See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/378629402

Plant Diseases Diagnosis with Artificial Intelligence (AI)

Chapter · March 2024

DOI: 10.1007/978-981-99-9621-6_12

CITATIONS READS

2 3,836

3 authors, including:

Syed Muzammil Munawar

Bharathidasan Engineering College, Anna University, Chennai, India
37 PUBLICATIONS 166 CITATIONS

SEE PROFILE

All content following this page was uploaded by Syed Muzammil Munawar on 07 March 2024.

The user has requested enhancement of the downloaded file.

Microorganisms for Sustainability 47
Series Editor: Naveen Kumar Arora

Aditya Khamparia
Babita Pandey
Devendra Kumar Pandey
Deepak Gupta Editors

Microbial Data
Intelligence and
Computational
Techniques
for Sustainable
Computing
Microorganisms for Sustainability

Volume 47

Series Editor
Naveen Kumar Arora, Environmental Microbiology, School for Environmental
Science, Babasaheb Bhimrao Ambedkar University, Lucknow, Uttar Pradesh, India
Microorganisms have been in existence since the origin of life on earth and can
survive the most extreme habitats or conditions on earth. Microorganisms are
involved in regulating biogeochemical cycles, maintaining plant and animal health,
and sustaining the global food chain. Moreover, they play crucial roles in addressing
the challenges of climate change and achieving the targets of Sustainable Develop-
ment Goals (SDGs).
This multidisciplinary book series captures the role of microbes towards building
a sustainable world, while encompassing cutting-edge technologies and current
needs across various fields such as agriculture sustainability, bioremediation, resto-
ration of degraded habitats and wastelands, and food security. Additionally, this
series explores microbial applications in industries, and building their utilization in
clean and green energy solutions. Furthermore, themes like microbial secondary
metabolites, extremophilic microbes and modern omics, including next generation
sequencing and metagenomics, are also covered in this series.
With contributions from researchers across the globe, this series addresses the
important call of ‘One Planet-One Health-One Future’. It comprises a collection of
diverse volumes that provides insights for scientists, young researchers, educators
and decision‐makers in the government, private sector, and non‐governmental
organizations, empowering their efforts to achieve the global goals.
The series invites, evaluates, and accepts book proposals to ensure a diverse,
inclusive, and evolving program. The final decision regarding acceptance rests with
the series editor.
Peer-review
This book series follows a stratified review process. Proposals for individual
volumes are reviewed by the series editor and then the editorial board members. On a
case-to-case basis, external reviewers are also invited for further evaluation of the
book proposal. Review of the chapters is the responsibility of the volume editor(s). A
manuscript submission platform has been recently made available to the authors,
volume editors and the series editor.
Aditya Khamparia • Babita Pandey •
Devendra Kumar Pandey • Deepak Gupta
Editors

Microbial Data Intelligence

and Computational
Techniques for Sustainable
Computing
Editors
Aditya Khamparia Babita Pandey
Department of Computer Science Department of Computer Science
Babasaheb Bhimrao Ambedkar University Babasaheb Bhimrao Ambedkar University
Amethi, Uttar Pradesh, India Lucknow, Uttar Pradesh, India

Devendra Kumar Pandey Deepak Gupta

School of Bioengineering and Biosciences Department of Computer Science
Lovely Professional University Maharaja Agrasen Institute of Technology
Phagwara, Punjab, India Delhi, Delhi, India

ISSN 2512-1901 ISSN 2512-1898 (electronic)

Microorganisms for Sustainability
ISBN 978-981-99-9620-9 ISBN 978-981-99-9621-6 (eBook)
https://doi.org/10.1007/978-981-99-9621-6

© The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore
Pte Ltd. 2024
This work is subject to copyright. All rights are solely and exclusively licensed by the Publisher, whether
the whole or part of the material is concerned, specifically the rights of translation, reprinting, reuse of
illustrations, recitation, broadcasting, reproduction on microfilms or in any other physical way, and
transmission or information storage and retrieval, electronic adaptation, computer software, or by
similar or dissimilar methodology now known or hereafter developed.
The use of general descriptive names, registered names, trademarks, service marks, etc. in this publication
does not imply, even in the absence of a specific statement, that such names are exempt from the relevant
protective laws and regulations and therefore free for general use.
The publisher, the authors, and the editors are safe to assume that the advice and information in this
book are believed to be true and accurate at the date of publication. Neither the publisher nor the authors or
the editors give a warranty, expressed or implied, with respect to the material contained herein or for any
errors or omissions that may have been made. The publisher remains neutral with regard to jurisdictional
claims in published maps and institutional affiliations.

This Springer imprint is published by the registered company Springer Nature Singapore Pte Ltd.
The registered company address is: 152 Beach Road, #21-01/04 Gateway East, Singapore 189721,
Singapore

Paper in this product is recyclable.

Preface

Microbes are ubiquitous in nature, and their interactions among each other is a key
strategy for colonizing diverse habitats. The core idea of sustainable computing is to
deploy algorithms, models, policies, and protocols to improve energy efficiency and
management of resources, enhancing ecological balance, biological sustenance, and
other services on societal contexts. This book offers a comprehensive intelligent and
computational techniques for microbial data associated with either plant microbe,
human microbes, etc. The readers will be able to understand the positive findings as
well as the negative findings obtained by the usage of computational AI and
distributed computing techniques for microbial data. It entails data extraction from
various sources followed by pre-processing of data, and how to make effective use of
extracted data for application-based research. The book also involves computer-
assisted tools for visualization and representation of complex microbial data. The
book explores the conventional methods as well as the most recently recognized
high-throughput technologies which are important for productive agroecosystems to
feed the growing global population.
The main reason behind the success rate of deep learning and biomedical data
analysis techniques is its ability to reason and learn in an environment of unique data
and real case studies. This book will focus on involvement of microbial data
intelligence assisted and plant treatment and care-driven intelligent computing
methods, state of arts, novel findings, and recent advances in different applications
and areas like drug and plant image classification with a wide range of theory and
methodologies has been investigated to tackle the complex and challenging
problems.
Gathering the contributions by active researchers in these fields, the book covers
the theories as well as important real-time practical considerations. This book also
includes the design of a set of AI hybrid algorithms in detail, showing how to use
them in practice to solve problems relating to genome and plant image classification,
data analysis, bioinformatics, and engineering control. It is intended as a reference
guide to advanced hybrid computational intelligence methods for graduate students
and researchers in applied mathematics and optimization, computer science, and

v
vi Preface

engineering. This book is of interest to teachers, researchers, microbiologist, com-

puter bioinformatics scientists, plant and environmental scientist, and those inter-
ested in environment stewardship around the world. The book also serves as an
additional reading material for undergraduate and graduate students of computer
science, biomedical, agriculture, human science, forestry, ecology, soil science, and
environmental sciences, and policymakers consider this a useful book to read.

Objective of the Book

The primary emphasis of this book is to introduce different computational

intelligence-assisted techniques, methodology, and intelligent algorithms applied
to categorize and classify microbial-assisted plant datasets. Gathering contributions
by active researchers in those fields, the book covers the theories as well as important
practical considerations. In turn, it provides an overview of microbial data-driven
image analysis, deep learning, computer vision, and chaotic optimization enabled
evaluation of the proposed solutions in the manufacturing sector and compares the
advantages and disadvantages related to the same. This book will endeavor to endow
with significant frameworks, theory, design methods, and the latest empirical
research findings in the area of intelligent computing.

Amethi, Uttar Pradesh, India Aditya Khamparia

Lucknow, Uttar Pradesh, India Babita Pandey
Phagwara, Punjab, India Devendra Kumar Pandey
Delhi, Delhi, India Deepak Gupta
Contents

1 The Contribution of Artificial Intelligence to Drug Discovery:

Current Progress and Prospects for the Future . . . . . . . . . . . . . . . . 1
Umesh Gupta, Ayushman Pranav, Anvi Kohli, Sukanta Ghosh,
and Divya Singh
2 Prediction of Plant Disease Using Artificial Intelligence . . . . . . . . . . 25
Manoj Ram Tammina, K. Sumana, Pavitar Parkash Singh,
T. R. Vijaya Lakshmi, and Sagar Dhanraj Pande
3 Computer Vision-based Remote Care of Microbiological
Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
Pritesh Kumar Jain and Sandeep Kumar Jain
4 A Comparative Study of Various Machine Learning (ML)
Approaches for Fake News Detection in Web-based
Applications . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
Mahabub Hasan Mahalat, Sushree Bibhuprada B. Priyadarshini,
Sandip Swain, Shobhit Sahoo, Atish Mohapatra, and Mangaldeep Das
5 Analytics and Decision-making Model Using Machine Learning
for Internet of Things-based Greenhouse Precision Management
in Agriculture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
Ashay Rokade, Manwinder Singh, Anudeep Goraya, and Balraj Singh
6 DistilBERT-based Text Classification for Automated Diagnosis
of Mental Health Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
Diwakar and Deepa Raj
7 An Optimized Hybrid ARIMA-LSTM Model for Time Series
Forecasting of Agricultural Production in India . . . . . . . . . . . . . . . 107
Babita Pandey, Arvind Shukla, and Aditya Khamparia

vii
viii Contents

8 An Exploratory Analysis of Machine Intelligence-enabled

Plant Diseases Assessment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 121
Ashis Pattanaik, Agniva Bhattacharya, and Sushruta Mishra
9 Synergizing Smart Farming and Human Bioinformatics
Through IoT and Sensor Devices . . . . . . . . . . . . . . . . . . . . . . . . . . 139
Sandeep Kumar Jain and Pritesh Kumar Jain
10 Deep Learning-Assisted Techniques for Detection and
Prediction of Colorectal Cancer From Medical Images
and Microbial Modality . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 151
Ravi Kumar, Amritpal Singh, and Aditya Khamparia
11 Smart Farming and Human Bioinformatics System Based
on Context-Aware Computing Systems . . . . . . . . . . . . . . . . . . . . . . 171
Sini Anna Alex, T. P. Pallavi, and G. C. Akshatha
12 Plant Diseases Diagnosis with Artificial Intelligence (AI) . . . . . . . . . 187
Syed Muzammil Munawar, Dhandayuthabani Rajendiran,
and Khaleel Basha Sabjan
13 Analyzing the Frontier of AI-Based Plant Disease Detection:
Insights and Perspectives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 195
Mridula Dwivedi, Babita Pandey, and Vipin Saxena
14 Fuzzy and Data Mining Methods for Enhancing Plant
Productivity and Sustainability . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205
Khalil Ahmed, Mithilesh Kumar Dubey, Devendra Kumar Pandey,
and Sartaj Singh
15 Plant Disease Diagnosis with Artificial Intelligence (AI) . . . . . . . . . 217
Muhammad Naveed, Muhammad Majeed, Khizra Jabeen,
Nimra Hanif, Rida Naveed, Sania Saleem, and Nida Khan
16 Sustainable AI-Driven Applications for Plant Care
and Treatment . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 235
Muhammad Naveed, Nafeesa Zahid, Ibtihaj Fatima, Ayesha Saleem,
Muhammad Majeed, Amina Abid, Khushbakht Javed,
Rehmana Wazir, and Amina Qasim
17 Use Cases and Future Aspects of Intelligent Techniques
in Microbial Data Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 259
Muhammad Naveed, Zaibun-nisa Memon, Muhammad Abdullah,
Syeda Izma Makhdoom, Arooj Azeem, Sarmad Mehmood,
Maida Salahuddin, Zeerwah Rajpoot, and Muhammad Majeed
Contents ix

18 Early Crop Disease Identification Using Multi-fork Tree

Networks and Microbial Data Intelligence . . . . . . . . . . . . . . . . . . . . 281
S. S. Ittannavar, B. P. Khot, Vibhor Kumar Vishnoi,
Swati Shailesh Chandurkar, and Harshal Mahajan
19 Guarding Maize: Vigilance Against Pathogens Early
Identification, Detection, and Prevention . . . . . . . . . . . . . . . . . . . . . 301
Khalil Ahmed, Mithilesh Kumar Dubey, and Sudha Dubey
20 Comprehensive Analysis of Deep Learning Models for
Plant Disease Prediction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319
Narendra Pal Singh Rathor, Praveen Kumar Bhanodia,
and Aditya Khamparia
21 Enhancing Single-Cell Trajectory Inference and Microbial
Data Intelligence . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341
Bhargavi Posinasetty, Mukesh Soni, Sagar Dhanraj Pande,
Krishnendu Adhikary, and Dhirendra Kumar Tripathi
22 AI-Assisted Methods for Protein Structure Prediction
and Analysis . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 365
Divya Goel, Ravi Kumar, and Sudhir Kumar
Editors and Contributors

About the Editors

Aditya Khamparia has expertise in teaching, entrepreneurship, and research and

development of a decade. He is currently working as an assistant professor and
coordinator of the Department of Computer Science, Babasaheb Bhimrao Ambedkar
University, India. He received his Ph.D. degree from Lovely Professional Univer-
sity, Punjab, in May 2018. He has completed his M. Tech. from VIT University and
B. Tech. from RGPV, Bhopal. He has completed his PDF from UNIFOR, Brazil. He
has more than 100 research papers along with book chapters including more than
20 papers in top journals with cumulative impact factor of above 100 to his credit.
Additionally, he has authored and edited a cumulative of 11 books. His research
interests include machine learning, deep learning, educational technologies, and
computer vision.

