer incidence or absence is the (Accuracy of 96.5% on chest X-rays) [4]
aim variable (binary variable: 0 for absence, Zhang, X., et al. (2023). A multi-scale
1 for presence). VGG19 network for lung cancer detection on
A review on-malignancy detection in lung nodules using CNN 753
CT scans. Journal of Healthcare Engineering. 2. Image Preprocessing:
(Accuracy of 97.7% on CT scans) [5] • Make all images the same size to work
well with the model.
2.2.3. InceptionV4 • Adjust the brightness and contrast of
images so they look similar.
Wu et al., 2022: Proposed a CAD system for • Use techniques like flipping, rotating,
lung cancer diagnosis on chest X-rays using and cropping to make more training
InceptionV4. A test batch of 150 photos examples and help the model learn
resulted in the system being accurate up to better.
97.5% [6] 3. Data Splitting:
Based on InceptionV4, Li et al. (2023) • Divide your prepared dataset into two
developed a deep learning model to predict parts: one for training and one for
lung nodule malignancy using chest CT scans. testing.
A model was tested on a batch of 250 photos • A common split is to use 80% of the
and got an accuracy rate of 98.3 % for the data for training and 20% for testing.
classifier [2] • Make sure the training data has a good
mix of cancer and non-cancer cases to
2.2.4. LeNet avoid mistakes.
X-ray_COMPILEEl-Gabry et al. [13] LeNetA 4. Model Training:
computer-aided CAD system for lung cancer • Pick a good model for the job, like
detection on chest X-rays in 2021. The sys- InceptionV4 or LeNet, which are good
tem was tested using a set of 120 photos and at recognizing things in pictures.
demonstrated an accuracy rate of 94.5% [7]: • Teach the model using the training
Zhang et al., 2022: the aim of this study data, showing it the prepared images
was the detection of lung cancer using low- and telling it whether they have cancer
dose CT scans and they built a deep learning or not.
model based on LeNet. When tested on a set • Adjust how fast it learns and other set-
of 200 photos, the model hit an accuracy rate tings to make it work the best.
of 95.3% [5].
5. Model Evaluation:
2.3. Methodology • Test how well the trained model works
on the part of the data it hasn’t seen
Let us break down the detailed approach for
before.
developing a deep literacy model for cancer
• Measure how often it gets things right
discovery using casket X-rays or CT reviews:
and wrong, like its accuracy and other
1. Data Collection: important numbers.
• Start by collecting a large and varied • See if the model works well for differ-
dataset of chest X-rays or CT scans ent groups of people or different types
that includes both cancer and non- of cancer to spot any problems.
cancer cases.
Additional Considerations:
• Get these images from different places
like hospitals, clinics, and public • Transfer Learning: Suppose about using
datasets. knowledge from models that formerly
• Make sure the dataset represents dif- know a lot about filmland. This can help
ferent types of people, health condi- your model work more, especially when
tions, and cancer stages. you do not have a lot of data.
754 Innovations in Computational Intelligence: Trends and Technologies
• Making It Explainable: Produce ways to encouraging issues, pressing the pledge of AI-
explain why the model makes the opin- driven medical opinion results.
ions it does. This can help understand We erected a solid frame for training and
how it works and find any problems or assessing our CNN model through the gath-
limits. ering, preprocessing, and scrupulous cura-
tion of a sizable dataset. The model passed
2.4. Critical analysis meticulous training, hyperparameter tuning,
and strict confirmation. It was erected with
The experimental outcomes, expressed in a personal armature. As a result, it has dem-
terms of different metrics, of the proposed onstrated a prominent position of proficiency
Computer-Aided Diagnosis (CADe) system in secerning between lung images that are
for early cancer detection are reported. These healthy and those that are reflective of malice.
metrics are assessed with respect to various The model’s performance was assessed,
designs, configurations, the amount of the and the results were positive. This offers a
classifier features, the type of classifier utilized, regard into the ground- breaking eventuality
and whether or not the suggested Genetic of AI in abetting croakers in the early discov-
Algorithm (GA) is applied (YES or NO). ery of cancer, perfecting patient issues, and
The performance standards evaluated for reducing the burden on the healthcare system.
vibrant pairings of classifiers and deep point
extractors include Accuracy [8].
References
3. Conclusion [1] “Early Lung Cancer Detection Using Deep
Learning Optimization.”doi:https://doi.
In this design, we delved the creation of a org/10.3991/ijoe.v16106.13657.
Convolutional Neural Network (CNN) for [2] J. Li, Mu et al., “Deep Learning Predicts
the pivotal task of lung cancer opinion from Malignancy and Metastasis of Solid Pul-
medical filmland, specifically CT reviews monary Nodules from CT Scans,” SSRN
and casket X-rays. Our work on this design Electron. J., no. 4, 2022, doi: 10.2139/
has handedTable perceptive information and ssrn.4285500.
107.1 Captain: [3] J. Chen et al., “Performance of Deep-Learn-
Performance Comparison of CNN ing Solutions on Lung Nodule Malignancy
Table 107.1. Models Using Different Classifiers Classification: A Systematic Review,” Life,
for Lung Cancer Detection vol. 13, no. 9, p. 1911, 2023, doi: 10.3390/
Model Classifier Accuracy
life13091911.
VGG16 Bayes 77.87 [4] J. Sun et al., “Neoadjuvant Camrelizumab
AQ1:
Please MLP 83.53 Plus Platinum-Based Chemotherapy vs
provide
table cap- Random 84.83 Chemotherapy Alone for Chinese Patients
tion cita- forest With Resectable Stage IIIA or IIIB (T3N2)
tion and
VGG19 Bayes 76.43 Non–Small Cell Lung Cancer,” JAMA
source.
Oncol., vol. 9, no. 569, pp. 1348–1355,
MLP 87.69 2023, doi: 10.1001/jamaoncol.2023.2751.
NOTE ADDED Random 87.11 [5] L. Zhang et al., “Anlotinib plus icotinib
forest as a potential treatment option for EGFR-
InceptionV4 Bayes 80.08 mutated advanced non-squamous non-small
MLP 87.83 cell lung cancer with concurrent mutations:
final analysis of the prospective phase 2, mul-
Random 85.81 ticenter ALTER-L004 study,” Mol. Cancer,
forest vol. 22, no. 1, pp. 1–11, 2023, doi: 10.1186/
Source: s12943-023-01823-w.
AUTHOR'S SELF GENERATED
A review on-malignancy detection in lung nodules using CNN 755
[6] C. Wu et al., “Altered temporal-parietal Dourado Júnior, and V. H. C. de Albuquer-
morphological similarity networks in non- que, “Lung nodule malignancy classification
small cell lung cancer patients follow- in chest computed tomography images using
ing chemotherapy: an MRI preliminary transfer learning and convolutional neu-
study,” Brain Imaging Behav., vol. 16, no. ral networks,” Neural Comput. Appl., vol.
6, pp. 2543–2555, 2022, doi: 10.1007/ 32, no. 15, pp. 11065–11082, 2020, doi:
s11682-022-00709-7. 10.1007/s00521-018-3895-1.
[7] E. A. Rhea El-Gabry et al., “Intratumoral [9] H. Akamatsu et al., The Japanese Lung
CD103+ CD8+ T cells predict response Cancer Society Guideline for non-small
to PD-L1 blockade.,” J. Immunother. can- cell lung cancer, stage IV, vol. 24, no. 7.
cer, vol. 9, no. 4, Apr. 2021, doi: 10.1136/ Springer Singapore, 2019. doi: 10.1007/
jitc-2020-002231. s10147-019-01431-z.
[8] R. V. M. da Nóbrega, P. P. Rebouças Filho,
M. B. Rodrigues, S. P. P. da Silva, C. M. J. M.
108 Next-generation parking system
using IoT
K. Geethaa, Gude Ganesh Satya Kumarb, Raghu Ram C.c, and
A. G. Sreedevid
Department of Computer Science and Engineering, Amrita School of Computing, Amrita Vishwa
Vidyapeetham, Chennai, India
Abstract: This project revolves around the implementation of a cutting-edge parking system utilizing Internet
of Things (IoT) technology. The system functions through an intricate network of sensors and devices, auto-
mating the parking process for efficiency and security enhancement. At its core, the system integrates IR sen-
sors to detect incoming vehicles, triggering servo motors to open the parking gate and allocate a slot for the
vehicle. When all parking slots are occupied, a clear indication is displayed on an LCD monitor, prohibiting
further entries until a space becomes available. Upon a vehicle exiting the parking area, the display dynamically
updates, signaling the availability of a slot. The project further ensures the security of the Arduino board by
incorporating measures against physical tampering. In the event of any unauthorized physical interference, the
IR sensor promptly activates a buzzer, emitting an alarm, fortifying the protection of the system. The entire
system is interconnected with a ThingSpeak account, facilitating remote access and monitoring capabilities for
users via smartphones or web interfaces through cloud integration. This seamless connectivity empowers users
to oversee parking availability and security status in real-time, enhancing convenience and peace of mind. By
harnessing IoT technology, this next-generation parking system not only optimizes parking management but
also prioritizes security measures, offering a comprehensive solution for modern parking infrastructure.
Keywords: Arduino board, IR Sensor, LCD monitor, buzzer, ThingSpeak, IOT, cloud
1. Introduction connectivity to automate the parking process
while prioritizing security measures. At its
The elaboration of technology has continu- core, the system employs Infrared (IR) detec-
ally revolutionized colorful angles of our tors in confluence with servo motors to descry
lives, and the realm of parking systems is incoming vehicles, allowing entry and alloca-
no exception. In this period of interconnect- tion of available parking places. A vital point
edness and robotization, the integration of of this system is its dynamic monitoring capa-
Internet of effects (IoT) technology has paved bility, which incontinently updates the status
the way for innovative and effective results in of parking places on an TV examiner, enabling
managing parking spaces. This design encap- real-time mindfulness of parking vacuity also,
sulates the development and perpetration of a the design emphasizes robust security meas-
slice- edge coming- generation parking system ures by incorporating mechanisms within the
exercising IoT principles, aimed at revolu- Arduino board to descry and warn against
tionizing the conventional parking structure. any physical tampering attempts. Any unau-
The design’s abecedarian ideal is to produce a thorized hindrance triggers an alarm through
sophisticated parking system that seamlessly the IR detector, icing the system’s integrity
integrates detectors, bias, and pall- grounded and fortifying its resistance to tampering or
a
[email protected]; [email protected]; [email protected]; dag_sreedevi@
ch.amrita.edu
DOI: 10.1201/9781003606635-108
Next-generation parking system using IoT 757
sabotage. likewise, the integration of pall of parking slot availability. Here, a binary or
technology via ThingSpeak regard connec- categorical variable model can represent the
tivity extends the system’s functionality to status of parking slots—occupied or vacant.
druggies’ smartphones or web interfaces. Implementing a time-series model enables
This aspect empowers druggies with remote dynamic updates on the LCD monitor, ensur-
access to cover parking vacuity and security ing real-time visibility of available parking
status in real- time, enhancing convenience spaces based on changes in slot status. The
and furnishing a comprehensive result to security feature of the system necessitates a
ultramodern parking operation challenges. mathematical model to assess the likelihood
In substance, this design represents an emul- of physical tampering attempts. Probability
sion of technological invention and practical modeling could help estimate the probability
mileage, aiming to review traditional parking of such events occurring, while response time
systems. By using IoT technology, it trials to analysis would determine the system’s reac-
streamline parking operations, enhance secu- tion time to trigger the alarm upon detecting
rity measures, and offer druggies an intuitive, any unauthorized interference.
connected experience in managing parking Cloud connectivity and remote monitoring
spaces. add complexity to the model. Network mod-
els are essential to understand the dynamics
1.1. Mathematical model of next- of data transmission via cloud integration,
factoring in bandwidth, latency, and commu-
generation parking systems
nication protocols to enable real-time moni-
Developing a mathematical model for a next- toring on smartphones or web interfaces.
generation parking system using IoT involves Reliability and system performance are
intricate considerations across various com- critical considerations. Mathematical mod-
ponents and functionalities. One crucial els for reliability could involve assessing the
aspect to model is the arrival rate of vehi- individual reliability of system components,
cles, which can be represented statistically or such as the Arduino board, IR sensors, and
through probability distributions to estimate servo motors. These models may encompass
the frequency of cars entering the parking reliability block diagrams or fault tree analy-
area. Incorporating sensor response times, sis to understand and optimize the system’s
particularly those of the IR sensors detecting overall reliability.