Babita Pandey working as an associate professor in the Department of Computer

Science, Babasaheb Bhimrao Ambedkar University, Lucknow, India. Her research
interests include biomedical engineering, e-learning, computational intelligence, and
security systems. She has published more than 100 publications and conferences
including more than 40 SCI Indexed Journals.

Devendra Kumar Pandey is currently working as a professor at Lovely Profes-

sional University, India. He obtained his Ph.D. in biochemical engineering from the
Indian Institute of Technology, India. His main area of interest is related to pharma-
cology and toxicology, plant and soil sciences, and molecular sciences. His area of
expertise includes plant biotechnology, plant–microbe interaction, chromatography
techniques, i.e., HPTLC, HPLC, LC-MS, molecular markers and bioactive com-
pound markers for medicinal plants, and bioactive compounds. He has published
more than 100 articles in international journals with papers also in national and
international conferences contributed as author/co-author.

xi
xii Editors and Contributors

Deepak Gupta received a B.Tech. in 2006 from the Guru Gobind Singh
Indraprastha University, India. He received M.E. in 2010 from Delhi Technological
University, India, and Ph. D. in 2017 from Dr. APJ Abdul Kalam Technical
University, India. He has completed his post-doc from Inatel, Brazil. With
13 years of rich expertise in teaching and 2 years in the industry, he focuses on
rational and practical learning. He has contributed massive literature to the fields of
intelligent data analysis, biomedical engineering, artificial intelligence, and soft
computing. He has served as editor-in-chief, guest editor, and as associate editor in
various reputed journals. He has actively been organizing various reputed interna-
tional conferences. He has authored/edited 43 books. He has published 200 scientific
research publications including more than 100 SCI Indexed Journals.

Contributors

Muhammad Abdullah Biodiversity Park, Cholistan Institute of Desert Studies

(CIDS), The Islamia University of Bahawalpur, Bahawalpur, Pakistan
Amina Abid Department of Biotechnology, Faculty of Science and Technology,
University of Central Punjab, Lahore, Punjab, Pakistan
Krishnendu Adhikary Centurion University of Technology and Management,
Bhubaneswar, Odisha, India
Khalil Ahmed School of Computer Application, Lovely Professional University,
Phagwara, Punjab, India
G. C. Akshatha Department of CSE (AI & ML), Ramaiah Institute of Technology,
Bangalore, Karnataka, India
Sini Anna Alex Department of CSE (AI & ML), Ramaiah Institute of Technology,
Bangalore, Karnataka, India
N. Ashwini Department of Computer Science and Engineering, BMS Institute of
Technology and Management, Bengaluru, Karnataka, India
Arooj Azeem Department of Biotechnology, Faculty of Science and Technology,
University of Central Punjab, Lahore, Punjab, Pakistan
Praveen Kumar Bhanodia Computer Science Engineering, Acropolis Institute of
Technology and Research, Bhopal, Madhya Pradesh, India
Agniva Bhattacharya Kalinga Institute of Industrial Technology, Deemed to be
University, Bhubaneswar, Odisha, India
G. Bhavya Department of Information Science and Engineering, BMS Institute of
Technology and Management, Bengaluru, Karnataka, India
Editors and Contributors xiii

Swati Shailesh Chandurkar Pimpri Chinchwad College of Engineering, Pune,

India
Mangaldeep Das Computer Science & Information Technology, Institute of Tech-
nical Education and Research, Siksha ‘O’ Anusandhan Deemed to be University,
Bhubaneswar, India
Diwakar Diwakar University (A Central University), Lucknow, Uttar Pradesh,
India
Mithilesh Kumar Dubey School of Computer Application, Lovely Professional
University, Phagwara, Punjab, India
Sudha Dubey Department of Sociology, Lovely Professional University,
Phagwara, Punjab, India
Mridula Dwivedi Department of Computer Science, Babasaheb Bhimrao
Ambedkar University, Lucknow, Uttar Pradesh, India
Ibtihaj Fatima Department of Botany, University of Education, Lahore, Punjab,
Pakistan
Sukanta Ghosh SCSAI, SR University, Warangal, Telangana, India
Divya Goel Department of Biotechnology, H.N.B. Garhwal University, Srinagar
Garhwal, Uttarakhand, India
Anudeep Goraya School of Computer Science and Engineering, Lovely Profes-
sional University, Phagwara, Punjab, India
Umesh Gupta SCSET, Bennett University, Greater Noida, Uttar Pradesh, India
Nimra Hanif Department of Biotechnology, Faculty of Science and Technology,
University of Central Punjab, Lahore, Punjab, Pakistan
S. S. Ittannavar ECE Department, Hirasugar Institute of Technology, Belgaum,
India
Khizra Jabeen Department of Biotechnology, Faculty of Science and Technology,
University of Central Punjab, Lahore, Punjab, Pakistan
Pritesh Kumar Jain Department of Computer Science and Engineering, Shri
Vaishnav Vidyapeeth Vishwavidyalaya, Indore, Madhya Pradesh, India
Sandeep Kumar Jain Department of Computer Science and Engineering, Shri
Vaishnav Vidyapeeth Vishwavidyalaya, Indore, Madhya Pradesh, India
Khushbakht Javed Department of Biotechnology, Faculty of Science and Tech-
nology, University of Central Punjab, Lahore, Punjab, Pakistan
Aditya Khamparia Department of Computer Science, Babasaheb Bhimrao
Ambedkar University, Satellite Centre, Amethi, Uttar Pradesh, India
xiv Editors and Contributors

Nida Khan Department of Botany, University of Science and Technology Bannu,

Bannu, Khyber Pakhtunkhwa, Pakistan
B. P. Khot ECE Department, Hirasugar Institute of Technology, Belgaum, India
Anvi Kohli SCSET, Bennett University, Greater Noida, Uttar Pradesh, India
Ravi Kumar Department of Computer Science Engineering, Lovely Professional
University, Jalandhar, Punjab, India
Department of Computer Science Engineering, Jawaharlal Nehru Government Engi-
neering College, Sundernagar, Himachal Pradesh, India
Sudhir Kumar Department of Biotechnology, H.N.B. Garhwal University, Srina-
gar Garhwal, Uttarakhand, India
T. R. Vijaya Lakshmi Mahatma Gandhi Institute of Technology, Gandipet,
Hyderabad, India
Harshal Mahajan Indira College of Engineering and Management, Pune, India
Mahabub Hasan Mahalat Computer Science & Information Technology, Institute
of Technical Education and Research, Siksha ‘O’ Anusandhan Deemed to be
University, Bhubaneswar, India
Muhammad Majeed Department of Botany, University of Gujrat, Gujrat, Pakistan
Syeda Izma Makhdoom Department of Biotechnology, Faculty of Science and
Technology, University of Central Punjab, Lahore, Punjab, Pakistan
Sarmad Mehmood Department of Biotechnology, Faculty of Science and Tech-
nology, University of Central Punjab, Lahore, Punjab, Pakistan
Zaibun-nisa Memon Department of Zoology, Shah Abdul Latif University,
Khairpur Mirs, Sindh, Pakistan
Sushruta Mishra Kalinga Institute of Industrial Technology, Deemed to be Uni-
versity, Bhubaneswar, Odisha, India
Atish Mohapatra Computer Science & Information Technology, Institute of Tech-
nical Education and Research, Siksha ‘O’ Anusandhan Deemed to be University,
Bhubaneswar, India
Syed Muzammil Munawar Department of Biochemistry, C. Abdul Hakeem Col-
lege (Autonomous), Melvisharam, Vellore, Tamil Nadu, India
Muhammad Naveed Department of Biotechnology, Faculty of Science and Tech-
nology, University of Central Punjab, Lahore, Punjab, Pakistan
Rida Naveed Department of Biotechnology, Faculty of Science and Technology,
University of Central Punjab, Lahore, Punjab, Pakistan
Editors and Contributors xv

T. P. Pallavi Department of CSE (Cyber Security), Ramaiah Institute of Technol-

ogy, Bangalore, Karnataka, India
Sagar Dhanraj Pande School of Computer Science and Engineering, VIT-AP
University, Amaravati, Andhra Pradesh, India
Babita Pandey Babasaheb Bhimrao Ambedkar (Central) University, Lucknow,
Uttar Pradesh, India
Department of Computer Science, Babasaheb Bhimrao Ambedkar University, Luck-
now, Uttar Pradesh, India
Devendra Kumar Pandey School of Computer Application, Lovely Professional
University, Phagwara, Punjab, India
Ashis Pattanaik Kalinga Institute of Industrial Technology, Deemed to be Univer-
sity, Bhubaneswar, Odisha, India
Bhargavi Posinasetty Department of Masters in Public Health, The University of
Southern Mississippi, Hattiesburg, MS, USA
Ayushman Pranav SCSET, Bennett University, Greater Noida, Uttar Pradesh,
India
Sushree Bibhuprada B. Priyadarshini Computer Science & Information Tech-
nology, Institute of Technical Education and Research, Siksha ‘O’ Anusandhan
Deemed to be University, Bhubaneswar, India
Amina Qasim Department of Botany, Minhaj University Lahore, Lahore, Pakistan
Deepa Raj University (A Central University), Lucknow, Uttar Pradesh, India
Zeerwah Rajpoot Department of Biotechnology, Faculty of Science and Technol-
ogy, University of Central Punjab, Lahore, Punjab, Pakistan
Narendra Pal Singh Rathor Computer Science Engineering, Acropolis Institute
of Technology and Research, Bhopal, Madhya Pradesh, India
Ashay Rokade School of Electronics and Electrical Engineering, Lovely Profes-
sional University, Phagwara, Punjab, India
Shobhit Sahoo Computer Science & Information Technology, Institute of Techni-
cal Education and Research, Siksha ‘O’ Anusandhan Deemed to be University,
Bhubaneswar, India
Maida Salahuddin Department of Biotechnology, Faculty of Science and Tech-
nology, University of Central Punjab, Lahore, Punjab, Pakistan
Ayesha Saleem Department of Biotechnology, Faculty of Science and Technology,
University of Central Punjab, Lahore, Punjab, Pakistan
Sania Saleem Department of Plant Sciences, Quaid-i-Azam University, Islamabad,
Pakistan
xvi Editors and Contributors

Vipin Saxena Department of Computer Science, Babasaheb Bhimrao Ambedkar

University, Lucknow, Uttar Pradesh, India
Arvind Shukla Babasaheb Bhimrao Ambedkar (Central) University, Lucknow,
Uttar Pradesh, India
Amritpal Singh Department of Computer Science Engineering, Lovely Profes-
sional University, Phagwara, Punjab, India
Balraj Singh School of Computer Science and Engineering, Lovely Professional
University, Phagwara, Punjab, India
Divya Singh SCSET, Bennett University, Greater Noida, Uttar Pradesh, India
Manwinder Singh School of Electronics and Electrical Engineering, Lovely Pro-
fessional University, Phagwara, Punjab, India
Pavitar Parkash Singh Department of Management, Lovely Professional Univer-
sity, Phagwara, Punjab, India
Sartaj Singh School of Computer Application, Lovely Professional University,
Phagwara, Punjab, India
Mukesh Soni Department of CSE, University Centre for Research & Development,
Chandigarh University, Mohali, Punjab, India
K. Sumana Department of Microbiology, JSS AHER, Mysuru, Karnataka, India
Sandip Swain Computer Science & Information Technology, Institute of Techni-
cal Education and Research, Siksha ‘O’ Anusandhan Deemed to be University,
Bhubaneswar, India
Manoj Ram Tammina Innovation, Bread Financial, Columbus, OH, USA
Dhirendra Kumar Tripathi Sri Satya Sai University of Technology and Medical
Sciences, Sehore, MP, India
Vibhor Kumar Vishnoi College of Computing Sciences and Information Tech-
nology, Teerthanker Mahaveer University, Moradabad, India
Rehmana Wazir Department of Biotechnology, Faculty of Science and Technol-
ogy, University of Central Punjab, Lahore, Punjab, Pakistan
Nafeesa Zahid Department of Botany, University of Kotli, Kotli, Azad Jammu and
Kashmir, Pakistan
Chapter 1
The Contribution of Artificial Intelligence
to Drug Discovery: Current Progress
and Prospects for the Future

Umesh Gupta , Ayushman Pranav , Anvi Kohli , Sukanta Ghosh ,

and Divya Singh

Abstract The swift progress of artificial intelligence (AI) is fundamentally altering

the terrain of drug discovery, carrying the substantial potential to accelerate the
pinpointing of new drugs and improve the effectiveness and efficiency of the drug
development process. Across various stages of drug discovery, AI methodologies
are proving instrumental:
1. Target Identification and Validation: AI demonstrates prowess by sifting
through extensive genomic and proteomic datasets to discern and affirm fresh
drug targets. This computational prowess enables the identification of potential
candidates for therapeutic intervention (Kim et al., Biotechnol Bioprocess Eng
25:895–930, 2020).
2. Virtual Screening: The application of AI extends to efficiently screening vast
compound libraries. AI predictions of binding affinities and pertinent properties
offer a streamlined approach to identifying promising drug candidates
(Sahayasheela et al., Nat Prod Rep 39:2215, 2022).
3. Drug Design: AI’s capabilities span the design phase, aiding in creating innova-
tive drug molecules with specified attributes like enhanced potency, selectivity,
and pharmacokinetics (Blanco-Gonzalez et al. Pharmaceuticals 16:891, 2023).
4. Drug Repurposing: AI breathes new life into existing drugs, uncovering alter-
nate applications. This strategy is a cost-effective and time-sensitive avenue for
developing new treatment avenues (Ren et al., Chem Sci 14:1443–52, 2023).
5. Clinical Trial Design: Leveraging AI, clinical trial frameworks can be opti-
mized. AI empowers the precise selection of patients, appropriate dosages, and

U. Gupta (✉) · A. Pranav · A. Kohli · D. Singh

SCSET, Bennett University, Gr. Noida, Uttar Pradesh, India
e-mail: [email protected]
S. Ghosh
SCSAI, SR University, Warangal, Telangana, India

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 1
A. Khamparia et al. (eds.), Microbial Data Intelligence and Computational
Techniques for Sustainable Computing, Microorganisms for Sustainability 47,
https://doi.org/10.1007/978-981-99-9621-6_1
2 U. Gupta et al.

predictive assessments of trial success probability (Keshavarzi Arshadi et al.,

Artif Intell 65, 2020).
While AI’s integration into drug discovery remains relatively nascent, it holds the
potential to revolutionize the field. Recent strides in AI technology have enabled the
resolution of complex challenges, such as identifying targets for refractory ailments
and engineering drugs with heightened efficacy and reduced toxicity. The trajectory
of AI in drug discovery appears promising. Its influence is poised to intensify,
driving expedited drug discovery, refining the efficiency of the developmental
journey, and hastening the availability of novel treatments for patients.

Keywords Drug discovery · Artificial intelligence · Machine learning · sustainable

computing · Deep learning

1.1 Introduction

In this chapter, we undertake a detailed exploration that dives into the interdependent
connection between drug discovery and the transformative capabilities of artificial
intelligence (AI). The journey of drug discovery, a complex voyage aimed at
discovering new chemical entities (NCEs) with the potential to drive therapeutic
advancements, traverses a terrain characterized by carefully defined stages. These
encompass the meticulous identification of disease-triggering molecular targets, the
astute curation and refinement of NCEs, the arduous passage through preclinical
scrutiny, and the exacting challenge of human clinical trials. Amid the range of
challenges and uncertainties along this journey, AI’s indelible mark as the conduit
for pioneering treatments remains unwavering. Within this narrative, the emergence
of AI as a catalyzing agent takes center stage, orchestrating enhancements that
streamline process efficiency and augment outcomes. Our expedition unfurls, cast-
ing a luminous spotlight on the manifold ways AI’s prowess resonates across pivotal
dimensions, spanning the domains of target identification, lead discovery, optimi-
zation, preclinical assessment, and clinical trials. Furthermore, we delve into the
resurgence of microbes, hallowed sources of medicinal innovation, as AI infuses a
renewed vitality into their exploration within the realm of drug discovery. As AI’s
trajectory seamlessly converges with the path of drug discovery, it heralds a para-
digm shift—an era in which the fusion of innovation and computational brilliance
forges novel pathways within the domain of therapeutics. This chapter is a testament
to the symbiosis of scientific ingenuity and AI’s transformative potential, poised to
unveil unprecedented horizons within the expansive field of drug discovery.
1 The Contribution of Artificial Intelligence to Drug Discovery:. . . 3

1.2 Historical Evolution of Drug Discovery

Conventional avenues of drug discovery have long relied on a trial-and-error

methodology, entailing the meticulous screening of expansive compound libraries
to identify those exhibiting sought-after biological activities. However, this
approach is arduous, time-intensive, and often yields compounds with undesirable
side effects or toxicity.
The landscape shifted with the advent of computational methodologies in the
1990s, which introduced predictive computer models to gauge compound properties
and biological potentials. This infusion of computational prowess expedited drug
discovery and augmented the success rate of subsequent drug development
endeavors.
In recent times, artificial intelligence (AI) has started to gain attention in drug
discovery. The wide array of tools offered by AI, including machine learning and
deep learning, brings forth the ability to analyze extensive data repositories, identify
complex patterns, and generate predictive insights. This convergence of AI and drug
discovery promises to fundamentally revolutionize the process, ushering in
enhanced efficiency and efficacy. Illustrating this synergy are specific instances
where AI is leaving its indelible mark on drug discovery:
• Drug Repurposing
AI assumes a pivotal role in reimagining the potential uses of existing drugs.
By scrutinizing comprehensive datasets, AI identifies drugs ripe for repurposing,
as exemplified by identifying remdesivir, an Ebola-originated drug, for COVID-
19 treatment (Kim et al. 2020).
• Target Identification
AI’s analytical capabilities come to the fore in deciphering the intricate
landscape of disease-associated proteins and molecules. By analyzing vast geno-
mic and proteomic datasets, AI zeroes in on potential drug targets (Sahayasheela
et al. 2022).
• Lead Optimization
AI breathes fresh life into the lead optimization process, tailoring novel
compounds that exhibit heightened potency, superior safety profiles, and boosted
viability in clinical trials (Blanco-Gonzalez et al. 2023).
• Virtual Screening
In virtual screening, AI acts as an accelerator, sieving through expansive
compound libraries with heightened precision to identify compounds more likely
to exhibit desired biological activities (Ren et al. 2023).
Although the incorporation of AI into drug discovery is at an early phase, its
capacity to revolutionize the industry is unquestionable. The capabilities of AI are
poised to bring about a fresh era characterized by enhanced efficiency, effectiveness,
and innovative breakthroughs in the field of drug discovery. As the synergistic
partnership between AI and drug discovery gathers momentum, the future holds
great potential for uncovering new targets and ground-breaking treatments for
various diseases.
4 U. Gupta et al.

1.3 Fundamentals of Artificial Intelligence in Drug

Discovery

Contained within this chapter is an investigation into the fundamental principles of

AI applied within the field of drug discovery. The rapid evolution of AI holds the
transformative potential to reshape various sectors, and drug discovery is no excep-
tion. Integrating AI techniques has opened avenues to streamline intricate processes
such as data mining, virtual screening, and molecular design, accelerating drug
discovery and mitigating the financial burdens of new drug development. Central
to AI’s influence in drug discovery are vital concepts that warrant exploration:
• Machine Learning: A cornerstone of AI, machine learning enables computers to
glean insights from data without explicit programming. By identifying patterns
and making predictions, machine learning algorithms contribute to tasks ranging
from identifying potential drug targets to predicting drug candidates’ efficacy and
toxicity and optimizing the design of novel drugs (Kim et al. 2020).
• Deep Learning: Within the purview of machine learning lies deep learning, a
paradigm utilizing artificial neural networks to glean insights from data. Inspired
by the human brain, these networks unravel intricate data patterns. Deep learning
is essential in drug discovery to dissect molecular data and ascertain potential
drug targets (Sahayasheela et al. 2022).
• Reinforcement Learning: A transformative AI concept, reinforcement learning
enables agents to learn optimal behavior within an environment through trial and
error, reinforced by rewards for favorable actions and penalties for unfavorable
ones. In drug discovery, reinforcement learning holds promise for designing
effective and safe drugs through iterative learning (Blanco-Gonzalez et al. 2023).
The very fabric of AI in drug discovery is interwoven with mathematical foun-
dations encompassing algorithms and statistical methods:
1. Linear Regression: A statistical tool to predict a continuous variable from a set of
independent variables, it finds utility in estimating drug candidate efficacy from
molecular properties (Ren et al. 2023). The equation for linear regression is.

y = mx þ b

where y is the predicted value, m is the slope, b is the y-intercept, and x is the
independent variable.
2. Support Vector Machines: A machine learning algorithm, support vector
machines are adept at classification and regression tasks, effectively identifying
potential drug targets and predicting drug candidate toxicity (Keshavarzi Arshadi
et al. 2020).
The equation for support vector machines is:
1 The Contribution of Artificial Intelligence to Drug Discovery:. . . 5

fðxÞ = w  x þ b

where f(x) is the predicted value, w is the weight vector, b is the bias, and x is
the independent variable.
3. Artificial Neural Networks: These are also inspired by the human brain and excel
in learning complex data patterns, presenting a practical approach for tasks such
as image and video processing and natural language processing. Within drug
discovery, they play a pivotal role in molecular data analysis and target identifi-
cation (Jiménez-Luna et al. 2021). The equation for artificial neural networks is:

y = fðWx þ bÞ

where y is the predicted value, W is the weight matrix, x is the input vector, b is
the bias vector, and f() is the activation function.
Exploring the differences between machine learning and deep learning, the
chapter unravels the equations that underpin these approaches, offering a mathemat-
ical insight into their mechanisms. Machine learning algorithms, being more
straightforward, and deep learning algorithms, with their complexity and ability to
learn from unlabeled data, are utilized in various tasks, including classification,
regression, image recognition, and language processing.
In essence, this chapter illuminates the ever-evolving landscape where AI’s
potential converges with the intricacies of drug discovery. AI’s rapid evolution is
poised to revolutionize industries, and its potential in drug discovery is undeniable.
Through computational ingenuity, AI unlocks the door to expedite drug discovery,
propelling innovation and efficiency to unveil novel therapeutic avenues.