vehicles, requires modeling to determine the Each component’s mathematical model
time elapsed from detection to gate opening requires a deep understanding of the system’s
and parking slot allocation.Another critical characteristics, including sensor response
element is the monitoring and management times, data transmission protocols, and
AQ1:
Please
provide
figure
captions
and
source
for all Figure 108.1. Figure 108.2.
figures. Source: Source:
Parking area Car Parking Spaces
AUTHOR'S SELF GENERATED
AUTHOR'S SELF GENERATED
visualization Availability
758 Innovations in Computational Intelligence: Trends and Technologies
technical specifications. Employing various system incorporates an amalgamation of
mathematical techniques such as queuing sophisticated cameras and alert systems. This
theory, control theory, probability distribu- setup not only deters potential threats but
tions, and statistical analysis, the models aim also ensures complete traceability of events,
to predict, analyze, and optimize the system’s thereby enhancing the security and peace
performance, efficiency, and reliability. of mind for both drivers and their vehicles
Developing a comprehensive mathematical within the parking facility.
model entails empirical data collection, theo- Moreover, this paper delivers a holistic
retical analysis, and computational simulations proposal for a smart parking architecture,
to validate and refine the models. Ultimately, encompassing an intricate blend of hardware
these models contribute to enhancing the park- and software components. This comprehen-
ing system’s effectiveness, reliability, and over- sive proposal outlines the necessary equip-
all performance in real-world applications. ment required for deploying an effective
Thing speak cloud platform real data smart parking system. It goes a step further
This paper introduces groundbreaking contri- by providing detailed flowcharts that meticu-
butions designed to overhaul the traditional lously delineate essential equipment and ser-
parking experience into a seamlessly efficient vice features. These visuals offer an exhaustive
and user-centric environment. At the fore- understanding of the proposed system’s func-
front is the transformative approach to pay- tionality, showcasing its capability to seam-
ment systems, catering comprehensively to lessly integrate technology, equipment, and
diverse user groups. The implementation of a services to optimize parking operations while
versatile payment system addresses the needs prioritizing user convenience and safety.
of both subscribed and non-subscribed users.
Subscribers benefit from the convenience of 1.2. Architectural diagram
automatic payments facilitated by personal-
ized badges, while non-subscribers have access
to designated terminals for manual transac-
tions, complemented by electronic payment
options. Furthermore, the paper emphasizes
the evolution of parking reservation mecha-
nisms by introducing a cutting-edge smart
parking system. This system empowers users
by enabling remote spot reservations via a
mobile application. Beyond mere reserva-
tions, it incorporates advanced features like
guidance and vehicle location services, signifi-
cantly enhancing the overall parking experi-
ence by providing seamless navigation and
real-time information to drivers.
In addition to user-centric advancements,
the paper places a paramount focus on ele-
vating parking security standards. Recogniz-
ing the significance of safeguarding parked
vehicles and ensuring driver safety, the imple-
mented security system stands as a testament
to this commitment. With a comprehensive
design tailored to address both external
intrusions and internal safety concerns, the
Figure 108.3:
Entrance and Exit
Scenario:
AUTHOR'S SELF GENERATED
Architectural diagram
Next-generation parking system using IoT 759
2. Conclusion System based on Wireless Sensor Networks,
in Proceedings of the 1st International Vir-
In conclusion, this paper has presented a tual Conference on Intelligent Transporta-
transformative vision for the parking indus- tion Systems, Slovakia, 26–30 August 2013.
try, focusing on enhancing user experience, [2] Vasieis Karagiannis, A Survey on application
bolstering security measures, and proposing layer protocols for the Internet of Things,
a comprehensive smart parking architecture. Transaction on IoT and Cloud Computing
The overarching goal has been to revolution- 2015, ISSN: 2331-4753 (Print) ISSN: 2331-
ize the conventional parking paradigm into 4761 (Online).
a user-friendly, efficient, and secure environ- [3] Thanh Nam Pham, Ming-Fong Tsai, Duc
ment. By introducing a versatile payment Bing Nguyen, Chyi-Ren Dow and Der-Jiunn
Deng, A Cloud-Based Smart-Parking System
system catering to various user categories,
Based on Internet-of-Things Technologies.
including subscribers and non-subscribers, IEEE Access, Vol 3, September 2015.
the paper addresses diverse needs and pref- [4] M. Fengsheng Yang, Android Application
erences, optimizing convenience for all. Development Revelation, China Machine
The integration of a smart parking system, Press, 2010.
allowing remote reservations and providing [5] Yanfeng Geng and Christos G. Cassandras,
guidance services through a mobile applica- A New Smart Parking System Based on
tion, represents a significant leap forward Optimal Resource Allocation and Reserva-
in accommodating modern drivers’ expecta- tions. IEEE Transaction on Intelligent Trans-
tions. Emphasizing the paramount impor- portation Systems, Vol 14, April 2013.
tance of security, the paper has underscored [6] Cui Shiyao, Wu Ming, Liu Chen, and Rong
the implementation of a robust security sys- Na, The Research and Implement of the
tem, leveraging advanced cameras and alarm Intelligent Parking Reservation Manage-
ment System Based on ZigBee Technology.
systems to ensure comprehensive protection
Measuring Technology and Mechatronics
for parked vehicles and drivers. Furthermore, Automation (ICMTMA), January 2014.
the comprehensive proposal for a smart park- [7] K. Ashokkumar, Baron Sam, R. Arshad-
ing architecture, detailing both hardware and prabhu, and Britto, Cloud Based Intelligent
software components with insightful flow- Transport System. Procedia Computer Sci-
charts, serves as a blueprint for deploying an ence, Vol 50, pp. 58–63, 2015.
efficient and technologically advanced park- [8] Z. Ji, I. Ganchev, M. O’Droma, and X.
ing system. Overall, this paper’s contribu- Zhang, A cloud-based intelligent car parking
tions offer a holistic approach to transform services for smart cities, in Proc. 31st URSI
parking operations, laying the groundwork General Assembly Sci. Symp (URSI GASS),
for a future where parking experiences are Aug. 2014.
safer, more efficient, and tailored to meet the [9] Hamada R. H. AI-Absi, and Patrick Sebas-
diverse needs of today’s drivers. tian, Vision-Based Automated Parking Sys-
tem, in 10th International Conference on
Information science, 2010.
Acknowledgment [10] Sarfraz Nawaz, Christos Efstratiou, and
Celia Mascolo, Parksense: A smartphone
The authors gratefully acknowledge the stu- based sensing system foron street parking, in
dents, staff, and authority of IOT department Cambridge University International Journal
for their cooperation in the research. of Computer Applications (0975–8887), Vol
169, No. 1, July 2017, 18.
[11] B. K. Konstantinos Domdouzis and C.
References Anuba, An experimental study of the effects
[1] Samaras, A., Evangeliou, N., Arvanitopou- of different medium on the performance of
los, A., Gialelis, J.; Koubias, S., Tzes, A., RFID system, Vol 21. Advanced Engineering
KATHODIGOS – A Novel Smart Parking Informatics, 2011.
109 A review on smart health diagnosis
Kishan Kumar Rauniyara, Vijay Yadavb, Padmini Mishrac, Sanskar
Singhd, and Ashish Kumar Yadave
Computer Science and Engineering, Babu Banarasi Das Institute of Technology and Management,
Lucknow, India
Abstract: The field of Machine Learning has experienced a boost in popularity, finding extensive use across
various sectors, including the healthcare industry. The Early identification of Disease helps in the proper and
well-timed medication. Smart Health Diagnosis is a disease prediction system in which the user provides the
symptoms as an input. The system analyzes these symptoms and come up with the probability of the disease as
an output. Disease Prediction is done by implementing various Supervised Machine Learning Algorithms like
Naïve Bayes, Random Forest, KNN, Decision Tree, etc.
Keywords: Machine learning, healthcare, symptoms based disease prediction
1. Introduction to patients. The disease prediction system,
predicated upon patient symptoms, prognos-
Machine learning entails instructing comput-
ticates various ailments. A large number of
ers to enhance performance by leveraging
current models focus on a single and complex
example data or historical data. It encom-
disease per analysis. Examples of such anal-
passes the exploration of computer systems
yses include those for diabetes, cancer, and
capable of acquiring knowledge from both
skin conditions. There are very few systems
data and experience. It two facets: Training
and Testing, Train/Test methodology serves in place that can examine multiple diseases.
as a mechanism for gauging the precision of Therefore, our Smart Healthcare Diagnosis
your model through the delineation of train- focus on general and common diseases giv-
ing and testing phases, thereby elucidating the ing customers precise and instantaneous dis-
efficacy and robustness of the model’s predic- ease forecasts, including the documentation
tive capabilities. The training data, constitut- that may succor in addition to the projected
ing the largest subset of the original dataset, sickness. We are going to investigate approxi-
serves to train the machine learning model by mately 40 commonly known diseases such as
imparting knowledge to the algorithms, ena- Influenza (Flu), Common Cold Gastroenteri-
bling them to formulate predictions for the tis, Hypertension, Asthma, diarrhea. Diverse
designated task. Subsequently, the test data- methodologies employed in disease predic-
set, an independent subset with akin features tion encompass the utilization of Machine
and class probability distribution, functions Algorithms, including Naïve Bayes, Decision
as a benchmark for evaluating the model Tree, Random Forest, and the k-mean algo-
post-training. The healthcare industry strate- rithm. Furthermore, constructing a disease
gically employs Machine Learning to propel prediction system often involves leveraging
advancements in its methodologies, thereby Big Data. Traditional disease risk model pre-
enhancing the delivery of superior services diction typically entails the application of
[email protected]; [email protected]; [email protected];
a
[email protected]; [email protected]
d
DOI: 10.1201/9781003606635-109
A review on smart health diagnosis 761
Machine Learning and supervised learning the prepared dataset. Figure 109.1: Flow Dia-
algorithms, utilizing labeled training data to gram The effectiveness of the trained models
train the models. One of the primary advan- is assessed using critical metrics like precision,
tages of smart health diagnosis is its acces- recall, F1-score, and accuracy. To validate the
sibility. People, especially those in remote models, a distinct test set is employed. The sub-
or underserved areas, can consult health- sequent step involves creating a user-friendly
care professionals without the need to travel web application that enables real-time predic-
long distances. Smart health diagnosis is a tions based on user-input symptoms. Ethical
revolutionary approach that promotes early considerations are integral, addressing pri-
detection and monitoring of various medi- vacy, security, and transparency concerns in
cal conditions. Wearable devices and mobile accordance with healthcare regulations. The
applications for smartphones enable real- process is completed by instituting a feedback
time tracking of essential health indicators, loop for ongoing enhancement, allowing the
including heart rate, blood pressure, and glu- model to adapt over time by incorporating
cose levels. Analyzing this data helps in iden- new data. This iterative approach ensures the
tifying health issues promptly, allowing for machine learning system remains flexible and
timely intervention and preventive measures. responsive to emerging trends and advance-
The incorporation of artificial intelligence ments in the field, ultimately contributing to
and machine learning in recent times has more precise and dependable predictions of
significantly broadened the horizon of intel- various diseases.
ligent health diagnosis. The future of predict-
ing diseases with machine learning includes
using genetic, wearable, and IoT data for
remote monitoring. Innovations like block-
chain secure data, and explainable AI makes
models easier to understand. Federated learn-
ing keeps information private, and continu-
ous improvement involves advanced imaging.
This combination aims for more accurate,
personalized, and proactive healthcare, trans-
forming how we predict and manage diseases.