1.4 Data-centric Approaches in Artificial Intelligence

for the Field of Drug Discovery

• The significance of extensive datasets in drug discovery (Kim et al. 2020;

Sahayasheela et al. 2022; Blanco-Gonzalez et al. 2023)
The intricacies of drug discovery, characterized by its duration and costs, often
confront hurdles due to limited accessible data. Integrating big data can prove
transformative, furnishing researchers with deep insights into drug targets, dis-
ease mechanisms, and potential candidates. This repository of information can
drive the identification of novel drug targets, enable the design of fresh pharma-
ceuticals, and optimize the developmental trajectory. For instance, the infusion of
big data can facilitate:
– Identifying novel drug targets via comprehensive analyses of genomic,
transcriptomic, and proteomic datasets
6 U. Gupta et al.

– Design of innovative drugs by sifting through extensive compound libraries to

identify those interacting with specific targets
– Enhancement of the developmental trajectory through the utilization of data
for predicting the safety and effectiveness of prospective drugs
• Preprocessing and normalization of data: Mathematical frameworks (Kim et al.
2020; Ren et al. 2023):
Before employing big data for drug discovery, the preparatory steps encom-
pass data pre-processing and normalization. This entails refining the data, purging
errors and anomalies, and adapting it to a format amenable to machine learning
algorithms. Several mathematical models are instrumental in data pre-processing
and normalization, including:
– Principal Component Analysis (PCA): A statistical technique that condenses
dataset dimensionality while conserving crucial information
– K-means clustering: A machine learning algorithm that clusters similar data
points together
– Gaussian Mixture Models (GMMs): A class of probabilistic models
representing data point distributions
• Overfitting mitigation and regularization through mathematical equations (Kim
et al. 2020; Keshavarzi Arshadi et al. 2020):
In machine learning for drug discovery, overfitting arises when a model
excessively learns from the training data, hampering its capacity to apply that
knowledge to new data and resulting in inaccurate predictions. To counter
overfitting, an array of regularization techniques exists, including:
– L1 Regularization: Penalizes models for substantial coefficients, curtailing
overfitting by curbing reliance on training data
– L2 Regularization: Discourages significant squared coefficients, curbing
overfitting’s impact on training data
– Dropout: A technique that randomly eliminates certain features during train-
ing, ensuring that no single feature is relied upon
The mathematical formulations underpinning these regularization techniques are
as follows:
• L1 Regularization:

J ðθÞ = LðθÞ þ α θ 1

where J(θ) is the cost function of the model. L(θ) is the loss function of the
model. α is the regularization parameter. θ is the vector of model parameters
• L2 Regularization:
1 The Contribution of Artificial Intelligence to Drug Discovery:. . . 7

J ðθÞ = LðθÞ þ α θ 2

• Dropout:

PðdropoutÞ = 1 - p

where P(dropout) is the probability that a feature will be dropped out. p is the
dropout rate
The ascendancy of data-driven methodologies in drug discovery is palpable.
These approaches surmount the challenges intrinsic to the drug discovery process,
such as data scarcity and overfitting. By synergizing big data with machine learning,
researchers expedite the quest for novel pharmaceuticals, amplifying the efficiency
and efficacy of drug discovery processes.

1.5 Data-driven Approaches in AI for Drug Discovery

Microbial Natural Products (MNPs) encompass various compounds synthesized by

microorganisms, including bacteria, fungi, and algae. These compounds exhibit a
spectrum of biological activities, spanning antimicrobial, anticancer, and antiviral
properties. With a history rooted in traditional medicine, many MNPs have found
their place in modern therapeutic practices.
Exemplary MNPs include:
• Penicillin: A renowned antibiotic effective against diverse infections
• Taxol: A cancer-fighting agent used in breast cancer and other malignancies
• Artemisinin: An antimalarial weapon combatting malaria
• Sirolimus: An immunosuppressive tool aiding organ transplant acceptance
• The Transition: Traditional vs. AI-Powered Discovery from Microbial Sources
Historically, the discovery of MNPs relied on a blend of screening and
bioactivity tests, a time-intensive and resource-demanding endeavor often hin-
dered in identifying novel MNPs with desired attributes. Artificial intelligence
(AI) has recently become a driving force in accelerating the discovery of MNPs.
AI can mechanize several steps of traditional drug discovery, encompassing:
– Screening extensive compound libraries for bioactivity
– Predicting MNP structures and properties
– Identifying potential MNP targets
– Designing novel MNPs with specific characteristics
• The Contribution of Mathematical Models in Microbial Drug Discovery Driven
by Artificial Intelligence
The exploration of AI-driven microbial drug discovery pivots on using math-
ematical models. These models enable:
8 U. Gupta et al.

– Bioactivity prediction based on MNP chemical structures

– Prediction of MNP target proteins
• Tailoring MNPs with desired attributes
Prominent mathematical models in this domain encompass:
– Quantitative Structure-Activity Relationship (QSAR) models: Leverage MNP
chemical structures to predict bioactivity
– Molecular docking models: Anticipate MNP-protein binding affinities
– Virtual screening models: Screen extensive compound libraries for potential
bioactivity
• AI’s Applications in Microbial Natural Product Drug Discovery
AI catalyzes MNP discovery through diverse avenues, including:
– Automated Screening: AI automates compound library screening, reducing
time and costs.
– Target Identification: AI identifies potential MNP targets, streamlining
research efforts.
– MNP Design: AI aids in designing MNPs with desired properties, enhancing
effectiveness and safety.
– Drug Repurposing: AI identifies new uses for existing drugs, accelerating
treatment availability.
• Navigating Challenges and Opportunities
The usage of AI in microbial natural product drug discovery has its own chal-
lenges and opportunities: data availability, MNP complexity, and the evolving
nature of AI techniques. Opportunities encompass accelerated drug discovery, new
drug target exploration, and tailored drug design. AI holds immense potential in
reshaping microbial natural product drug discovery. Through addressing challenges
and leveraging opportunities, AI stands as a transformative force, poised to expedite
the development of novel drugs to combat a myriad of diseases.

1.6 Hurdles and Prospects in Artificial Intelligence

for the Field of Drug Discovery

1.6.1 Navigating Challenges and Embracing Prospects

in AI-driven Drug Discovery

The traditional drug discovery processes are lengthy, expensive, and frequently
inefficient, entailing a significant span of 10–15 years of research and finally getting
a new drug on the shelves. However, AI holds the transformative potential to
expedite this labyrinthine process through task automation, encompassing:
1 The Contribution of Artificial Intelligence to Drug Discovery:. . . 9

• Data Mining: AI’s prowess in sifting through vast biological and chemical
datasets to unearth potential drug targets and candidates
• Virtual Screening: Leveraging AI to comb through extensive compound libraries
in search of promising drug candidates
• Molecular Modeling: AI’s ability to forecast drug candidate structures, proper-
ties, and their interactions with target proteins
• Drug Design: Employing AI to craft novel drug candidates tailored to specific
properties

1.6.2 Pinpointing Bottlenecks in Traditional Drug Discovery

The traditional drug discovery process faces several roadblocks:

• Escalating Costs: Research and development costs are on a relentless upward
trajectory, exacerbated by the escalating complexity of the entire process.
• Time Intensive Journey: A new drug’s journey to the market spans 10–15 years,
contending with arduous regulatory hurdles and extensive clinical trials.
• Merger Success Rate: Only a minute fraction of drug candidates that tread the
clinical trial path receive FDA approval due to the intricate challenges of devel-
oping safe and efficacious drugs.

1.6.3 Untangling AI Implementation Challenges

While AI has the potential to alleviate some of the challenges, it presents its own set
of hurdles in drug discovery implementation:
• Data Dearth: AI models thrive on copious volumes of high-quality data, a luxury
often absent in the drug discovery arena.
• Process Complexity: Drug discovery is a multifaceted process demanding
nuanced representation in AI models to ensure efficacy.
• Model Interpretability: AI models can be enigmatic, complicating trust and
usability in drug discovery due to the opacity of their predictions.

1.6.4 Glimmers of Possibilities Unleashed by AI

In the face of challenges, AI heralds transformation by:

• Expediting Drug Discovery: Automation empowered by AI slashes time and
costs, expediting the journey from lab to market.
10 U. Gupta et al.

• Elevating Success Rates: AI identifies safer and more effective drugs, improving
the probability of successful outcomes.
• Personalizing Medicine: Tailoring drugs to individual patient needs becomes
feasible through AI’s predictive capabilities.
• Tackling Neglected Diseases: AI’s potential extends to crafting treatments for
challenging diseases, including rare conditions.

1.6.5 A New Dawn in Drug Discovery

AI, a symbol of innovation, has the capacity to reshape the narrative of drug
discovery. By confronting challenges head-on and embracing opportunities, AI
can streamline the process, bringing novel and efficacious treatments to patients
faster and more efficiently.

1.7 Case Study: AlphaFold’s Acceleration in Drug

Discovery

Within the dynamic realm of biomedical research, the ground-breaking protein

structure prediction tool AlphaFold, created by DeepMind, a subsidiary of Google
AI, has emerged as a revolutionary influence. Unveiled in 2020, its extraordinary
aptitude for accurately predicting protein structures has established it as a potent
accelerator in drug discovery. The AlphaFold stands ready to transform the tradi-
tional course of drug development by interpreting protein structures and unveiling
their enigmas.

1.7.1 Significance of AlphaFold

The importance of AlphaFold is fundamentally tied to its capacity to transform the

landscape of drug discovery. The conventional route to developing a new drug is
laborious, marked by extensive timelines and substantial expenses. AlphaFold
emerges as a promising beacon, aiming to accelerate this journey by providing
researchers with precise forecasts of protein structures. This fresh understanding
carries the potential to revolutionize the identification of drug targets and the creation
of novel medications. The arrival of AlphaFold signifies a pivotal moment, intro-
ducing a more streamlined and cost-effective pathway to advancing therapeutic
innovations.
1 The Contribution of Artificial Intelligence to Drug Discovery:. . . 11

1.7.1.1 Mathematical Mastery Behind AlphaFold’s Prognostic Abilities

AlphaFold’s exceptional predictive ability is built upon a sophisticated deep learning

model, carefully refined through exposure to an extensive collection of established
protein structures. Comparable to a virtual expert in proteomics, this model identifies
the recurring patterns that interconnect various proteins. This mastery equips
AlphaFold with the capability to foresee the structure of any protein, including
those that have never been observed before. The essence of AlphaFold’s predictions
lies in understanding the physics of protein folding. Proteins, the building blocks of
life, are intricate compositions of amino acids linked by peptide bonds. Folding, a
central phenomenon, is governed by the interactions among these amino acids.
AlphaFold employs its deep learning model to predict these interactions, ultimately
creating a blueprint of the protein’s structure.

1.7.2 Elevating Drug Discovery: The AlphaFold Impact

on CDK20 Inhibitor Discovery

A noteworthy case study highlights AlphaFold’s transformative role in drug discov-

ery. An innovative application was witnessed in the realm of cancer treatment,
specifically targeting CDK20 (Ren et al. 2023), a protein pivotal in cell division
and the proliferation of cancer cells. AlphaFold’s predictive prowess was harnessed
in this study to unravel CDK20’s structure. Armed with this crucial information,
scientists crafted a small molecule inhibitor tailored to bind to CDK20 and impede
its activity. The outcome was promising: the inhibitor effectively restrained cancer
cell growth in vitro, illuminating AlphaFold’s potential in hastening drug discovery.

1.7.3 AlphaFold’s Multi-dimensional Drug Discovery Impact

Beyond CDK20, AlphaFold’s footprint in drug discovery extends across various

diseases, encompassing Alzheimer’s, Parkinson’s, and cancer. For instance,
AlphaFold’s insights into the structure of tau, a protein implicated in Alzheimer’s
disease, enabled the design of molecules that counteract harmful aggregations.
However, AlphaFold’s journey is far from complete. Still evolving, it has already
engendered a seismic shift in drug discovery. As its capacities continue to grow, it
stands ready to play an even more crucial role in shaping the trajectory of upcoming
drug development.
12 U. Gupta et al.

1.7.4 Navigating Challenges and Seizing Opportunities

However, as AlphaFold strives to revolutionize drug discovery, it faces particular

challenges. Its high computational demands can pose an obstacle to expansive drug
discovery initiatives. Additionally, its predictions are not immune to errors, occa-
sionally resulting in misinterpretations that could lead to ineffective drug designs.
Nonetheless, the potential encapsulated by AlphaFold remains resolute. Equipped
with its strengths and the insights gained from overcoming challenges, AlphaFold
remains an invaluable resource for researchers in drug discovery. As it progresses
and evolves, its transformative influence is destined to grow stronger.

1.7.5 The Eclipsing Horizon

In conclusion, AlphaFold has surpassed its status as a mere computational instru-

ment. It symbolizes optimism, heralding a revolutionary shift in drug discovery.
Through its capacity to unveil protein structures and steer drug design, AlphaFold
has indelibly influenced the realm of pharmaceutical advancement. As it progresses,
it carries the potential to expedite the creation of innovative therapies, ushering us
into a future where drug discovery is not only accelerated but also remarkably
efficacious.

1.8 AI in the Era of Pandemics: Case of COVID-19

The global impact of the COVID-19 pandemic has been deeply distressing, causing
numerous fatalities and widespread economic turmoil. Amid these difficult times,
artificial intelligence (AI) emerged as a powerful weapon in the battle against the
pandemic, providing efficient solutions to tackle the challenges presented by the
virus.

1.8.1 Overview of the COVID-19 Pandemic

SARS-CoV-2, the virus behind the respiratory ailment that caused the world to shut
down in 2020, is a highly contagious pathogen. It primarily spreads through respi-
ratory droplets, often released when an infected person coughs or sneezes or when a
person encounters a contaminated surface. The illness presents a range of symptoms,
from mild to severe, including fever, cough, shortness of breath, and fatigue. In
critical instances, COVID-19 can lead to conditions like pneumonia, acute respira-
tory distress syndrome, and even fatality. Emerging in Wuhan, China, in December
1 The Contribution of Artificial Intelligence to Drug Discovery:. . . 13

2019, the virus rapidly crossed borders, escalating into a worldwide emergency by
March 2020. By March 2023, over 600 million individuals had contracted COVID-
19, with the virus claiming the lives of more than 15 million people worldwide.

1.8.2 AI’s Crucial Role in Drug Discovery and Vaccine

Development

In the dark times of the pandemic, the cutting-edge solutions offered by AI made it a
formidable weapon in the fight against COVID-19, yielding significant contribu-
tions, particularly in the realms of drug discovery and vaccine development.
• Drug Repurposing: One pivotal application of AI lies in drug repurposing, where
existing drugs are identified and evaluated for their potential efficacy against
COVID-19. For instance, the drug remdesivir, initially formulated for combating
the Ebola virus, has proven effective in treating COVID-19 patients.
• Virtual Screening: The formidable computational power of AI has facilitated the
virtual screening of vast databases of potential drug molecules. The approach
accelerates the drug discovery timeline by swiftly identifying compounds with
the potential to combat COVID-19.
• Structural Biology and AI: The proficiency of AI in structural biology has
facilitated the exploration of the complex framework of the SARS-CoV-2 virus.
This comprehension is crucial for identifying potential drug targets and providing
a clear path for drug design.
Moreover, AI benefited greatly in the faster development of the COVID-19
vaccines. Among these, the Pfizer-BioNTech vaccine relies on mRNA technology
and was developed using AI methodologies. This represents a remarkable fusion of
innovation and medical science.

1.8.3 Leveraging Mathematical Models for Drug Prediction

Alongside its contributions to drug and vaccine development, AI has fostered the
creation of intricate mathematical models to predict the potential effectiveness of
drug molecules against COVID-19. These models, driven by AI algorithms, play a
pivotal role in sifting through extensive compound libraries to identify those with
promising therapeutic attributes. An exemplary instance is the DeepDTA model,
which employs deep learning techniques to predict the binding affinity of
drug molecules to the SARS-CoV-2 virus. The help of this model in predicting
drug efficacy led to faster, more cost-effective client delivery of several promising
drug candidates. Another notable mathematical model is the aforementioned
AlphaFold, which employs deep learning to forecast the three-dimensional
14 U. Gupta et al.

structures of proteins. In the context of COVID-19, AlphaFold’s predictions have

facilitated the identification of potential drug targets by unraveling its structural
complexities.

1.8.4 A Glimpse into the Future

In summary, AI proved its utility in drug discovery during the COVID-19 pandemic.
Its multifaceted applications encompass expediting drug and vaccine discovery,
shaping predictive mathematical models, and enhancing diagnostic and treatment
approaches for COVID-19 patients. As the pandemic endures, the role of AI is
poised to expand even further. AI’s potential to hasten the development of novel
treatments and vaccines, improve patient care, and deepen our comprehension of the
virus holds tremendous promise. In the quest to conquer COVID-19, AI stands as a
symbol of innovation and a steadfast ally in the collective endeavor to overcome the
challenges posed by the pandemic.

1.9 Deep Learning in Antibiotic Discovery

The urgent need for solutions to the worldwide health crisis of antibiotic resistance is
undeniable. Conventional methods for discovering antibiotics are marked by their
slow pace and substantial expenses. Recently, the adoption of artificial intelligence
(AI) methods, particularly deep learning, is gaining traction as an approach to
accelerate and improve the efficiency of the processes of traditional methods.
1. Deep Learning: Deep learning, a subset of machine learning, harnesses artificial
neural networks to learn intricate patterns from data. Inspired by the human brain,
these networks excel in deciphering complex data patterns. Across diverse
domains, deep learning has demonstrated ground-breaking achievements, from
image recognition and natural language processing to drug discovery.
2. Leveraging Deep Learning in Antibiotic Discovery: Deep learning’s potential in
antibiotic discovery is multifaceted and promising:
(a) Identification of Novel Antibiotic Targets: Deep learning models can
identify potential antibiotic targets by analyzing extensive protein datasets.
These models are trained to predict whether a protein could serve as an
antibiotic target.
(b) Design of Enhanced Antibiotics: Deep learning facilitates the design of
antibiotics that combat antibiotic-resistant bacteria more effectively. Models
can predict essential properties like binding affinity to antibiotic targets and
toxicity.
1 The Contribution of Artificial Intelligence to Drug Discovery:. . . 15

(c) Prediction of Antibiotic Properties: By learning from known antibiotic

datasets, deep learning models predict crucial antibiotic properties, including
binding affinity to antibiotic targets and toxicity.
(d) Efficient Antibiotic Screening: Deep learning expedites antibiotic screening
by evaluating massive compound libraries for potential antibiotic activity.
Models predict the likelihood of a compound possessing antibiotic
capabilities.
3. Advantages of Deep Learning in Antibiotic Discovery: Several merits underscore
the deployment of deep learning:
(a) Big Data Analysis: Deep learning efficiently analyzes vast bacterial datasets
and their interactions with antibiotics, which is pivotal for effective antibiotic
discovery.
(b) Pattern Recognition: Deep learning excels in discerning complex patterns,
thereby aiding in identifying fresh antibiotic targets and creating innovative
antibiotics.
(c) Rapid Screening: Deep learning accelerates antibiotic screening, enabling
quick identification of potential candidates.
4. Challenges of Deep Learning in Antibiotic Discovery: While promising, deep
learning in antibiotic discovery faces challenges:
(a) Data Quantity: Deep learning mandates substantial data for model training,
which can be arduous and costly.
(b) Complexity and Interpretability: Complex models pose challenges in
interpretability, hindering the comprehension of decision-making processes.
(c) Bias Concerns: Deep learning models can exhibit bias, leading to inaccurate
predictions.
Despite these hurdles, deep learning is a promising avenue to address the global
antibiotic resistance crisis, given its potential to expedite antibiotic discovery and
counteract antibiotic-resistant strains.
1. Diverse Deep Learning Techniques and Models: Several deep learning tech-
niques and models serve antibiotic discovery:
(a) Convolutional Neural Networks (CNNs): Particularly adept at image rec-
ognition, CNNs scrutinize protein structures to identify potential targets for
antibiotics.
(b) Recurrent Neural Networks (RNNs): Specializing in sequential data,
RNNs predict new antibiotic properties, aiding the design of novel
compounds.
(c) Generative Adversarial Networks (GANs): GANs generate new data and
can be applied to screening antibiotics by generating new compounds likely
to possess antibiotic activity.
2. Mathematical Foundations of Deep Learning Predictions: The predictive power
of deep learning models hinges on essential equations:
16 U. Gupta et al.

(a) Loss Function: Measures model performance, minimized during training

(b) Optimizer: Algorithm updating model parameters during training
(c) Backpropagation: Calculates gradients of the loss function regarding model
parameters, pivotal for the optimizer’s updates
3. Conclusion: Deep learning’s integration into antibiotic discovery holds immense
promise. Its potential to hasten the identification of antibiotics and combat
antibiotic resistance is unparalleled. Despite obstacles, the combination of deep
learning’s computational prowess and antibiotic discovery’s urgency suggests a
symbiotic relationship that could reshape the landscape of antibiotic research and
offer hope in the face of escalating antibiotic resistance.

1.10 AI Techniques in Antibiotic and Antiviral

Development

In the rapidly evolving landscape of drug discovery, a slew of cutting-edge AI

techniques is taking center stage, offering ground-breaking solutions for the formi-
dable challenges posed by the development of antibiotics and antiviral agents. These
emerging AI techniques can potentially revolutionize the future of medicine by
expediting drug discovery, tailoring treatments to individuals, and ushering in a
new era of predictive mathematical models.
1. Emerging AI Techniques in Drug Discovery: Drug discovery has been revolu-
tionized with the introduction of Artificial Intelligence methods such as
(a) Machine Learning (ML): Harnessing ML algorithms, extensive datasets
containing biological information and chemical compounds can be scruti-
nized to identify potential drug targets and design novel drug molecules.
ML’s prowess spans from target selection to compound optimization,
streamlining the discovery process (Kim et al. 2020; Sahayasheela et al.
2022; Blanco-Gonzalez et al. 2023; Ren et al. 2023; Keshavarzi Arshadi
et al. 2020; Jiménez-Luna et al. 2021; Paul et al. 2021; Zhang et al. 2017).
(b) Deep Learning (DL): Nested within ML, DL employs artificial neural net-
works to glean insights from data. Renowned for tasks like image recognition
and natural language processing, DL is increasingly applied to drug discov-
ery. It exhibits remarkable potential in forecasting drug toxicity and gauging
the binding affinity of drugs with their intended targets, thus expediting
optimization processes (Jiménez-Luna et al. 2021; Paul et al. 2021; Zhang
et al. 2017).
(c) Generative Adversarial Networks (GANs): A subset of DL, GANs generate
fresh drug molecules akin to existing compounds while boasting enhanced
properties. This innovation facilitates the creation of drug candidates with
improved attributes (Jiménez-Luna et al. 2021; Zhang et al. 2017).
1 The Contribution of Artificial Intelligence to Drug Discovery:. . . 17