2. Methodology
To construct a machine learning model for
the anticipation of various illnesses based on
symptoms, the initial phase involves gather-
ing a diverse dataset that encompasses both
symptoms and diagnosed conditions. This
dataset undergoes meticulous processing,
including cleansing, standardization, and
normalization to guarantee data quality. Sub-
sequently, relevant symptoms are prioritized
through feature selection using statistical
AQ1:
methods. Following this, models for multi- Please
class classification, such as decision trees or Figure 109.1. Data flow diagram. provide
source for
random forests, are selected and trained with Source:
AUTHOR'S SELF GENERATED the figure
762 Innovations in Computational Intelligence: Trends and Technologies
3. Literature Review concerns and desiring a more in-depth insight
into their physiological states. It places spe-
Rajneesh Thakur, Mansha, Pranjal Sharma, cific emphasis on those facing mental health
Dhruv: This document introduces an struggles, such as depression and anxiety. Our
advanced computational system crafted to system plays a crucial role in aiding them to
anticipate potential health conditions based navigate these obstacles and attain a higher
on symptoms, employing sophisticated math- quality of daily living. An outstanding feature
ematical and technological approaches. Its of the system lies in its adeptness at forecasting
primary objective is to mitigate hospital con- diseases based on an individual’s symptoms.
gestion and streamline healthcare procedures, This predictive capability can significantly
simplifying the work for medical profession- impact disease detection, providing individu-
als. The system employs four distinct method- als with the opportunity to seek timely medi-
ologies, highlighting an impressive accuracy cal intervention and enhance their prospects
rate of 94%, underscoring its considerable of recovery. By presenting a user-friendly and
proficiency. The integration of this technology comprehensive approach to health manage-
provides numerous benefits, including early ment, our system aims to empower people to
identification of illnesses, expedited access to lead healthier and more enriching lives, par-
treatment, and improved patient outcomes. ticularly those grappling with mental health
Additionally, the system’s capability to store challenges and those simply seeking to opti-
user input data for future enhancements is a mize their well-being.
valuable characteristic. It utilizes diverse tech- Anuj Kumar, Mr. Analp Pathak: The pri-
niques, such as decision trees and deep learn- mary objective of this review paper is to
ing, to address a broad range of diseases and develop a robust disease prediction system
presents information in a readily understand- using Machine Learning algorithms, primar-
able format. Looking forward, it is essential ily based on symptoms reported by patients.
to explore innovative approaches to enhance The paper employs four distinct Machine
the system further. Learning algorithms to make predictions and
Raj H. Chauhan, Daksh N. Naik, Rinal A. has achieved an impressive mean accuracy
Halpati, Sagar Kumar Patel, Mr. A. D. Praja- rate of over 95%. This high accuracy not
pati: This review article aims to assist indi- only represents a significant improvement
viduals who experience heightened concerns over previous work but also enhances the
about their well-being and seek insights into system’s reliability, thereby ensuring a higher
their bodily functions. Our primary objec- level of user satisfaction when compared to
tive is to promote their health, particularly existing solutions. One notable feature of the
for those grappling with mental health chal- system is its data retention capability, where
lenges such as depression and anxiety. This it stores the information provided by users,
platform can aid in overcoming these issues along with the diagnosed disease, in a data-
and enhance the overall comfort of their base. This stored data serves as a valuable his-
daily lives. Furthermore, our system excels in torical record, facilitating future treatments
forecasting illnesses based on an individual’s and contributing to more efficient health care
symptoms and provides uplifting messages management. In addition, the paper intro-
and visuals. The notable aspect is its inclusiv- duces a user- friendly Graphical User Inter-
ity, as it is accessible to everyone without any face (GUI) to enhance the system’s interaction
restrictions, catering to those keen on main- with users, making it exceptionally easy to
taining their health operate
Shweta Agarwal, Dr. Chander Prabha, Dr. Gopi Battineni, Getu Gamo Sagara,
Meenu: This platform serves as a valuable tool Nalini Chinatalapudiand Francesco
for individuals contending with health-related
A review on smart health diagnosis 763
Amenta: In this paper, the focus is on the with valuable insights into potential diseases,
critical issue of diagnosing chronic diseases, making their job easier and more accurate.
a task of immense importance in health- Dr CK Gomathy, Mr. A. Rohith Naidu:
care. To address this challenge, it is para- The primary objective of this disease predic-
mount to select the most suitable methods tion system is to accurately forecast diseases
and models to ensure accurate and reli- based on user-input symptoms. By collecting
able decision-making. The report rightfully symptoms from users, the system generates
underscores the potential dangers associ- disease predictions with an impressive aver-
ated with some computer models, which age prediction accuracy probability of 100%,
can be susceptible to manipulation with indicating its high reliability. This system
the addition of fake data. Such vulner- was effectively implemented using the Grails
abilities in diagnostic models could have framework and is designed to offer a user
life threatening consequences, sparking a friendly and easily accessible environment.
broader debate on the appropriateness of Being a web-based application, this system
relying on these computer-driven systems provides users the flexibility to access it from
in healthcare. These reviews of prediction anywhere and at any time, enhancing its con-
models play a pivotal role in identifying the venience and availability. The system employs
most robust and trustworthy methods for various Machine Learning algorithms, includ-
diagnosing chronic diseases. They serve as ing Decision Trees, Random Forests, and the
safeguards to minimize the risks associated Naïve Bayes Algorithm, to predict diseases
with incorrect or manipulated diagnoses. based on user-provided symptoms. Data pro-
In the rapidly evolving landscape of health- cessing is carried out using Machine Learning
care technology, emerging smart computer techniques, such as Random Forest, Decision
techniques like machine learning, cognitive Tree, and Naïve Bayes, which have collec-
computing, and deep learning are becoming tively yielded an impressive system accuracy
increasingly indispensable for comprehend- of 98.3%. These machine learning capabili-
ing the complexities of chronic diseases. ties are designed to not only predict diseases
Priyanka J. Panchal, Sheenah A. Mhaskar, but also to successfully anticipate disease
Tejal S. Ziman: This paper discusses an algo- outbreaks.
rithm used for predicting diseases based on
a person’s symptoms. In this system, there’s
a menu where you can choose from various 4. Recommendation and Future
symptoms. You can select any five symptoms Work
from the list, and the algorithm will then pre- The future of disease prediction using machine
dict the disease that might be causing these learning holds great promise for revolution-
symptoms. Additionally, the system can even izing healthcare. Machine learning, a subset
suggest common medications that are usu- of artificial intelligence, is increasingly being
ally prescribed for the identified disease. The harnessed to enhance our ability to forecast,
main goal of this system is to catch diseases prevent, and manage various diseases. The
in their early stages. It helps in diagnosing ill- potential lies not only in expanding the data-
nesses before they become too serious. This is set to encompass a broader spectrum of dis-
important because early diagnosis often leads eases but also in leveraging the health reports
to more effective treatment and better out- of patients for model training, enhancing the
comes for the patient. Moreover, doctors can accuracy and scope of predictive algorithms.
also use this system to assist them in making A noteworthy addition to this landscape is
accurate diagnoses. It can help healthcare pro- the incorporation of live consultations with
fessionals avoid confusion and provide them specialized doctors, providing a real-time
764 Innovations in Computational Intelligence: Trends and Technologies
and personalized dimension to healthcare 98.3%, showcasing its proficiency in predict-
services. This interactive approach fosters a ing diseases. The utilization of machine learn-
more comprehensive understanding of indi- ing expertise ensures successful prediction of
vidual health profiles and facilitates timely disease outbreaks, underlining the system’s
interventions. Furthermore, the integration efficacy.
of healthcare chatbots stands as a valuable
resource, addressing general queries and
offering accessible information to users. This References AQ2:
Please
multifaceted approach, combining advanced [1] Rajneesh Thakur, Mansha, Pranjal Sharma, provide
machine learning algorithms with augmented Dhruv. 2022. Diseases Prediction Using in-text
citation
patient- doctor interactions and AI-driven Classification Algorithm. International Jour- for all
chatbot support, holds great promise in nal for Research in Applied Science and references
revolutionizing disease prediction, making Engineering Technology (IJRASET) ISSN:
2321-9653; IC Value: 45.98; SJ Impact Fac-
healthcare more proactive, personalized, and
tor: 7.538 Volume 11, Issue VIII.
responsive to the diverse needs of individuals. [2] Raj H. Chauhan, Daksh N. Naik, Rinal A.
Halpati, Sagar Kumar J. Patel, and A. D.
5. Conclusion Prajapati. 2020. Disease Prediction using
Machine Learning. Volume: 07. Interna-
The primary objective of this disease pre- tional Research Journal of Engineering and
diction system is to forecast diseases based Technology (IRJET).
on user-provided symptoms. The system [3] Shweta Agarwal, Chander Prabha, and
takes user symptoms as input and generates Meenu Gupta. 2021. Chronic Diseases Pre-
a disease prediction with an average accu- diction Using Machine Learning – A Review.
racy probability of 100%. The user-friendly Volume 25. Annals of R.S.C.B., ISSN:
interface ensures easy accessibility, allowing 1583-6258.
users to employ the system remotely at their [4] Anuj Kumara, and Analp Pathakb. 2021. A
Machine Learning Model for Early Predic-
convenience. The system, being web-based,
tion of Multiple Diseases to Cure Lives. Vol-
enhances accessibility, enabling users to uti- ume 12. Turkish Journal of Computer and
lize it anytime, anywhere. This comprehen- Mathematics Education.
sive review aims to assess the performance, [5] Gopi Battineni, Getu Gamo Sagaro, Nalini
constraints, and potential applications of Chinatalapudi, and Francesco Amenta.
healthcare software. The insights gained 2020. Applications of Machine Learning
from this review can guide future developers Predictive Models in the Chronic Disease
in creating Disease Predictability Software, Diagnosis. Journal of personalized medicine.
fostering personalized patient care. The sys- [6] Priyanka J. Panchal, Shaeezah A. Mhaskar,
tem employs Decision Tree, Random Forest, and Tejal S. Ziman. APR 2020. Disease Pre-
and Naïve Bayes algorithms for disease pre- diction Using Machine Learning. Volume 3.
diction, with data processed through machine IRE Journals.
learning algorithms such as Random Forest, [7] C. K. Gomathy and A. Rohith Naidu. Oct–
2021. The Prediction of Disease Using
Decision Tree, and Naive Bayes. Notably,
Machine Learning. Volume 05 International
the system achieves a high accuracy rate of Journal of Scientific Research in Engineering
and Management (IJSREM).
110 Android app threat analysis techniques:
a review
Ujjwal Shuklaa, Gyanendra Kumarb, Syed Haidar Abbasc, and
Vaibhav Paliwald
Computer Science and Engineering, Babu Banarasi Das Institute of Technology and Management,
Lucknow, India
Abstract: The realm of cybersecurity undergoes constant evolution, characterized by an ongoing struggle
between malicious entities and defenders. Within this context, the identification of malicious Android applica-
tions emerges as a pivotal defense measure. This comprehensive review navigates the diverse landscape of meth-
odologies employed to detect and counteract these threats. The exploration encompasses three primary avenues
for detection: signature-based, behavior-based, and machine-learning approaches. Signature-based methods
rely on predefined patterns, behavior-based strategies scrutinize real-time actions and machine learning tech-
niques leverage data patterns. Each avenue is thoroughly dissected, shedding light on its strengths, limitations,
and evolving nuances. Delving further, the paper investigates the intricacies of feature extraction and selec-
tion, essential steps in preparing data for effective machine learning models. Feature extraction condenses raw
data, while feature selection hones in on the subset of attributes critical for distinguishing between benign and
malicious behavior. Lastly, the exploration extends to dynamic analysis and behavioral patterns, revealing the
importance of real-time scrutiny and the identification of anomalous behavior. This synthesis of methodologies
and techniques aims not only to encapsulate the present state of the field but also to pave the way for future
advancements in fortifying Android ecosystems against malicious intrusions.