(d) Bayesian Optimization: This statistical method is pivotal in experiment

design optimization. In the context of drug discovery, Bayesian optimization
identifies promising drug compounds for clinical trials, optimizing resource
allocation (Jiménez-Luna et al. 2021; Paul et al. 2021).
(e) Natural Language Processing (NLP): The interplay between computers and
human languages constitutes the essence of NLP. Its potential in drug dis-
covery lies in dissecting medical literature and clinical trial data to unearth
potential drug targets and evaluate the safety and efficacy of novel drugs.
These emerging AI techniques epitomize the promise of rapid innovation in
drug discovery. As AI advances, we can anticipate even more revolutionary
applications in drug discovery.
2. Potential of Personalized Medicine Through AI: AI holds the transformative
power to revolutionize personalized medicine, which hinges on tailoring medical
interventions to suit individual patient needs. AI interventions encompass:
(a) Precision Patient Selection: By analyzing genetic data, AI identifies genes
linked to specific diseases. This knowledge, in turn, aids in identifying drugs
that are likely to be effective for treating the identified disease.
(b) Treatment Response Prediction: AI employs patient data to forecast individ-
ual responses to drugs, enabling the optimization of treatment regimens for
better patient outcomes.
(c) Customized Dosages: Leveraging AI’s analytical prowess, personalized dos-
ages can be calculated based on an individual’s unique characteristics and
responses to treatment.
The realm of personalized medicine, buoyed by AI, promises to enhance
treatment outcomes, minimize costs, and reshape medical care by tailoring
therapies to individual patients.
3. Predictive Mathematical Models Shaping Drug Discovery: Predictive mathemat-
ical models are a cornerstone of modern drug discovery. These models facilitate
the antimicrobial drug compound properties, revolutionizing the process. Key
predictive models include:
(a) Quantitative Structure-Activity Relationship (QSAR): By utilizing statistical
approaches, Quantitative Structure-Activity Relationship (QSAR) models
establish connections between the structures of drug compounds and their
biological functions. This process assists in foreseeing toxicity and effective-
ness across a diverse range of drug compounds.
(b) Molecular Dynamics (MD) Simulations: Through computational models,
MD simulations predict the behavior of molecules within biological systems.
These simulations extend to forecasting drug-target binding and the effects of
drug compounds on biological entities.
Predictive mathematical models offer profound insights into the behavior of drug
compounds, streamlining drug development and facilitating target identification.
18 U. Gupta et al.

In conclusion, AI’s influence on antibiotic and antiviral development is transfor-

mative. The convergence of emerging AI techniques like ML, DL, GANs, Bayesian
optimization, and NLP facilitates the identification of drug targets, the design of
novel molecules, and the personalization of drug treatments. Alongside these,
predictive mathematical models enrich drug discovery by anticipating compound
properties. As AI’s horizons expand, its potential to hasten the development of
innovative and effective drugs addressing many diseases shines ever brighter. The
interdependent partnership between AI and drug discovery is on the brink of
reshaping the healthcare landscape.

1.11 Applications of AI in Drug Discovery

The incorporation of artificial intelligence (AI) has transformed the field of drug
discovery, pushing for the beginning of a fresh era characterized by heightened
efficiency and effectiveness. AI-powered technologies are revamping multiple
stages of drug development, encompassing tasks ranging from identifying and
validating targets to screening and designing compounds.

1.11.1 Target Identification and Validation

AI leverages vast genomic, transcriptomic, and proteomic datasets to identify and

validate novel drug targets. AI algorithms delve into disease-specific genetic muta-
tions and intricate biological pathways, shedding light on potential therapeutic
avenues. Additionally, AI assesses the multidrug targets’ structural and functional
attributes.

1.11.2 Compound Screening and Design

The power of AI is evident in its ability to sift through expansive compound

libraries, pinpointing these with a high likelihood of interacting with target proteins.
AI also plays a role in designing new compounds with heightened effectiveness and
minimal side effects. Predicting binding affinities to target proteins and assessing
compound toxicity are among AI’s contributions.
1 The Contribution of Artificial Intelligence to Drug Discovery:. . . 19

1.11.3 Clinical Trial Optimization

AI optimizes clinical trials through refined patient selection, streamlined protocol

design, and advanced result analysis. Among AI’s functions are predicting patient
responsiveness to specific treatments and determining optimal drug dosages. By
enhancing trial efficiency, AI accelerates the pace of drug development.

1.11.4 Concrete Examples of AI’s Impact

In 2020, AI spotlighted a potential drug target for Alzheimer’s disease, the protein
BACE1, integral in forming amyloid plaques characteristic of the ailment (Kim et al.
2020). In 2021, AI was pivotal in designing a compound to combat drug-resistant
tuberculosis. The algorithm factored in target protein structure and the properties of
effective tuberculosis drugs to craft the compound (Sahayasheela et al. 2022). In
2022, AI optimized a clinical trial for a novel cancer drug, expertly identifying
promising trial participants and refining trial protocols (Blanco-Gonzalez et al.
2023). These instances merely scratch the surface of AI’s role in drug discovery.
As AI’s capabilities burgeon, its potential to reshape and expedite drug development
remains boundless.

1.12 The Future of AI in Drug Discovery

Artificial intelligence (AI) is swiftly reshaping the landscape of drug discovery,

offering the potential to expedite processes, enhance efficiency, and bolster the
efficacy of novel drug development.

1.12.1 Emerging AI Techniques in Drug Discovery

Several cutting-edge AI techniques are currently making strides in drug discovery:

• Machine Learning (ML): ML algorithms analyze vast chemical and biological
datasets, unveiling new drug targets and potential candidates (Kim et al. 2020).
• Deep Learning (DL): A subset of ML, DL algorithms excel in deciphering
intricate data patterns. Their effectiveness spans tasks like predicting compound
toxicity and identifying new drug targets (Sahayasheela et al. 2022; Blanco-
Gonzalez et al. 2023).
• Natural Language Processing (NLP): NLP techniques dissect textual data from
medical records and scientific literature, revealing drug targets and potential side
effects (Ren et al. 2023).
20 U. Gupta et al.

• Computer-Aided Drug Design (CADD): CADD employs simulations to predict

properties like pharmacokinetics, pharmacodynamics, and toxicity of potential
drug compounds (Keshavarzi Arshadi et al. 2020).

1.12.2 The Potential of Personalized Medicine via AI

AI’s potential extends to enabling personalized medicine—tailoring drug treatment

for individual patients. AI scrutinizes genetic profiles, medical histories, and more to
pinpoint optimal and safe drug treatments (Jiménez-Luna et al. 2021).

1.12.3 Predictive Mathematical Models Shaping Drug

Discovery’s Future

AI is instrumental in crafting predictive mathematical models for drug discovery.

These models forecast promising drug targets and candidates and the success or
failure of new drug development programs (Stokes et al. 2020; Melo et al. 2021;
Askr et al. 2023; Volkamer et al. 2023; Luukkonen et al. 2023).
• Anticipating the Future Role of AI in Drug Discovery
• The usage of AI in drug discovery remains in its nascent stages but holds
revolutionary potential. Its current acceleration of drug discovery is just the
beginning, with prospects for further transformation. The following is a glimpse
into AI’s future impact on drug discovery:
• Discovering Novel Drug Targets: Researchers will uncover new drug targets by
harnessing AI to analyze vast biological and chemical data. This process, cur-
rently labor-intensive, can be automated, drastically enhancing efficiency
(Kaushik and Raj 2020; Gupta et al. 2022a).
• Designing Effective Drug Compounds: AI’s prowess in screening vast libraries of
compounds will be harnessed to design novel drugs with enhanced efficacy and
safety (Paul et al. 2021; Askr et al. 2023).
• Predicting Development Program Outcomes: AI-driven predictive models will
evolve to project the success or failure of drug development programs. This
information will guide decision-making processes (Zhang et al. 2017; Prasad
and Kumar 2021; Gupta et al. 2021).
• Personalized Treatment: Leveraging AI’s insights into genetic profiles, medical
histories, and other factors, personalized medicine will become more refined and
effective (Prasad and Kumar 2021; Gupta et al. 2022c).
The prospects for AI in drug discovery are promising. As AI advances, its
capacity to transform the field, enhance efficiency, and introduce personalized
medicine will become increasingly evident. The path toward discovering new
1 The Contribution of Artificial Intelligence to Drug Discovery:. . . 21

drugs and advancing personalized treatments is set to benefit significantly from the
ongoing evolution of AI (Gupta et al. 2021, 2022a, 2022b).

1.13 Conclusion

1.13.1 Revolutionizing Microbial Drug Discovery

with Artificial Intelligence

The incorporation of artificial intelligence (AI) into the realm of microbial

drug discovery is rapidly gaining momentum, set to redefine the landscape of
novel drug development. AI methodologies are positioned to accelerate every aspect
of the drug discovery journey, encompassing tasks ranging from identifying targets
to designing and refining drugs.
• Target Identification: AI’s potential is showcased in identifying novel drug
targets by scrutinizing extensive genomic and proteomic datasets. AI offers
ground-breaking insights by discerning targets vital for pathogenic microbes’
survival, distinct from those in human cells (Kim et al. 2020; Sahayasheela et al.
2022; Luukkonen et al. 2023).
• Drug Design: The power of AI extends to crafting new drugs that optimize
efficacy and mitigate side effects. AI empowers the screening of vast compound
libraries, identifying candidates with sought-after attributes (Blanco-Gonzalez
et al. 2023; Ren et al. 2023).
• Drug Optimization: AI-driven drug optimization thrives in perfecting crucial
attributes like solubility, stability, and pharmacokinetics. By enhancing these
properties, AI contributes to the safety and efficacy of novel drugs (Keshavarzi
Arshadi et al. 2020; Volkamer et al. 2023).

1.13.2 Anticipating Future Trends and Breakthroughs

The trajectory of AI in microbial drug discovery is brimming with promise. As AI

techniques grow more sophisticated and potent, their application across the drug
discovery spectrum catalyzes the creation of novel drugs that are more potent, safer,
and expeditiously developed (Gupta et al. 2022c; Sarkar et al. 2023; Askr et al.
2023).
22 U. Gupta et al.

1.13.3 Potential Breakthroughs Envisioned by AI

• Novel Drug Target Identification: AI is essential to discovering drug targets

indispensable for pathogenic microbes’ survival while sparing human cells.
This breakthrough is poised to revolutionize treatment strategies.
• Precision Drug Design: AI’s prowess in compound library screening for effica-
cious candidates that minimize side effects translates to ground-breaking drug
design—the result being drugs that combat microbes effectively and safeguard
patient well-being.
• Optimized Drug Properties: AI’s contribution in refining drug attributes like
solubility, stability, and pharmacokinetics ensures the creation of safer and
more effective medications. This leap improves patient outcomes while minimiz-
ing adverse effects.
• Innovative Drug Delivery: AI’s application extends to reimagining drug delivery
methods, bolstering the safety and efficacy of new drugs. Innovative approaches
can lead to more targeted and efficient treatment.
• Accelerated Drug Discovery: The integration of AI drives the entire journey of
drug discovery, spanning from target identification to regulatory approval, with
the potential to accelerate the whole process.
In conclusion, the infusion of AI into microbial drug discovery is poised to bring
about an evolutionary shift. AI’s capacity to swiftly analyze vast datasets, optimize
drug properties, and predict efficacious compounds sets the stage for ground-
breaking advancements. Considering the trajectory of AI’s progress and implemen-
tation, the outlook for microbial drug discovery is undeniably promising. This holds
the potential to yield treatments that are safer, more effective, and readily accessible.

References

Askr H, Elgeldawi E, Aboul Ella H, Elshaier YA, Gomaa MM, Hassanien AE (2023) Deep learning
in drug discovery: an integrative review and future challenges. Artif Intell Rev 56(7):5975–6037
Blanco-Gonzalez A, Cabezon A, Seco-Gonzalez A, Conde-Torres D, Antelo-Riveiro P, Pineiro A,
Garcia-Fandino R (2023) The role of ai in drug discovery: challenges, opportunities, and
strategies. Pharmaceuticals 16(6):891
Gupta D, Choudhury A, Gupta U, Singh P, Prasad M (2021) Computational approach to clinical
diagnosis of diabetes disease: a comparative study. Multimed Tools Appl 80:1–26
Gupta M, Srivastava D, Pantola D, Gupta U (2022a) Brain tumor detection using improved Otsu’s
thresholding method and supervised learning techniques at early stage. In: Proceedings of
emerging trends and technologies on intelligent systems: ETTIS 2022. Springer Nature Singa-
pore, Singapore, pp. 271–281
Gupta U, Gupta D, Agarwal U (2022b) Analysis of randomization-based approaches for autism
spectrum disorder. In Pattern recognition and data analysis with applications. Springer Nature
Singapore, Singapore, pp. 701–713
1 The Contribution of Artificial Intelligence to Drug Discovery:. . . 23

Gupta U, Sharma S, Jyani U, Bhardwaj A, Sharma M (2022c) Sign language detection for deaf and
dumb students using deep learning: dore Idioma. In: 2022 2nd international conference on
innovative sustainable computational technologies (CISCT). IEEE, pp. 1–5
Jiménez-Luna J, Grisoni F, Weskamp N, Schneider G (2021) Artificial intelligence in drug
discovery: recent advances and future perspectives. Expert Opin Drug Discovery 16(9):
949–959
Kaushik AC, Raj U (2020) AI-driven drug discovery: a boon against COVID-19? AI Open 1:1–4
Keshavarzi Arshadi A, Webb J, Salem M, Cruz E, Calad-Thomson S, Ghadirian N et al (2020)
Artificial intelligence for COVID-19 drug discovery and vaccine development. Frontiers. Artif
Intell 65
Kim H, Kim E, Lee I, Bae B, Park M, Nam H (2020) Artificial intelligence in drug discovery: a
comprehensive review of data-driven and machine learning approaches. Biotechnol Bioprocess
Eng 25:895–930
Luukkonen S, van den Maagdenberg HW, Emmerich MT, van Westen GJ (2023) Artificial
intelligence in multi-objective drug design. Curr Opin Struct Biol 79:102537
Melo MC, Maasch JR, de la Fuente-Nunez C (2021) Accelerating antibiotic discovery through
artificial intelligence. Commun Biol 4(1):1050
Paul D, Sanap G, Shenoy S, Kalyane D, Kalia K, Tekade RK (2021) Artificial intelligence in drug
discovery and development. Drug Discov Today 26(1):80
Prasad K, Kumar V (2021) Artificial intelligence-driven drug repurposing and structural biology for
SARS-CoV-2. Curr Res Pharmacol Drug Discov 2:100042
Ren F, Ding X, Zheng M, Korzinkin M, Cai X, Zhu W et al (2023) AlphaFold accelerates artificial
intelligence powered drug discovery: efficient discovery of a novel CDK20 small molecule
inhibitor. Chem Sci 14(6):1443–1452
Sahayasheela VJ, Lankadasari MB, Dan VM, Dastager SG, Pandian GN, Sugiyama H (2022)
Artificial intelligence in microbial natural product drug discovery: current and emerging role.
Nat Prod Rep 39:2215
Sarkar C, Das B, Rawat VS, Wahlang JB, Nongpiur A, Tiewsoh I et al (2023) Artificial intelligence
and machine learning technology driven modern drug discovery and development. Int J Mol Sci
24(3):2026
Stokes JM, Yang K, Swanson K, Jin W, Cubillos-Ruiz A, Donghia NM et al (2020) A deep learning
approach to antibiotic discovery. Cell 180(4):688–702
Volkamer A, Riniker S, Nittinger E, Lanini J, Grisoni F, Evertsson E et al (2023) Machine learning
for small molecule drug discovery in academia and industry. Artif Intell Life Sci 3:100056
Zhang L, Tan J, Han D, Zhu H (2017) From machine learning to deep learning: progress in machine
intelligence for rational drug discovery. Drug Discov Today 22(11):1680–1685
Chapter 2
Prediction of Plant Disease Using Artificial
Intelligence

Manoj Ram Tammina, K. Sumana, Pavitar Parkash Singh,

T. R. Vijaya Lakshmi, and Sagar Dhanraj Pande

Abstract Plant diseases are a persistent threat to global food security due to their
ability to damage crops. They account for 20–40% of loss of global food trade every
year. The exploding global food trade, coupled with climate change, has led to the
sustainability of native plant pests in the new environment, worsening the condition.
Additionally, new plant pests and diseases continue to threaten staple crops. This
sheds light on the need for the implementation of novel techniques to diagnose plant
diseases to tackle the global food crises. Implementation of artificial intelligence
(AI)-based methods such as machine learning (ML), deep learning (DL), and
artificial neural networks can aid in overcoming such challenges by conducting
early diagnosis of plant pests and diseases. In recent years, many research investi-
gations conducted on plant disease detection using AI have offered valuable insights
for agriculturists, botanical researchers, practitioners, and industrial professionals.
The applications DL and ML methods for plant disease detection are growing
rapidly. This chapter will shed light on recent cutting-edge research in this field,
including the latest advancements involving AI-based plant disease detection. It will
also address the trials and limitations related to the usage of AI-based methods for
plant disease diagnosis.

M. R. Tammina (✉)
Innovation, Bread Financial, Columbus, OH, USA
K. Sumana
Department of Microbiology, JSS AHER, Mysuru, India
e-mail: [email protected]
P. P. Singh
Department of Management, Lovely Professional University, Phagwara, Punjab, India
e-mail: [email protected]
T. R. V. Lakshmi
Mahatma Gandhi Institute of Technology, Gandipet, Hyderabad, India
e-mail: [email protected]
S. D. Pande
School of Computer Science and Engineering, VIT-AP University, Amaravati, Andhra Pradesh,
India

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 25
A. Khamparia et al. (eds.), Microbial Data Intelligence and Computational
Techniques for Sustainable Computing, Microorganisms for Sustainability 47,
https://doi.org/10.1007/978-981-99-9621-6_2
26 M. R. Tammina et al.

Keywords Plant pest infection · Deep learning · Artificial intelligence · Neural

network

2.1 Introduction

The global starvation rate crossed 950 billion in 2022. Over 100 billion individuals
are suffering from varying types of starvation and malnutrition. However, while
growing population needs is the major attributing factor, plant diseases are one of the
greatest threats to the agricultural industry and global food security. Every year 40%
of crops are affected due to plant pests, rendering them a persistent threat to crop
health. Climatic changes, lack of agricultural responses, and sustainability of plant
pests are threatening global food security. Plant diseases caused by pathogens and
pest attacks are one of the major risk factors for global food crises. Annual loss of
potatoes, peas, maize, rice, soya beans, and tomatoes due to pathogenic infection
contributes to 10–25% of crop loss per annum. The crisis due to crop diseases is not
restricted to food security, but is also the result of considerable economic loss and
reduced agricultural yield. Roland et al. have described that crop loss is the decrease
of the crop yield in terms of quality and quantity. Qualitative crop loss assessment
involves the analysis of efficiency of current crop protection practices, assessing the
green crop yield strategies and pest protection strategies and integrated management
of plant diseases. Quantitative crop loss evaluation involves assessment of biotic and
abiotic factors. Assessing crop losses might offer a better insight into the problem
and pave the way for development of efficient identification techniques.
Agriculture is critical to guaranteeing global food security and sustaining global
economies. Plant diseases are one of the most serious threats to agricultural output
and quality. Plant illnesses produced by pathogens such as fungi, bacteria, viruses,
and other biotic agents result in significant yield losses, worse crop quality, and
higher production costs. Whole harvests can be destroyed due to crop diseases,
threatening food supply, food security, and economic stability (Jones 2009).To
avoid these hazards and preserve sustainable agricultural practices, effective plant
disease control is required. The cornerstone of such management techniques is
prompt and accurate detection of plant diseases. The timely identification of
disease-causing agents enables targeted actions such as the implementation of
relevant therapies, the selection of resistant plant types, and the modification of
agricultural practices. This not only decreases crop losses but also the need for
chemical pesticides, resulting in more sustainable and economically viable agricul-
tural methods (Kamilaris and Kartakoullis 2021).
Artificial intelligence (AI) has created disruptive opportunities in a variety of
industries, including agriculture. The algorithms used for machine learning like
supervised learning methods and models fueled by data that may acquire knowledge
from trends and then make predictions or judgments are examples of AI. In the field
of plant disease diagnostics, AI has emerged as a game-changing technology capable
of supplementing and even outperforming traditional approaches (Savary et al.
2019).
2 Prediction of Plant Disease Using Artificial Intelligence 27

AI provides a distinct edge in dealing with the complex and diversified character
of plant diseases. Convolutional neural networks, also called neural networks
(CNNs), for example, can analyze plant photos to identify visual signals linked to
illnesses and have substantially improved image-based diagnosis. These artificial
intelligence models can quickly scan and analyze hundreds of photos, identifying
small differences in leaf colour, texture, and form that may signal disease prevalence
(Singh and Shrivastava 2018). Furthermore, AI may combine data from additional
sources, including sensor tracking environmental conditions, to offer a comprehen-
sive evaluation of illness risks.
AI has the capacity to predict disease and identify it early on. AI models can
forecast disease outbreaks with surprising accuracy by using past data, weather
trends, and other pertinent variables. Farmers should take preventive actions ahead
of time, optimizing resource allocation and minimizing crop losses. In essence,
artificial intelligence (AI) provides farmers and agricultural players with practical
knowledge that leads to better decision-making and disease control tactics (Elad and
Pertot 2014).
The goal of this chapter is to describe how AI is changing detection of plant
diseases in the context of agriculture and food security. This chapter seeks to give a
full picture of how technology might transform the agricultural environment by
looking into the promise and problems of incorporating AI approaches into disease
detection practices. This chapter’s focus includes many aspects of AI-driven plant
infection diagnostics. It will cover the principles of plant disease diagnostics,
emphasizing the variety of pathogens including their impact on crops. Traditional
diagnostic procedures will be explored, emphasizing their limits and the need for
novel alternatives. The focus will then move to the development of AI and its
components, specifically on how AI approaches may revolutionise illness detection
by improving precision, rapidity, and portability.
Further sections will shed light on the process of data gathering and preparation,
emphasizing the need for reliable information for training strong AI models. They
will dig into the complexities of constructing and educating AI models, shedding
light on algorithm selection and feature extraction subtleties. The chapter will also
go into depth on image-based and sensor data-based illness diagnostic methodolo-
gies, as well as successful examples and the influence of AI on continuous surveil-
lance and disease prediction.
In addition, the chapter will go through the difficulties of installing AI-based
detection techniques in the field, taking into account variables like data protection,
connection, and farmer approval. Ethical and societal ramifications will be discussed
as well, emphasizing the importance of responsible AI usage in agriculture. The
chapter will conclude by looking ahead to future directions and advances in this area,
reviewing new trends, and imagining how AI will revolutionize precision farming
and green agricultural practices.
Finally, this chapter aims to shed light on the symbiotic link that exists between
plant disease detection, agriculture, and AI. It intends to encourage more study,
cooperation, and implementation of AI technologies for the improvement of global
28 M. R. Tammina et al.

food security and agricultural practices by revealing the revolutionary potential of AI

in disease control.

2.2 Fundamentals of Plant Disease Diagnosis

Plant diseases produced by pathogens such as fungi, microbes, viruses, and other
biological agents represent a substantial danger to global food and agricultural
production. Proper plant disease identification is critical for successful disease
management and ensuring optimal crop production and quality (Kamilaris et al.
2021). This section delves into the core ideas of plant illness diagnosis, including
different kinds of plant illnesses, their symptoms, standard diagnostic procedures,
and the demand for novel methods.

2.2.1 Plant Disease Types

Plant diseases include a wide spectrum of problems that affect plant health and
production. Plant diseases are broadly classified into infectious and non-infectious
based on the causal agent. Living organisms such as fungi, viruses, microbes, and
nematodes cause infectious diseases, but non-living causes such as nutritional
deficits, environmental pressures, and chemical imbalances cause non-infectious
diseases (Agrios 2005). Biotic diseases, especially those caused by microorganisms,
have received a great deal of attention because of their potential to inflict massive
production losses (Savary et al. 2019).