Keywords: Cybersecurity, Android, malware detection, machine learning approaches, feature extraction and
selection
1. Introduction become a focal point for malicious entities
seeking to exploit vulnerabilities and com-
“Securing the Android ecosystem is not promise user security. In response to this
just a technological challenge; it’s an escalating threat landscape, the development
ongoing narrative of innovation and and deployment of sophisticated cybersecu-
adaptation. As we delve into the intricate rity measures have become imperative. This
landscape of malicious Android applica- comprehensive review delves into the intri-
tion detection, we embark on a journey cate strategies and methodologies employed
where vigilance meets sophistication, and for the detection and mitigation of malicious
the security of digital experiences is not Android applications.
just a goal but an evolving imperative.” The exploration commences by scrutiniz-
ing the foundational pillar of signature-based
The ubiquitous use of Android devices has detection, delving into how predefined patterns
significantly transformed the way we interact and signatures are leveraged to identify known
with digital information and services. How- malware. Moving beyond the limitations of sig-
ever, this widespread integration has also nature-based approaches, the review navigates
a
[email protected]; [email protected]; [email protected]; dpaliwalvaibhavkns@
gmail.com
DOI: 10.1201/9781003606635-110
766 Innovations in Computational Intelligence: Trends and Technologies
through the dynamic terrain of behavior-based articles to choose from various Machine Learn-
detection. Here, the focus is on understanding ing approaches that are to be included for mal-
how the analysis of application behavior in ware detection in an Android application [1].
real-time can unveil previously unseen threats All these criteria allowed for the discern-
based on deviations from normal patterns. ing of the methodologies that are being used
A pivotal turn in the narrative directs in the present day and age. To fulfil this pur-
attention toward the transformative power pose a broad search was done using various
of machine learning in the realm of malware sources, such as Google Scholar, and multiple
detection. Machine learning models, with other online databases such as IEEE Xplore,
their adaptive and evolving nature, showcase IJERT, arXiv, Preprints, etc.
promising capabilities in discerning complex The following are the criteria that were
patterns and anomalies indicative of mali- used for selection:
cious intent. This segment scrutinizes various
machine learning algorithms, shedding light • Admissible: The criteria for the selec-
on their strengths and limitations in the con- tion of the articles were done based on
text of Android application security. being relevant to malware detection in an
The journey doesn’t conclude with detec- Android application.
tion methodologies alone; it extends to the • Published recently (last 5 years): It was
crucial realms of feature extraction and selec- the main criterion for selection because it
tion. These processes are fundamental in dis- is vital to keep track of the latest techno-
tilling the most relevant and discriminative logical developments.
characteristics that enable accurate identifi- • Experimental only: This included studies
cation of malicious signatures or behaviors. must be experimental.
Furthermore, the review unwraps the layers
of dynamic analysis, emphasizing the real- 2.2. Dataset collection
time examination of application behavior
The dataset was obtained from CCCS-CIC-
during execution, and the scrutiny of behav-
AndMal-2020 [2]. CCCS-CIC-AndMal-2020
ioral patterns that signify malicious intent.
is publicly available dataset produced in 2020
As we traverse through these methodolo-
by jointly Canadian Centre for Cyber Secu-
gies, it becomes apparent that the landscape
rity and the Canadian Institute for Cyberse-
of Android application security is dynamic
curity. This dataset consists of 200K benign
and continually evolving. The intricate inter-
and 200K malware samples that combine to
play between these detection methods, their
become 400K Android applications with 14
effectiveness, and their adaptability to emerg-
prominent malware categories and 191 emi-
ing threats forms the crux of this exploration.
nent malware families.
Ultimately, this review seeks not only to dis-
The dataset is labelled and categorized
sect these methodologies but also to provide
into corresponding families. The malware
insights into their cohesive integration, offer-
applications are divided into eight categories
ing a holistic understanding that fortifies the
such as sensitive data collection, media, hard-
Android ecosystem against the ever-evolving
ware, actions/activities, internet connection,
landscape of cybersecurity threats.
CandC, antivirus and storage and settings.
2. Methodology 2.3. Training and testing through
algorithms
2.1. Overview of the selection criteria
After the classification the dataset is divided
Multiple selection criteria were set up to provide into malicious and benign categories [3]. We
the template for the selection of the studies or then, start to train a machine learning model
Android app threat analysis techniques: a review 767
using the permissions feature of Android. 3.3. Dynamic analysis and
Grouping or Stacking algorithms is a learn- behavioral patterns
ing technique, which combines multiple
individual models to create a better overall The method used in cybersecurity to under-
prediction. The training of the model is done stand the behavior of a program or code
by using the output of several models as input in real-time is called dynamic analysis. In
for one model to produce better predictions. dynamic analysis, there is no need to analyze
the static code instead the software is exe-
cuted in a controlled environment to observe
3. Thematic Overview its behavior, it is often called a sandbox.
Behavioral patterns are the characteristics,
3.1. Detection techniques actions, and activities of a malicious appli-
There are various techniques used for the cation. These applications always showcase
detection of malware in Android applications identifiable behavior that separates them
including signature-based, behavior-based, from benign applications.
Machine Learning, and Deep Learning-based
techniques and approaches. 3.4. Feature extraction and selection
In a signature-based approach, the prede- Feature extraction plays a very important role
fined patterns or signatures are used to iden- in the preparation of data for machine learn-
tify malicious Android applications. It is done ing and pattern recognition. In this process,
by comparing their code and behavior against the raw data is converted into a condensed
a known database or malware signatures. set of pertinent features, to encapsulate vital
The behavior-based approach is used information for model training,
to assess an Android application’s runtime The curation of a subset containing the
behavior to determine its potential threat. most pertinent features from the extracted set
This approach flags applications that exhibit is the essence of feature selection. Acknowl-
suspicious or harmful actions. edging that not all features contribute equally
The Machine Learning technique enables to a machine learning model’s performance,
the detection of malicious Android applica- and some might introduce noise or redun-
tions, based on the patterns and characteris- dancy, feature selection methods come into
tics learned from large datasets. It is capable play.
of evolving and improving over time.
3.2. Machine learning models for 4. Critical Analysis
malware detection This thorough review delves into the world of
Android malware detection, breaking down
For malware detection, the machine-learning
the primary methods used to identify mali-
models utilize algorithms and computational
cious applications. The paper looks at three
techniques to identify patterns and features
key approaches: signature-based, behavior-
associated with a malicious application. These
based, and machine learning. These methods
machine-learning models can gain predictive
have their strengths and weaknesses, and the
ability by learning from large datasets of benign
review provides a clear understanding of how
and malicious applications. Some of the com-
they work.
monly used machine learning models are:
The discussion on dynamic analysis and
• Decision Trees recognizing behavioral patterns is a key focus.
• Random Forests This means not just looking at what an app
• Support Vector Machines (SVM) is, but also how it behaves—a crucial aspect
768 Innovations in Computational Intelligence: Trends and Technologies
when dealing with ever-changing threats. One noteworthy implication is the signifi-
The review also touches on using machine cance of dynamic analysis, where the focus
learning to enhance detection, combining it shifts from static characteristics to the behav-
with smart feature selection for more precise ior of applications. This proactive approach is
outcomes. crucial in identifying new, previously unseen
However, it could give more insight into threats. The paper suggests that incorporat-
the challenges these methods face and how ing machine learning into this dynamic analy-
they play out in the real world. While it men- sis further refines the detection process.
tions a hybrid approach, a deeper dive into The implications extend to the future
the hurdles and potential benefits of such a of Android malware detection. Transpar-
strategy would add more depth. ent machine learning models are proposed,
The future work section is forward-think- ensuring that the inner workings of detec-
ing, suggesting areas for further research. tion algorithms are understandable and trust-
It emphasizes the need for new, innovative worthy. Real-time threat intelligence sharing
approaches, transparent machine learn- emerges as a collaborative solution, fostering
ing models, real-time threat intelligence, a collective defense against emerging threats.
collaborative efforts, and educating users. Educating users about safe practices becomes
These suggestions pave the way for ongo- integral in this ecosystem.
ing improvements in dealing with malicious In practical terms, the synthesis indicates
Android applications. a need for constant innovation in detection
In conclusion, this review is a valuable methods. The hybrid approach proposed in
resource for those in cybersecurity, research, the paper, combining multiple techniques,
and development. It provides a clear under- sets the stage for further research and devel-
standing of Android malware detection and opment. The ultimate goal is to create a holis-
offers insights into strategies for countering tic defense mechanism that can adapt to the
evolving threats. Its analysis and recommen- ever-changing tactics of malicious Android
dations contribute to the ongoing efforts to applications.
stay ahead of malicious Android applications. This synthesis not only provides a roadmap
for improving Android malware detection
but also highlights the broader implications
5. Synthesis and Implications for the cybersecurity landscape. It emphasizes
This review paper weaves together insights the need for collaboration, transparency, and
from various approaches to detect malicious ongoing education to safeguard users against
Android applications. By combining signa- emerging threats.
ture-based, behavior-based, and machine-
learning methods, the synthesis reveals a
comprehensive strategy for dealing with the 6. Recommendations for Future
dynamic landscape of cybersecurity threats Work
on Android devices. Despite significant strides in Android mal-
The synthesis highlights the importance of a ware detection, there exist promising avenues
multi-faceted approach. While signature-based for further exploration and advancement.
methods are effective against known threats, The following areas present opportunities for
behavior-based analysis steps in to identify future research and development:
unfamiliar, suspicious activities. Machine
learning, with its ability to adapt and learn 1. Innovative hybrid approaches:
from patterns, enhances the overall detection Investigate unconventional hybrid method-
capability. This combination acts as a robust ologies that seamlessly integrate signature-
shield against the evolving nature of malware. based, behavior-based, and machine-learning
Android app threat analysis techniques: a review 769
techniques to optimize their synergies and The pivotal role played by dynamic analy-
mitigate individual limitations. sis and behavioral pattern recognition in mal-
2. Transparent machine learning models: ware detection cannot be overstated. These
Develop machine learning models that techniques provide real-time insights into
prioritize interpretability and transpar- application activities, empowering a proactive
ency, facilitating a deeper understanding response to emerging security threats. Addi-
of decision-making processes and instill- tionally, the strategic use of feature extraction
ing trust in automated detection systems. and selection significantly enhances the effi-
3. Real-time threat intelligence fusion: cacy of machine learning models, contribut-
Explore the fusion of real-time threat ing to the development of more accurate and
intelligence feeds into detection systems resilient detection systems.
to enable swift adaptation to emerg- As we navigate the intricate realm of
ing threats, vulnerabilities, and evolving Android security, a holistic comprehension of
attack patterns. these detection methods becomes crucial for
4. Collaborative Android ecosystem: cybersecurity professionals, researchers, and
Foster collaboration among cybersecurity developers. By harnessing the strengths of sig-
experts, app developers, and stakeholders nature-based, behavior-based, and machine-
in the Android ecosystem to co-create a learning approaches, a more robust defense
more secure and resilient environment for can be forged against the persistent evolution
end-users. of malicious Android applications.
5. User empowerment through education:
Investigate strategies to enhance user
awareness and education on secure References
mobile practices, empowering them to [1] Md Naseef-Ur-Rahman Chowdhury,
make informed decisions and proactively Ahshanul Haque, Hamdy Soliman, Moham-
prevent malware infections. mad Sahinur Hossen, Tanjim Fatima, and
Imtiaz Ahmed. Android Malware Detec-
Diligent attention to these aspects will tion using Machine Learning: A Review.
undoubtedly contribute to the ongoing evo- arXiv:2307.02412 (2023).
lution of adaptive and effective strategies for [2] Pinar G. Balikcioglu, Melih Sirlanci, Ozge A.
Kucuk, Bulut Ulukapi, Ramazan K. Turkmen
countering the menace of malicious Android
and Cengiz Acarturk. Malicious code detec-
applications, ensuring sustained security tion in android: the role of sequence charac-
within the Android ecosystem. teristics and disassembling methods. Int. J.
Inf. Secure. 22, 107–118 (2023). https://doi.
7. Conclusion org/10.1007/s10207-022-00626-2.
[3] Aasthaa Bohra, Gayatri Shahane, Sakshi
In summary, this review delves into the persis- Shelke, and Shalu Chopra. Android Malware
tent security challenges posed by the evolving Detection. International Research Journal of
landscape of malicious Android applications. Engineering and Technology (IRJET), Vol-
The exploration and scrutiny of three primary ume 10, Issue 04, 2023.
detection methodologies—signature-based, [4] Eralda Caushaj and Vijayan Sugumaran.
behavior-based, and machine learning— Classification and Security Assessment of
underscore the complexity of combating Android Apps. Springer, 2023.
[5] Madihah Mohd Saudi, Muhammad Afif
dynamic threats. Each method exhibits dis-
Husainiamer, Azuan Ahmad, and Mohd
tinct merits and drawbacks, emphasizing the Yamani Idna Idris. iOS mobile malware
necessity of a comprehensive and multifac- analysis: a state-of-the-art. Springer, 2023.
eted approach to tackle Android malware’s
ever-changing nature.
111 AI Powered Legal Documentation
Assistant
Yogesh Shekhawata, Utkarsh Tiwarib, Syed Hasan Mehdic, and
Himanshu Vaishyd
Department of Computer Science and Engineering, Babu Banarasi Das Institute of Technology and
Management, (Affiliated to AKTU Lucknow, Uttar Pradesh and Approved by AICTE, New Delhi),
Lucknow, India
Abstract: The AI-powered legal documentation assistant for India is an innovative platform that incorporates a
fine-tuned version of GPT-3.5, specifically tailored for legal applications. Integrated through the OpenAI API,
this advanced language model enhances the platform’s capabilities for the creation, review, and management
of legal documents in India. The user-friendly tool facilitates the generation of accurate and compliant legal
documents for legal professionals, individuals, and businesses through guided prompts. The fine-tuned GPT-
3.5 model offers an intelligent layer, providing context-aware suggestions and corrections. With features such
as cloud storage for document accessibility and mobile app capabilities for offline work, the project aims to
optimize efficiency, reduce the risk of legal disputes, and democratize the legal documentation process in India.
Keywords: AI powered Documentation, AI legal assistant, GPT-3.5, Open AI API, GPT in legal Documenta-
tion, Legal Chatbot, Legal copilot
1. Introduction highlight the swift and remarkable progress
in large language model performance but also
Legal documents often feature complex and emphasize the potential for such models to
extensive legal jargon, posing a formidable revolutionize the provision of legal services in
challenge for lawyers navigating past cases. society [1].
Understanding such documents can be time- Responding to these challenges, the AI-
consuming, creating difficulties for individu- powered legal documentation assistant for
als lacking a legal background [2]. India emerges as a groundbreaking solution,
Recent advancements in language mod- seamlessly blending advanced technology
els, exemplified by GPT-3.5, have exhib- with user-friendly design. This project har-
ited exceptional prowess in the legal realm. nesses the capabilities of artificial intelligence,
In a groundbreaking experiment assessing incorporating the versatility of the Django
zero-shot performance, GPT-3.5 displayed backend, the cross-platform features of the
remarkable superiority in the entire uniform React Native frontend, and the ChatGPT
bar examination, surpassing both human model to redefine the handling of legal docu-
test-takers and previous models. On the ments. Large language models like ChatGPT
multistate bar examination, GPT-3.5 show- play a pivotal role in this innovation, utiliz-
cased an impressive 26% improvement over ing their pre-training on extensive text data
ChatGPT, outperforming humans in five out to offer intelligent, context-aware support for
of seven subject areas. These findings not only generating and reviewing legal documents.
[email protected]; [email protected]; [email protected];
a
DOI: 10.1201/9781003606635-111
AI Powered Legal Documentation Assistant 771
By leveraging the power of GPT-based
models, the AI-powered legal documentation
assistant excels in generating, reviewing, and
suggesting improvements for legal documents
with remarkable precision and efficiency. At
its core, this project aims to provide a com-
prehensive and user-centric platform empow-
ering legal professionals, individuals, and
businesses to streamline the creation, review,
and management of legal documents.
This paper serves as an in-depth explora-
tion of this innovative tool, delving into the AQ1:
development process, unraveling technical Please
intricacies, and dissecting core features that Figure 111.1. Project architecture. provide
source
position this platform as a game-changer in Source:
AUTHOR'S SELF GENERATED for all the
the legal landscape. Specifically, it highlights figures
intelligent document generation capabili-
ties, AI-driven assistance using ChatGPT for comprehensive collection of 35,000 Indian
context-aware suggestions and corrections, Supreme Court cases annotated with origi-
secure cloud-based document storage, and nal court decisions. Notably, a subset of this
seamless collaboration features. Additionally, corpus serves as a separate test set, enriched
the paper examines the extensive benefits this with gold standard explanations meticulously
solution offers, including significant time and curated by legal experts. Primarily designed
cost savings, heightened accuracy in legal for Court Judgment Prediction and Explana-
documentation, and enhanced accessibility. tion (CJPE), this dataset challenges the model
to deliver an automated system capable of
predicting and explaining the outcomes of
2. Methodology legal cases.
The fine-tuning process involves exposing
2.1. Introduction to the research the model to the intricacies of Indian legal
design language, refining its comprehension, and
The methodology employed in this research augmenting its proficiency in generating con-
aims to provide a systematic and comprehen- textually relevant legal documents. The diver-
sive framework for the development of an AI- sified nature of the ILDC dataset, ranging
powered legal documentation assistant. The from contracts and wills to deeds and affida-
research design combines both technical and vits, ensures that the fine-tuned model pos-
user-centric approaches to ensure the effec- sesses the versatility to generate text across a
tiveness and reliability of the system. multitude of legal documents. This fine-tun-
ing enhances the model’s adaptability, mak-
ing it adept at understanding and generating
2.2. Technological framework content in alignment with the specific nuances
2.2.1. AI-Powered document and requirements of diverse legal contexts.
generation (OpenAI GPT-3.5)
Utilize OpenAI’s GPT-3.5 as the core lan- 2.2.2. Platform architecture
guage model for document generation, fine- Design the front end using React Native for a
tuned through the ILDC dataset (Indian Legal responsive and intuitive user interface. Imple-
Documents Corpus). The ILDC dataset is a ment secure user authentication through
772 Innovations in Computational Intelligence: Trends and Technologies
tokenization or OAuth. Employ Django with Optimize the database schema for efficient
Django Rest Framework for the back-end to data storage and retrieval.
handle user requests, data management, and
AI integration. - Ensure seamless integration 2.5.2. Scalability and reliability
with external legal resources and databases
for accurate and up-to-date information. Continuously monitor and optimize AWS
RDS configurations to ensure scalability with
growing user demands. Implement automated
2.3. User interaction and services scaling mechanisms to adapt to fluctuations
2.3.1. Document review and in user activity.
customization
Develop a user-friendly interface for document 2.6. Analysis and evaluation
review, allowing users to request changes and 2.6.1. User feedback and performance
revisions. - Implement customization features metrics
enabling users to tailor generated documents
to their specific needs. Collect user feedback through surveys and
interviews to evaluate the user experience.
Define performance metrics, including docu-
2.3.2. Legal advice integration ment accuracy, response time, and system
Implement an innovative feature where GPT- reliability.
3.5, the powerful language model, automati-
cally generates initial legal advice based on 2.6.2. Statistical analysis and machine
user queries. Leverage GPT-3.5’s natural lan- learning models
guage processing capabilities to provide con-
text-aware and relevant legal insights. Employ statistical methods and machine
learning models for in-depth analysis of user
interactions and system performance. Con-
2.4. Document storage and delivery
duct iterative evaluations, refining the system
2.4.1. Secure document storage based on user feedback and emerging legal
requirements.
Implement secure storage protocols on the
server, ensuring data privacy and compliance
with legal standards. 3. Thematic Overview
This research provides a comprehensive
2.4.2. User access and download exploration of the evolutionary journey of
Allow users to access and download final- an AI-powered legal documentation assis-
ized documents through the application. - tant, delving into crucial thematic domains
Implement multi-factor authentication for essential for understanding the project’s con-
enhanced document access security. ception and implementation. It commences
by shedding light on the challenges posed
by intricate legal language, emphasizing the
2.5. Database management
pressing need for innovative solutions. Tak-
(PostgreSQL on AWS RDS) ing center stage is GPT-3.5, a recent language
2.5.1. Data storage model developed by OpenAI, distinguished as
a large multimodal model proficient in pro-
Utilize PostgreSQL on AWS RDS to store user cessing both image and text inputs, showcas-
data, preferences, and generated documents. ing human-level performance across diverse
AI Powered Legal Documentation Assistant 773
benchmarks. Its transformer-style architec- between users and legal experts for consulta-
ture, incorporating an attention mechanism tions, is thoroughly discussed.
to discern the relevance of data segments, The paper culminates by synthesizing key
enhances the understanding of word relation- insights and contributions, addressing future
ships in text [13]. directions for the Legal Documentation Assis-
A recent study employs GPT in a compre- tant, and recognizing its dynamic nature in
hensive system for abstractive summarization the ever-evolving legal technology landscape
across multi-document collections, demon- [5]. This thematic overview serves as a read-
strating comparable performance to other er’s roadmap, offering insights into major
summarization approaches [10]. OpenAI’s components and discussions in subsequent
API, residing on Microsoft Azure, serves as sections of the research paper. The develop-
a cloud-based interface, not only granting ment of this AI-powered solution aligns with
access to pre-trained AI models but also facil- contemporary technological trends and paves
itating direct engagement with GPT-3.5 in its the way for transformative advancements in
raw form. This unique capability enables pre- the legal documentation sphere [5].
cise fine-tuning of GPT-3.5, aligning it with
specific use case requirements and effectively
bridging the gap between the cloud-based AI 4. Critical Analysis
capabilities of OpenAI API and the customi- Introduced to widespread acclaim in March
zation potential inherent in GPT-3.5’s bare 2023, GPT-3.5 represented a notable leap
metal form [14]. forward from its predecessor, GPT-3.5, which
Technologically, the backend thrives on had hitherto stood as the pinnacle of Ope-
Django, a Python-based web framework, nAI’s language models, catalyzing the inaugu-
while the frontend embraces React Native, ral release of ChatGPT. Despite the attention
a cross-platform framework catering to IOS garnered by the improvements in GPT-3.5,
and Android platforms. The intricate techno- a closer examination reveals significant
logical framework is meticulously explored,
highlighting the integration of React Native
for the frontend and Django with Django
Rest Framework for the backend. The paper
accentuates the fine-tuning process of GPT-
3.5 for legal document comprehension,
underscoring the project’s reliance on cutting-
edge technological advancements [5].
A user-centric design ethos permeates the
development, crafting an intuitive interface
with React Native, incorporating inclusive
user input mechanisms, and securing authen-
tication processes. The integration with legal
resources is elucidated, detailing how the
backend establishes connections with exter-
nal legal databases to ensure document accu-
racy and compliance [5].
Further exploration delves into the docu-
ment review process, empowering users to
request revisions and changes. The integration Figure 111.2. Workflow flowchart.
of legal advice features, fostering connections Source:
AUTHOR'S SELF GENERATED
774 Innovations in Computational Intelligence: Trends and Technologies
reservations about its professed reasoning data, may lack awareness of the latest legal
capabilities [16]. updates, making real-time information
Large language models (LLMs), epito- updates a significant challenge.
mized by GPT, operate on a fundamental While GPT models offer the flexibility
concept: training neural architectures on of fine-tuning, accommodating diverse user-
extensive datasets to create potent language specific preferences and customization needs
models. Many LLMs, rooted in the trans- proves challenging. Individual users may have
former architecture, showcase versatility varying requirements for document genera-
across various natural language processing tion and review, and the model may encoun-
(NLP) tasks, surpassing the state of the art in ter difficulty adapting seamlessly to these
numerous challenges [8]. individual preferences.
GPT-3, in showcasing its capabilities, dis- Furthermore, the inclusion of sensitive and
played discernible biases linked to various confidential information in legal documents
demographic factors, including race, gen- raises valid concerns about data privacy and
der, and religion. These biases, evident in the security. The use of a language model neces-
model’s responses and outputs, inadvertently sitates rigorous measures to prevent inad-
perpetuate societal prejudices and stereotypes vertent exposure of confidential information,
related to racial identities, gender roles, and adding a critical layer of complexity [17].
religious affiliations [16]. A recent examina-
tion assessing the ethical aspects of GPT-3.5
revealed a substantial correlation between 5. Synthesis and Implications
human and AI responses. However, notable The development of LexGPT and Law GPT
and systematic disparities indicate existing models signifies a groundbreaking advance-
limitations in GPT-3.5’s accurate modeling of ment in the field of legal AI [3][10]. It is cru-
human decision-making processes [9]. cial to highlight that while LexGPT may not
In the context of our specific use case, GPT be tailor-made for the nuances of the Indian
models confront limitations arising from a legal system, it lays the foundation for future
lack of domain-specific knowledge. Legal applications, sparking considerations for
documents necessitate a profound under- potential refinements to align with regional
standing of intricate legal concepts, and the legal frameworks. This approach finds paral-
model may struggle to interpret and gener- lels in initiatives like Cicero, which fine-tuned
ate content accurately without specialized a transformer specifically for the Italian legal
knowledge of legal contexts. system, and Lawformer, a Longformer-based
Furthermore, inherent biases in legal language model extensively pre-trained on
texts present an additional challenge. If the criminal and civil case documents, showcas-
training data includes biased legal docu- ing significant progress in LegalAI [4][6].
ments, the model may generate summaries Additionally, a separate research paper
that are skewed, impacting the impartiality details an individual’s creation of a model
and equity of the content it generates. The using BERT, customized for the specific intri-
notorious ambiguity of legal language poses cacies of the Indian legal system [7]. In con-
a significant hurdle for GPT models, making trast, our innovative approach capitalizes on
accurate interpretation of ambiguous legal GPT-3.5, already equipped with extensive
terms or phrases challenging and potentially internet-sourced data, including Indian legal
leading to inaccuracies in document summa- databases. The advantage lies in leveraging
ries or advice. pre-existing training, requiring only fine-tun-
The dynamic nature of legal landscapes, ing for our specific use case.
with frequent changes in laws, poses another Significantly, our project goes beyond
challenge. GPT models, reliant on pre-existing document drafting; it empowers users to edit
AI Powered Legal Documentation Assistant 775
existing documents, seek simplified summa- pioneering solution poised to transform the
ries, and comprehend complex legal language. realm of legal documentation. Through the
This inclusive feature directly addresses the seamless integration of Django, React Native,
challenge faced by individuals lacking a and the cutting-edge GPT-3.5 via OpenAI
legal background, providing assistance in API technologies, this project stands as a
understanding and navigating intricate legal testament to the capacity of artificial intel-
documents. In doing so, it ensures informed ligence to revolutionize intricate processes.
decision-making in legal matters. By providing a user-friendly, accurate, and
efficient tool, it not only empowers legal
professionals but also extends its benefits to
6. Recommendations for Future the broader public, facilitating a more acces-
Research sible approach to navigating India’s legal
In the forthcoming phases of development, landscape. The amalgamation of these tech-
our focus will center on pivotal aspects to nologies not only signifies a substantial leap
enhance the efficacy and inclusivity of our forward in the efficiency of legal services but
AI-powered legal documentation assis- also sets a new standard for legal assistance in
tant. Firstly, the integration feature will be the digital age. The transformative potential
expanded to facilitate secure collaboration of this project is poised to redefine the legal
and document sharing among multiple par- documentation landscape in India, marking
ties, fostering teamwork and streamlining a significant stride towards enhanced acces-
document management. Secondly, we aspire sibility and efficiency in legal processes.
to broaden the scope of legal document gen-
eration, encompassing a diverse array of doc- References
ument types to address a wider spectrum of
legal needs and requirements. Additionally, the [1] Katz DM. GPT-3.5 passes the bar exam.
Published in SSRN; 2023.
incorporation of multilingual support will be
[2] Dalal S, Singhal A, and Lall B. LexRank and
pursued, ensuring accessibility for users from Pegasus transformer for summarization of
various linguistic backgrounds. Our commit- legal documents. Published in SpringerLink;
ment extends to seamless integration with 1970.
E-signature platforms, thereby empowering [3] Lee J-S. LexGPT 0.1: Pre-trained GPT-J
users to electronically sign the generated legal models with pile of law. Published in arXiv.
documents and establishing a comprehensive org; 2023.
end-to-end solution for document creation [4] Xiao C, et al. Lawformer: A pre-trained lan-
and execution. Furthermore, the exploration guage model for Chinese legal long docu-
of advanced NLP techniques is on the agenda ments; 2021.
to refine document generation accuracy and [5] Arora J. Artificial intelligence as a legal
offer nuanced customization options based research assistant. Published in Academia.
on user input. These future endeavors under- edu; 2022.
[6] Luzi FD, et al. Cicero: An AI-based writing
score our dedication to continuous improve-
assistant for legal users. Published in Spring-
ment and the expansion of our AI-powered erLink; 1970.
solution to meet evolving user needs in the [7] Jin X, and Wang Y. Understand legal docu-
legal landscape. ments with contextualized large language
models. Published in arXiv.org; 2023.
[8] Liga D, et al. Fine-tuning GPT-3 for legal
7. Conclusion rule classification. Published in Comput
In this paper, we introduced the AI-powered Law Secur Rev; 2023.
legal documentation assistant for India as a
776 Innovations in Computational Intelligence: Trends and Technologies
[9] Almeida GFCF, et al. Exploring the psychol- [14] Sofiia. What is an OpenAI API, and how to
ogy of GPT-3.5’s moral and legal reasoning. use it? Published in Addepto; 2023.
Published in arXiv.org; 2023. [15] Burman A, and Bradford E. Building an opti-
[10] Nguyen HT. A brief report on LawGPT 1.0: mized algorithm that provides summaries of
A virtual legal assistant based on GPT-3. legal documents. Published in ResearchGate;
Published in arXiv.org; 2023. 2022.
[11] Liu S, and Healey CG. Abstractive summa- [16] Tamkin. Understanding the capabilities,
rization of large document collections using limitations, and societal impact of large lan-
GPT. Published in arXiv.org; 2023. guage models. Published in arXiv.org; 2021.
[12] Gupta Y. Chat GPT and GPT-3 detailed [17] Sun AY. Does fine-tuning GPT-3 with the
architecture study: deep NLP HORSE. Pub- OpenAI API leak personally-identifiable
lished in Medium; 2023. information? Published in arXiv.org; 2023.
[13] Wagh A. What’s new in GPT-3.5: An over- [18] Liu Y. Roberta: A robustly optimized Bert
view of the GPT-3.5 architecture and capa- pretraining approach. Published in arXiv.
bilities of next-generation AI. Published in org; 2019.
Medium; 2023.
112 Ideas to boost fitness activities and
assist in keeping fit
Neha Chauhana, Vasu Guptab, Harshita Bajpaic, and Kim Guptad
Department of Computer Science and Engineering, Babu Banarasi Das Institute of Technology and
Management, Lucknow, Uttar Pradesh, India
Abstract: This study highlights the important role of a user-centered design approach that aligns app goals with
user needs and incorporates feedback through wire-frames and prototypes. The technology stack selection pro-
cess has been carefully considered with a focus on cross-platform development and scalability. Collaboration
with fitness experts drives the development of evidence-based training plans and nutritional algorithms,
enhanced by the integration of wearable technology to improve user engagement. Gamification elements and
motivational features are used strategically, complemented by a focus on security, privacy, iterative develop-
ment, and a comprehensive launch and marketing strategy. This comprehensive methodology ensures the devel-
opment of dynamic, user-friendly fitness apps that thrive in the ever-evolving digital environment.
Keywords: Fitness, Pseudo Code, Personalization, Cross-platform Development, Web development, Accessibil-
ity, Customer Participation
1. Introduction data analysis techniques such as c-svc and
v-svc to improve the Android development
In recent years, the landscape of fitness appli- process and optimize fitness results [1].
cations has undergone remarkable changes Building on this, a 2018 research paper
due to rapid advances in information tech- highlights the integrative nature of physical
nology. The convergence of mobile internet, activity, recognizing its positive impact on
4G communications, and data technology social and mental well-being. However, certain
has made smart-phones an essential tool for population groups are also considered, such as:
physical activity. The growing popularity of Some people, such as the elderly and postpar-
fitness software on smart-phones, noted in tum women, face difficulties when exercising
a 2023 research report, highlighted the sim- regularly at the gym or outdoors. This article
plicity, effectiveness, and broad application explores how technology can bridge this gap
potential of these tools [1,2]. by promoting exercise at home through the
A 2023 paper identifies flaws in the exist- motivation to achieve better strength and bal-
ing fitness app market, citing issues such as ance, especially in older adults. Although the
boring content, confusing logic, and shoddy potential is highlighted, the existing limitations
production. These limitations are due to a and research gaps in IT-based home training
lack of user-centered design and research. In solutions are also highlighted [6].
response, this paper proposes a solution based Fast forward to 2020, and this year’s
on an analysis of the needs of fitness apps. research paper shows that we are entering an
The goal is to redesign the app’s structural era where applications have become an essen-
model and functional modules and integrate tial part of daily life. The evolution from apps
[email protected]; [email protected]; [email protected];
a
DOI: 10.1201/9781003606635-112
778 Innovations in Computational Intelligence: Trends and Technologies
for email and basic utilities to a vast world process. This information is important to
of mobile gaming, GPS services, ticket book- understand physical condition of user and
ing, and more is remarkable. In this context, make decision based on it.
fitness apps have emerged as a solution for 2. Data Processing: The system then applies
people with busy lives, providing convenient algorithm to process the user to give the
access to health and wellness resources at any best results.
time [2]. 3. Condition Checking: The condition is
This comprehensive overview reflects then monitored whether user is perfectly
the dynamic evolution of fitness applica- normal physically or not and then based
tions, from addressing structural and con- on that data is passed forward.
tent challenges to leveraging technology for 4. Workout Plan and Nutrition Algorithm
comprehensive and accessible home train- Development: Collaborate with fitness
ing solutions. This journey summarizes the experts and nutritionists to design evi-
app’s evolution from simple functionality to dence-based training plans and nutritional
an essential companion to a healthier, more algorithms. Implement machine learning
active lifestyle [3]. or data analysis techniques to personalize
recommendations based on your profile,
preferences, and progress.
2. Methodology 5. Conditional exercise recommendation:
The methodology for developing a versatile The experts will then assign exercises
fitness app involves a systematic approach based on users conditions like if user
to meeting user needs and market require- works or have less time, then fast and
ments. The first phase will focus on com- short duration exercises will be recom-
prehensive market research and user needs mended and vice-versa.
analysis to understand trends, competitors, 6. Security and Privacy Measures: Prioritize
user demographics. The app’s goals and func- the security of user data by implementing
tionality are then clearly defined and aligned robust encryption protocols and ensuring
with user expectations. Choosing the right privacy compliance. Clearly communicate
technology stack ensures a seamless cross- your app’s privacy policy to your users
platform development process. User-centered and obtain their consent for data collec-
design principles guide the creation of intui- tion and processing.
tive user interfaces based on feedback gath- 7. Iterative Development and Testing: We
ered through an iterative design process. use an agile development approach with
Collaborate with fitness experts to develop iterative releases to gather user feedback
personalized, evidence-based training plans and make continuous improvements. We
and nutrition algorithms through machine perform thorough testing, including func-
learning and data analysis. Integration with tional, usability, and security testing to
wearable technology increases user engage- proactively identify and resolve issues.
ment and health monitoring. Gamification
elements and motivational features are inte-
grated to encourage user engagement, while 3. Thematic Analysis
robust security measures and transparent pri- The thematic analysis of the evolution of fit-
vacy policies prioritize the protection of user ness applications unravels a narrative woven
data. around a transformative shift from a tech-
nology-centric approach to a user-centric
1. Data Collection: The System collects design philosophy. The initial scrutiny in
the information about user for further 2019 pinpointed shortcomings in existing
Ideas to boost fitness activities and assist in keeping fit 779
fitness app market ranging from mundane
content to confusing logic and rough pro-
duction prompting a call for a fundamen-
tal reimagining of app development. This
reorientation involved a meticulous explora-
tion of user demands, steering the trajectory
towards a more holistic and engaging fitness
app experience. Simultaneously, the research
paper underscored the inclusive nature of
physical activity while recognizing the chal-
lenges faced by specific demographics, laying
the groundwork for technology to facilitate
home-based training solutions. This theme of
inclusivity and accessibility echoes in the per-
spective, where fitness apps are positioned as
indispensable tools in the daily lives of indi-
viduals, offering flexibility and convenience
to a diverse user base. Overall, this thematic
AQ1: analysis represents an effort characterized by
Please user-centricity, inclusivity, and a deep com-
provide
citation mitment to technology’s evolving role in
source for Figure 112.1. Methodological process flow democratizing access to fitness resources.
the figure chart.
and table
Source: AUTHOR'S SELF GENERATED
4. Critical Analysis on Previous Works
AUTHOR'S SELF GENERATED
Table 112.1. Critical analysis comparison
S.NO Paper Analysis
1 Francisco Martín,Jerónimo García According to the paper’s analysis of fitness
ernández,corresponding author, Manel applications by age and gender, Quadrant II
Valcarce-Torrente, Ainara Bernal- (high importance, high performance) aspects
García, Pablo Gálvez-Ruiz, and are prioritized by older users with higher
Salvador Angosto-Sánchez.“Importance- expectations. Items in Quadrant III (low
performance analysis in fitness apps. A importance, low performance) have little
study from the viewpoint of gender and priority, but these components help users feel
age “.Published online 2023 Aug 14. doi: satisfied and ought to be kept.
10.3389/fpubh.2023.1226888
2 Angosto, S., García-Fernández, J. and This study brings up to current the research on
Grimaldi-Puyana, M. A systematic the willingness to use and embrace fitness apps
review of intention to use fitness apps that has been released since 2020. It talks about
(2020–2023) the reasons why people want to utilize these
Humanit Soc Sci Commun 10, 512 kinds of technology and emphasis how popular
(2023). https://doi.org/10.1057/ fitness applications are becoming in the fitness
s41599-023-02011-3 industry.
(continued)
780 Innovations in Computational Intelligence: Trends and Technologies
Table 112.1. continued
S.NO Paper Analysis
3 Yanlong Guo, Xueqing Ma, Denghang With a focus on perceived ease of use, content
Chen“Factors Influencing Use of Fitness quality, technological grade, interaction efficacy,
Apps by Adults under Influence of system compatibility, economical and privacy
COVID-19” 2022 Nov 22. doi: 10.3390/ costs, this article investigates the factors
ijerph192315460 influencing adult use of fitness applications
during COVID-19. It emphasises how crucial
it is for makers of health apps to customise
tactics and material to each user’s particular
characteristics in order to take a more
individualised and methodical approach.
4 Manel Valcarce-Torrente, Vicente This study looks into how exercise apps affect
Javaloyes, Leonor Gallardo,Jerónimo athletes’ practices, levels of contentment, and
García-Fernández,“Influence of Fitness plans to stick around fitness centres. It offers
Apps on Sports Habits, Satisfaction, and suggestions for a thorough implementation
Intentions to Stay in Fitness Center Users: strategy for fitness facilities that takes into
An Experimental Study” 2021 Oct 2. doi: account employees, patrons, and their training
10.3390/ijerph181910393 regimens.
5 Yali Liua and Maria Avello”Status of the This work is a bibliometric study on fitness
research in fitness apps: A bibliometric applications, highlighting their interdisciplinary
analysis” 2020 Sep 23. doi: 10.1016/j. nature across various journals. Using the
tele.2020.101506 keyword “fitness app,” the study focuses on
scientific research for personal care. It offers
an overview of the research landscape, years of
publication, and research methodologies.
Source:
5. Recommendation for Future responsiveness. Create feedback loops to
refine features, ensuring continuous improve-
Work
ment and heightened user satisfaction.
It is important to fill the current gaps in fitness B. Community Growth and Engagement:
apps with future developments in mind. To Foster a sense of community through
improve the user experience, prioritize incor- interactive features and forums. Encour-
porating comprehensive mental health and age user-generated content and challenges
stress management features, recognizing the to boost engagement. Integrate social shar-
important role mental health plays in overall ing functionalities, enhancing connections
fitness. Consider integrating customized meal among users and sustaining motivation.
plans and real-time nutritional advice to pro- C. Integration of Emerging Technologies:
vide users with a holistic approach to health. Explore virtual or augmented reality fea-
Future apps should explore innovative ways tures for immersive experiences. Consider
to personalize the user’s workout routine AI-driven insights for personalized fitness
based on her personal preferences and ensure recommendations. Stay abreast of tech-
engagement and long-term commitment. nological trends to position the app as an
innovative frontrunner.
A. Enhanced User Experience: D. User-Friendly Interface Upgrades:
Innovate navigation for an intuitive user Simplify app navigation with clear and
journey and optimize app speed and intuitive menus. Optimize UI elements
Ideas to boost fitness activities and assist in keeping fit 781
for various devices, focusing on acces- strategy has the power to completely reshape
sibility features for inclusivity. Prioritize the fitness app landscape, surpassing con-
a seamless and enjoyable user interface sumer expectations and creating entirely new
experience. standards for the sector.
E. Making More Platforms Compatible:
Expand compatibility to diverse devices
and operating systems. Ensure seamless References
AQ2:
integration with popular fitness gadgets. [1] Francisco Martín, Jerónimo García
Please
Develop a web-based platform to broaden ernández,corresponding author, Manel Val- note that
accessibility beyond mobile devices. carce-Torrente, Ainara Bernal-García, Pablo the cross-
Gálvez-Ruiz, and Salvador Angosto-Sánchez reference
F. Advanced Personalization Algorithms: 4,5,7-14
Implement machine learning for in-depth “Importance-performance analysis in fitness is not
apps. A study from the viewpoint of gender listed
user behavior analysis. Tailor workout
and age “.Published online 2023 Aug 14. in-text.
plans, content, and challenges based on doi: 10.3389/fpubh.2023.1226888
individual data. Continuously refine per- [2] Angosto, S., García-Fernández, J. and Gri-
sonalization features to adapt to evolving maldi-Puyana, M. A systematic review
user needs and goals. of intention to use fitness apps (2020–
2023). Humanit Soc Sci Commun 10,
512 (2023). https://doi.org/10.1057/
6. Conclusion s41599-023-02011-3
The ever-changing landscape of fitness appli- [3] Yanlong Guo, Xueqing Ma, Denghang
cations presents fresh chances for improve- Chen“Factors Influencing Use of Fit-
ment and innovation. The current state of ness Apps by Adults under Influence of
the fitness app ecosystem needs to be filled COVID-19” 2022 Nov 22. doi: 10.3390/
by incorporating complete wellness features, ijerph192315460
[4] Manel Valcarce-Torrente, Vicente Jav-
providing individualised dietary guidance,
aloyes, Leonor Gallardo,Jerónimo García-
encouraging inclusionary programmes, and Fernández,“Influence of Fitness Apps on
utilising state-of-the-art technology. This Sports Habits, Satisfaction, and Intentions
strategy is revolutionising the fitness industry to Stay in Fitness Center Users: An Experi-
by promoting a comprehensive, user-centered mental Study” 2021 Oct 2. doi: 10.3390/
fitness experience. ijerph181910393
Future research may focus on improving [5] Yali Liua and Maria Avello”Status of
AI algorithms for more accurate personalisa- the research in fitness apps: A bibliomet-
tion, investigating virtual reality simulations ric analysis” 2020 Sep 23. doi: 10.1016/j.
for engaging in immersive fitness activities, tele.2020.101506
and integrating biometric data for continu- [6] Sakitha Anna Joseph, Reshma Raj
ous health monitoring. Stress management K., Sony Vijayan, “User’s Perspec-
tools and mindfulness practices that prior- tive about Mobile Fitness Applications”
8760038620/2020©BEIESP DOI.35940/
itize mental health could further strengthen
ijrte.F8760.038620, March 2020.
the integrative strategy. [7] Adria Muntaner-Mas, Antonio Martinez-
As developers, we are dedicated to more Nicolas, Carl J. Lavie, Steven N. Blair,
than just promoting physical health. Fitness Robert Ross, Ross Arena, and Francisco B.
apps have a major positive impact on peo- Ortega (2019). A Systematic Review of Fit-
ple’s general well-being and can help create ness Apps and Their Potential Clinical and
a society that is healthier and more cohe- Sports Utility for Objective and Remote
sive by addressing user needs and embracing Assessment of Cardiorespiratory Fitness.
technological advancements. This innovative
782 Innovations in Computational Intelligence: Trends and Technologies
Sports Medicine 2019, 49(4), 587-600. Motivation, Functionality, and Features.”
doi:10.1007/s40279-019-01084-y Journal of Medical Internet Research, 19(6),
[8] Jifeng Liang, “Research on Fitness APP” e210. doi:10.2196/jmir.6839
Xi’an Fanyi University Xi’an 710105 China [13] Direito, A., Carraça, E., Rawstorn, J., Whit-
(ICISS 2019). taker, R., Maddison, R. (2017). “mHealth
[9] Garcia, M., and Patel, R. “User-Centric Technologies to Influence Physical Activity
Design and Gamification in Fitness Apps: A and Sedentary Behaviors: Behavior Change
Comparative Analysis.” International Jour- Techniques, Systematic Review and Meta-
nal of Mobile Health Applications, 7(2), Analysis of Randomized Controlled Trials.”
87-102, 2019. Annals of Behavioral Medicine, 51(2), 226-
[10] Smith, A. “The Impact of Wearable Fitness 239. doi:10.1007/s12160-016-9836-7.
Technology on Physical Activity Habits: A [14] Laranjo, L., Arguel, A., Neves, A. L., Gal-
Longitudinal Study.” Journal of Health and lagher, A. M., Kaplan, R., Mortimer, N., and
Technology, 12(3), 123-136, 2018. Lau, A. Y. (2015). “The influence of social
[11] Johnson, L., and Turner, S. (2018). “Explor- networking sites on health behavior change:
ing the Role of Social Support in Sustaining a systematic review and meta-analysis.”
Engagement with Fitness Apps: A Qualita- Journal of the American Medical Informatics
tive Study.” Journal of Behavioral Medicine, Association, 22(1), 243-256. doi:10.1136/
15(4), 245-260. amiajnl-2014-002841
[12] Chen, Y., Zhuang, S., and Wang, C. (2017).
“A Comparative Analysis of Fitness Apps:
113 Using existing CCTV network for
crowd management, crime prevention
and work monitoring: A novel
approach for object detection
Manvi Agrawal
Computer Science and Engineering, Babu Banarasi Das Institute of Technology and Management,
Lucknow, India
Abstract: This paper presents a novel approach for the creation of an intelligent CCTV surveillance system,
leveraging advanced machine learning techniques, specifically the you only look once object detection algo-
rithm. The system aims to conduct real-time analysis of CCTV footage from Indian railways, with a focus on
enhancing crowd management, preventing criminal activities, and monitoring railway operations to ensure the
safety and security of passengers and commuters. The overarching goal is to improve the efficiency of day-to-
day railway operations. Through the integration of cutting-edge technology, this research contributes to the
advancement of safety measures and operational effectiveness in the context of Indian railways.
Keywords: Yolo algorithm, object detection, real-time analysis, supervised learning, SMART surveillance,
crowd density estimation
1. Introduction Indian Railways using machine learning (ML)
techniques. The system uses real time surveil-
The Indian Railways are the lifeline of the lance footage from CCTVs installed at the
country which is indicated by the billions of stations i.e. the existing CCTV infrastructure
passengers it carries throughout the day. To which is seen functioning commonly at major
manage such a volume of passengers on a Railway stations currently, and stores them
day-to-day basis in real-time by simple man- into a centralized cloud database for process-
ual monitoring is an extremely difficult task. ing and further analytics. This includes imple-
This includes monitoring not only the effec- menting computer vision for converting the
tive functioning of the railway operations video into individual frames using OpenCV
but also crowd management, crime preven- library and storing them with timestamps
tion, and work monitoring simultaneously by in the provided database and using the you
the authorities. Smart automation is rapidly only look once (YOLO) object detection
revolutionizing the whole world. Using ml algorithm for training the model in accurate
to transform railway operations is one such object detection in the frames generated using
smart solution proposed by the authors in already stored historical data for crowd den-
this paper. sity estimation, crime detection, work moni-
Essentially, this paper proposes to develop toring scenarios. Additionally, a user-friendly
an automated surveillance system for the alerting system or interface for the assistance
[email protected]
DOI: 10.1201/9781003606635-113
784 Innovations in Computational Intelligence: Trends and Technologies
of the authorities will timely prompt them in The outlined steps ensure a comprehensive
case of any anomaly detected or emergency approach to address the challenges associated
situations triggered like overcrowding at sta- with managing the vast volume of passengers
tions, detected by the trained ML model via and enhancing the overall operational effi-
CCTV real-time footage. ciency of the railway system.
This proposed solution not only reduces
the load on authorities by helping them focus 2.1. Data collection
their valuable resources like time, energy,
and money on more critical tasks but also CCTV footage acquisition: Real-time surveil-
enhances their efficiency by utilizing the auto- lance footage from existing CCTV infrastruc-
mation capabilities of an ML-based smart ture at major railway stations is collected.
system to their advantage. This also aids in This footage is essential for training and vali-
the real-time accurate detection of a num- dating the ML model.
ber of situations simultaneously via a single, Historical data compilation: Previously
simplified, easy-to-use, customizable, and recorded data is gathered to create a diverse
efficient system developed keeping in mind dataset. This historical data serves as a foun-
the problems faced by the Indian Railways dation for training the YOLO object detection
specifically. algorithm and refining the model’s accuracy.
2.2. Data processing
2. Methodology
Frame extraction: The OpenCV library is uti-
The methodology employed in this research lized to convert the video footage into indi-
aims to systematically design and imple- vidual frames. Each frame is timestamped
ment an automated surveillance system for and stored in a centralized cloud database for
the Indian Railways using ML techniques. further processing.
Database management: The cloud data-
base is organized to facilitate efficient storage
of and retrieval of video frames. Timestamped
frames are stored systematically to enable
seamless analytics.
2.3. Algorithm implementation
Object detection and training: The YOLO
object detection algorithm is implemented
for the accurate identification of objects in
the video frames. The model is trained using
historical data to optimize its performance
in crowd density estimation, crime detection,
AQ1: and work monitoring scenarios.
Please
provide
figure
Figure 113.1: 2.4. Analytics and alerting system
in-text
citation Caption: Crowd density estimation: The trained ML
figure
caption
"Architecture of model is employed for real-time crowd density
and Figure 113.1.
AI-powered estimation at railway stations, allowing for
source Surveillance System proactive measures to manage overcrowding.
Source: for Indian Railways"
AUTHOR'S SELF GENERATED
Using existing CCTV network for crowd management, crime prevention and work monitoring 785
Crime detection: Anomalies in video cameras at train stations. This system uti-
frames indicative of potential criminal activi- lizes real-time footage stored in a centralized
ties are identified using the ML model, trig- cloud database. It employs computer vision
gering alerts for prompt intervention by via the OpenCV library and the YOLO detec-
authorities. tion algorithm to facilitate accurate object
Work monitoring: The system moni- detection for crowd density estimation, crime
tors railway operations ensuring efficient detection, and work monitoring, informed
workflow and timely identification of any by historical data. This paper implements the
irregularities. integration of a user-friendly alerting system,
ensuring timely responses to anomalies or
2.5. User-friendly interface emergencies detected by the ML model. Apart
from lessening the burden on authorities, the
A user-friendly interface is designed to pro- proposed solution helps in resource optimiza-
vide authorities with timely alerts in case tion and efficiency, by allowing authorities to
of anomalies or emergency situations. This redirect their valuable resources toward more
interface ensures ease of use, customization, critical tasks. The outcome is a single, cus-
and efficient response to detected issues. tomizable, and efficient system, crafted spe-
cifically to address the major challenges faced
2.6. Validation and testing by the Indian Railways, ultimately enhancing
real-time authority’s awareness about what’s
The performance of the ML model is rigor-
happening in real-time and the operational
ously validated using both real-time and his-
efficiency.
torical data to ensure accurate and reliable
results.
4. Critical Analysis
2.7. Optimization and refinement The methodology proposed in this research
Based on feedback and real-world usage, the comes with its own unique challenges which
system undergoes iterative optimization to might pose as limitations if not dealt with in
enhance its accuracy, efficiency, and adapt- detail. Some of these bottlenecks or critical
ability to evolving scenario’s. examination factors crucial to be handled for
the successful implementation of this pro-
posed system are:
3. Thematic Analysis
This research aims to revolutionize Indian Acknowledgement
Railways by integrating ML in existing
CCTV surveillance systems at the stations. The authors would like to express their grati-
Indian Railways is vital for the country, car- tude for the invaluable and consistent guid-
rying a huge number of passengers daily. ance and support provided by Mr. Saurabh
Managing everything manually, like keep- Kumar Jain, the project guide, throughout the
ing operations smooth, managing crowds, development of this project. The authors also
preventing crimes, and overseeing work in take this opportunity to extend their appre-
real-time, which is very extensive and hard, ciation to all the mentors at BBDITM for
is this paper’s objective. The paper suggests encouraging them to work on this extensive
using ML, a smart technology, to change yet equally exciting project. The knowledge
how railways work. The idea is to create a and experience thus gained were invaluable
smart surveillance system using the existing assets in drafting this paper.
786 Innovations in Computational Intelligence: Trends and Technologies
AQ2: References [11] Pawar K, and Attar V. Deep learning
Please
approaches for video-based anomalous
provide
[1] Chen Y, Xu H, Zhang X, Gao P, Xu Z, and activity detection. 2018;May.
reference Huang X. An object detection method for [12] Haque MF, Lim HY, and Kang DS. Real time
citations bayberry trees based on an improved YOLO object detection based on YOLO with fea-
algorithm. Appl Sci. 2023;13(8):4956. ture filter bank. J KIIT. 2019;May.
[2] Kulkarni U, Naregal K, Farande V, Guttigoli [13] Khan K, Albattah W, Khan RU, Qamar
S, Angadi A, and Ujwane R. An object detec- AM, and Nayab D. Advances and trends
tion approach for automated detection of in real-time visual crowd analysis. Sensors.
groove line in tube yoke. ITM Web Conf. 2020;20(18):5073.
2023;53:01007. [14] Zhao L. Object detection algorithm
[3] Aldayri A, and Albattah W. Taxonomy of based on improved YOLOv3. Electronics.
anomaly detection techniques in crowd 2020;9(3):537.
scenes. Sensors. 2022;22(16):6080. [15] Gul MA, Yousaf MH, Nawaz S, Ur
[4] Wilson D, Manusankar C, and Prathibha P Rehman Z, and Kim H. Patient monitor-
H. Analytical study on object detection using ing by abnormal human activity recognition
the YOLO algorithm. Vol 7; 2022. based on CNN architecture. Electronics.
[5] Ren J, and Wang Y. Overview of object detec- 2020;9(12):1993.
tion algorithms using convolutional neural [16] Alex A, Sahu A, Tanwar A, and Rathi N.
networks. J Comput Commun. 2022;10(1). Abnormal event detection by machine
[6] Rezaee K, Rezakhani SM, Khosravi MR, and vision using deep learning. University of
Moghimi MA. A survey on deep learning- Hyderabad; 2020.
based real-time crowd anomaly detection [17] Khan A, Shah JA, Kadir K, Albattah W, and
for secure distributed video surveillance. Vol Khan F. Crowd monitoring and localization
2021;Article ID 9975700. using deep convolutional neural network: A
[7] Narejo S, Pandey B, Vargas DE, Rodriguez review. Appl Sci. 2020;10(14):4781.
C, and Anjum R. Weapon detection using [18] Viswanatha V, Chandana RK, and
YOLO V3 for smart surveillance system. Ramachandra AC. Real-time object detec-
[8] Ajala J. Object detection and recognition tion system with YOLO and CNN models:
using YOLO: Detect and recognize URL(s) A review. 2022.
in an image scene. St. Cloud State Univer- [19] Ajala J. Object detection and recognition
sity; 2021. using YOLO: Detect and recognize URL(s)
[9] Byrne J, and Marx G. Technological inno- in an image scene. St. Cloud State Univer-
vations in crime prevention and policing: A sity; 2021.
review of the research on implementation [20] Sharma P. A practical guide to object detec-
and impact. 2011. tion using the popular YOLO framework:
[10] Han C, Wang C, and others. YOLO-based Part III (with Python codes). Updated August
adaptive window two-stream convolutional 26, 2021.
neural network for video classification.
2017.
You might also like
- THEORY FILE - Internet of Things (5th Sem) .No ratings yetTHEORY FILE - Internet of Things (5th Sem) .13 pages
- Data Warehousing & Data Mining Unit-2 Notes100% (1)Data Warehousing & Data Mining Unit-2 Notes36 pages
- ds4015 Big Data Analytics Vignesh K NotesNo ratings yetds4015 Big Data Analytics Vignesh K Notes146 pages
- Predicting Hourly Boarding Demand of Bus Passengers 3.6.2100% (1)Predicting Hourly Boarding Demand of Bus Passengers 3.6.281 pages
- Car Parking Space Detection Using Yolo: A Project Report Submitted byNo ratings yetCar Parking Space Detection Using Yolo: A Project Report Submitted by60 pages
- SS ZG515 - Data Warehousing: BITS PilaniNo ratings yetSS ZG515 - Data Warehousing: BITS Pilani44 pages
- Final B.tech. Internship Report Sample Format-3100% (1)Final B.tech. Internship Report Sample Format-313 pages
- 1 L1 - Introduction To Data Structure and Algorithms100% (1)1 L1 - Introduction To Data Structure and Algorithms96 pages
- Ch-4 Processor Memory Modeling Using Queuing Theory100% (2)Ch-4 Processor Memory Modeling Using Queuing Theory19 pages
- Types of Assembly Lines:: 1.single Model Line 2.batch Model Line 3.mixed Model LineNo ratings yetTypes of Assembly Lines:: 1.single Model Line 2.batch Model Line 3.mixed Model Line10 pages
- Project Report "E-Commerce Recommendation"No ratings yetProject Report "E-Commerce Recommendation"20 pages
- R22 - IT - Python Programming Lab ManualNo ratings yetR22 - IT - Python Programming Lab Manual96 pages
- Digital Notes: (Department of Computer Applications)No ratings yetDigital Notes: (Department of Computer Applications)14 pages
- CCA3002 - FOG-AND-EDGE-COMPUTING - LT - 1.0 - 34 - Fog and Edge ComputingNo ratings yetCCA3002 - FOG-AND-EDGE-COMPUTING - LT - 1.0 - 34 - Fog and Edge Computing3 pages
- 25th August MCA New First Year Syllabus 2020No ratings yet25th August MCA New First Year Syllabus 202024 pages
- Smart India Hackathon 2019: Problem Statements-Software100% (1)Smart India Hackathon 2019: Problem Statements-Software8 pages
- 2.1: Advanced Processor Technology: Qn:Explain Design Space of Processor?No ratings yet2.1: Advanced Processor Technology: Qn:Explain Design Space of Processor?29 pages
- Tybscit - Business Intelligence-Unit 02 - 2021No ratings yetTybscit - Business Intelligence-Unit 02 - 202129 pages
- Te Aids - (Elective-I) Human Computer InterfaceNo ratings yetTe Aids - (Elective-I) Human Computer Interface2 pages
- Synopsis - Note Sharing Application Using DjangoNo ratings yetSynopsis - Note Sharing Application Using Django12 pages
- SMART WASTE MANAGEMENT SYSTEM USING IoTNo ratings yetSMART WASTE MANAGEMENT SYSTEM USING IoT46 pages
- Information Technology - MBA 1st Year 1st SemNo ratings yetInformation Technology - MBA 1st Year 1st Sem9 pages
- Dynamic Auto Selection and Auto Tuning of Machine Learning Models For Cloud Network Analytics SynopsisNo ratings yetDynamic Auto Selection and Auto Tuning of Machine Learning Models For Cloud Network Analytics Synopsis9 pages
- Introduction To AI & ML QUESTION BANK MODULEWISENo ratings yetIntroduction To AI & ML QUESTION BANK MODULEWISE3 pages
- InformationSystemsSecurityByNinaGodbole Good ReadNo ratings yetInformationSystemsSecurityByNinaGodbole Good Read4 pages
- Cyber Security: AICTE Sponsored One Week Online Faculty Development Programme OnNo ratings yetCyber Security: AICTE Sponsored One Week Online Faculty Development Programme On2 pages
- Helm (2005) : Section 32.4: Parabolic PdesNo ratings yetHelm (2005) : Section 32.4: Parabolic Pdes24 pages
- TS3 Trip Unit Test Set Instructions Manual: TS3 1SDH000670R0002 L3852 ENNo ratings yetTS3 Trip Unit Test Set Instructions Manual: TS3 1SDH000670R0002 L3852 EN44 pages
- Enclosures and Numbering/Lettering ParagraphsNo ratings yetEnclosures and Numbering/Lettering Paragraphs2 pages
- BBD - MyCampus - Attendance Not Submitted Part-2No ratings yetBBD - MyCampus - Attendance Not Submitted Part-24 pages
- Approved Amount Maruti Suzuki Wagon R Eligible Maximum Loan AmountNo ratings yetApproved Amount Maruti Suzuki Wagon R Eligible Maximum Loan Amount1 page
- Lecture07 Unit01 Call by Value and Call by ReferenceNo ratings yetLecture07 Unit01 Call by Value and Call by Reference4 pages
- EMSN-049: Reconstruction Monitoring of ST Martin and ST Barthélemy Islands (Post IRMA)No ratings yetEMSN-049: Reconstruction Monitoring of ST Martin and ST Barthélemy Islands (Post IRMA)3 pages
- Thorsten Altenkirch and Jonathan Grattage - A Functional Quantum Programming LanguageNo ratings yetThorsten Altenkirch and Jonathan Grattage - A Functional Quantum Programming Language10 pages