2.2.2 Symptoms and Signs

Plant disease symptoms present as apparent shifts in plant physical appearance,

structure, and function. Leaf withering, discoloration, retarded development, lesions,
and deformation are among the symptoms (Lucas 2008). Plant illnesses may also
create visual indicators such as fungal spores, bacterial slime, or viral inclusions.
These symptoms and indications act as markers of illness existence, allowing for
precise diagnosis and treatment.

2.2.3 Conventional Diagnostic Techniques

Visual inspection, microscopic examination, and laboratory testing are the mainstays
of traditional plant disease diagnosis techniques. Visual inspection entails close
2 Prediction of Plant Disease Using Artificial Intelligence 29

examination of plant symptoms and warning signals, sometimes requiring skilled

individuals to distinguish among diseases with similar symptoms (Schumann and
D’Arcy 2010). Microscopic analysis entails looking at spores, mycelium, and other
types of microscopic structures, particularly in the case of fungal illnesses. Enzyme-
linked immunosorbent assays (ELISAs) and polymerase chain reactions (PCR) are
examples of laboratory tests used to identify certain infections.

2.2.4 Limitations of Traditional Methods

Traditional methods have helped diagnose plant diseases, but they have drawbacks
as well. Windstam and Schmale (2018) revealed that visual analysis may be
subjective and greatly depends on the observer’s skill. Pathogen detection may be
time-consuming and necessitates specific tools and expertise for microscopic inves-
tigation.Additionally, these techniques might not be appropriate for quick, wide-
ranging disease screening, which is necessary to stop massive epidemics.

2.2.5 Need for AI-driven Innovative Methods

The problems posed by changing diseases, shifting climatic circumstances, and

international trade call for the development of novel, effective diagnostic methods.
Here, artificial intelligence (AI) starts to influence the game. By overcoming the
drawbacks of current techniques, AI has a chance to transform the detection of plant
diseases. Kamilaris and Prenafeta-Boldú (2018a) reported that convolutional neural
networks (CNNs), a type of deep learning model, are particularly good at evaluating
vast amounts of information and spotting complicated patterns. These algorithms
can analyze tens of thousands of plant photos, picking up on minute variations in
texture, hue, and form that may be signs of illness. To give a comprehensive
evaluation of illness risks, AI may also combine data from additional sources,
particularly networks of sensors monitoring environmental variables. Accuracy,
rapidity, scalability, and flexibility to shifting disease dynamics are promises of
plant disease detection powered by AI methods (Das et al. 2020).
The principles of plant diagnosis of diseases cover the many disease types, their
signs and symptoms, conventional diagnostic techniques, and the drawbacks of
these procedures. Innovative solutions are necessary as the agriculture industry
struggles to feed a growing global population (Kamilaris and Prenafeta-Boldú
2018b). An interesting opportunity to improve our capacity to quickly and reliably
detect illnesses is the use of AI in plant disease diagnostics. This will enable farmers
to make better choices regarding managing plant health more successfully.
30 M. R. Tammina et al.

2.3 Role of Artificial Intelligence in Plant Disease Diagnosis

One of the most exciting developments in recent years is the use of the application of
AI in plant disease diagnostics. The convergence of technology and agricultural
production has produced transformational improvements (Hughes and Salathé
2015). As plant pathogens continue to pose a threat to agricultural sustainability
and global food security, AI is emerging as a formidable tool with the potential to
completely transform disease identification, prediction, and control. This section
explains the critical significance of AI in diagnosing plant diseases while highlight-
ing its advantages, drawbacks, and practical applications (Mohanty et al. 2016).

2.3.1 AI: Changing the Diagnosis of Plant Disease

Artificial intelligence refers to a variety of methods that give computers the ability to
learn, reason, solve problems, and make decisions—tasks that traditionally require
human intellect. Artificial intelligence (AI) techniques like machine learning and
deep learning have excelled in diagnosing plant diseases. These tools analyze
enormous volumes of data, uncovering trends and connections that may elude
traditional methods (Mohanty et al. 2016).

2.3.2 Disease Diagnosis Based on Images

Image analysis is one of the most impressive ways in which AI is being used to
diagnose plant diseases. Convolutional neural networks (CNNs), a type of deep
learning algorithm, are particularly good at processing visual input. In this situation,
they can carefully examine plant photos for minute changes in hue, texture, and form
that indicate the presence of diseases. AI models are able to classify the severity of
illness as well as detect diseases. With the use of this computerized picture analysis,
illness detection for farmers may be done more quickly and accurately (Picon et al.
2020).

2.3.3 Sensor Data and Disease Forecasting

AI goes beyond imaging to help identify plant diseases. Sensors placed in farming
areas can gather information on environmental factors including temperature,
humidity, and moisture levels in the soil. These data may be processed by AI
systems, which correlate variations with the prevalence of disease. AI algorithms
can anticipate disease outbreaks remarkably accurately by examining historical
2 Prediction of Plant Disease Using Artificial Intelligence 31

trends and real-time inputs. Farmers may take preventative actions thanks to this
proactive strategy, decreasing crop losses and maximizing resource allocation (Picon
et al. 2020).

2.4 AI’s Advantages in Plant Disease Diagnosis

Beyond accuracy and speed, the use of AI for plant disease diagnostics offers a
variety of advantages.

2.4.1 Early Disease Detection and Prevention

AI-powered systems may identify illness in the very early stages, allowing for
prompt interventions and halting large-scale epidemics by implementing algorithms
(LeCun et al. 2015).

2.4.2 Accuracy and Scalability

To improve illness prediction and management, AI models examine data with high
accuracy and can be scaled to study enormous datasets.

2.4.3 Reduced Dependence on Chemicals

Early illness identification reduces the need for overuse of pesticides, encouraging
ecologically friendly and sustainable farming methods.

2.4.4 Improved Decision-making Capability

AI-generated insights enable farmers to make educated decisions about agriculture

and disease management measures.
32 M. R. Tammina et al.

2.5 Barriers to Implementing AI Techniques in Plant

Disease Diagnosis

Although AI has great potential for diagnosing plant diseases, there are a number of
issues that need to be resolved, described by Das et al. (2020).

2.5.1 Data Accuracy

For training, AI models need access to extensive, diversified, and reliable datasets.
For a model to work well, accurate data must be made available.

2.5.2 Model Interpretability

Deep learning models’ “black-box” nature might make it difficult to grasp how
choices are made. Transparency of the model must be maintained.

2.5.3 Infrastructure and Accessibility

It’s possible that many farmers lack the financial means or technological know-how
to use AI systems efficiently.

2.6 Current Trends in AI Involvement in Plant Disease

Diagnostics

Das et al. (2020) described that AI-based plant disease detection is already showing
promise in the following areas.
Cassava Disease Detection: AI models have been created in Sub-Saharan Africa
to identify cassava illnesses from leaf photos, assisting farmers in prompt disease
control.
Monitoring Grapevine Disease: Drones powered by AI and fitted with cameras
take high-resolution pictures of vineyards, allowing early identification and targeted
treatment of grapevine disease.
Smartphone Applications: Simple apps for smartphones have been created that
enable farmers to take pictures of plant symptoms and instantly diagnose diseases
and provide treatment suggestions.
2 Prediction of Plant Disease Using Artificial Intelligence 33

The use of artificial intelligence in the identification of plant diseases represents a

turning point for contemporary agriculture. The use of analyzing images, data
collected from sensors, and predictive modeling by AI provides farmers with
previously unheard of capabilities for effectively and sustainably battling plant
diseases. However, overcoming obstacles, advancing accessibility, and encouraging
partnerships between AI professionals, agricultural scientists, and farmers are all
necessary for a successful adoption. The path from AI-driven insights to real-world
effect contains the potential to secure the world’s food supply, ensure the success of
agricultural economies, and promote a sustainable future.

2.7 Data Collection and Pre-processing

Artificial intelligence (AI) has been shown to be a game-changer in a number of

industries, and its use in plant disease detection has enormous potential to transform
agriculture. However, the preparation and quality of the data are key factors in
whether AI models succeed in this situation. In-depth discussion of data collecting
and initial processing for artificial intelligence-based plant disease detection is
provided in this section, which also emphasizes the significance of the integrity of
data sources, pre-processing methods, and problems in achieving trustworthy and
accurate findings.

2.8 The Significance of Data Quality

Data quality is the basis for all AI projects. High-quality data is essential for training
models to reliably discriminate between healthy and sick plants in the context of
plant disease detection. The capacity of the model to generalize and make wise
evaluations is improved by having clean, well-labeled, and diversified datasets.
Biased or inaccurate data might provide inaccurate forecasts and unreliable results
(Picon et al. 2020).

2.9 Data Sources

Image Databases: Images are used to record illness symptoms and signs. These
photographs encompass a range of settings, lighting types, and plant species,
ensuring the model’s durability and flexibility (Mohanty et al. 2016).
Sensor Networks: Information from environmental sensors, such as relative
humidity, temperature, and soil moisture, is used to improve disease prediction
models. In turn, prediction accuracy is improved by linking these environmental
elements to illness incidence (Mohanty et al. 2016).
34 M. R. Tammina et al.

Historical Information: The construction of disease prediction algorithms is

aided by historical information on outbreaks of disease, weather trends, and agri-
cultural practices. Disease risks can be predicted with the use of patterns and trends
derived from previous data (Mohanty et al. 2016).

2.10 Techniques for Pre-processing Data

Data cleaning is necessary since raw data frequently contains noise, inaccuracies,
and discrepancies. To maintain the integrity of the dataset, duplicates must be
eliminated, errors must be fixed, and missing values must be filled in.

2.10.1 Data Augmentation

By performing changes like spinning, trimming, and scaling on pre-existing photos,

augmentation techniques artificially enhance dataset size. Through this method,
overfitting is decreased, and model generalization is improved (Das et al. 2020).

2.10.2 Standardization and Normalization

By making pixel values in photos the same scale, features are guaranteed to be
comparable. Convergence of the model is hastened by standardized procedures,
which focuses data near zero with a unit standard deviation.

2.10.3 Feature Extraction

Simplifying the complexity of data and enabling more efficient model learning are
two benefits of extracting relevant features from pictures, such as patterns or color
histograms.

2.10.4 Balancing Classes

Class balance is achieved by either oversampling minority classes or undersampling

dominant classes.
2 Prediction of Plant Disease Using Artificial Intelligence 35

2.11 Data Collection and Pre-processing Challenges

2.11.1 Labeling Complexity

Accurately classifying pictures as healthy or ill requires knowledge. The symptoms

of several disorders may overlap, making accurate identification difficult (Das et al.
2020).

2.11.2 Imbalanced Data

Imbalanced classes result from the fact that sick samples frequently exceed healthy
ones by a large margin. This imbalance may affect how well the model can identify
illnesses (Das et al. 2020).

2.11.3 Data Access and Privacy

Disseminating agricultural data presents privacy issues. Due to proprietary limita-

tions and data ownership, accessing a variety of datasets may be difficult (Das et al.
2020).

2.11.4 Environmental Variability

Changes in the backdrop, imaging, and lighting can add noise and have an influence
on the model’s generalizability.
The steps of data gathering and pre-processing are essential in developing AI
models for diagnosis of plant diseases. The dependability and accuracy of AI fore-
casts are fueled by high-quality data. AI models may recognize disease trends and
generate accurate forecasts by utilizing resources like imaging databases, networks
of sensors, and historical records together with efficient pre-processing algorithms.
The promise of AI-based plant disease diagnosis is still transformational, even
though issues like labeling intricacy, data disparity, and confidentiality must be
addressed. Agriculture stands to gain from better disease management, improved
crop yields, and sustainable farming methods by ensuring precise data collection and
rigorous pre-processing.
36 M. R. Tammina et al.

2.12 Building and Training AI Models

2.12.1 Comparative Analysis of Choosing Appropriate AI

Algorithms for Plant Disease Diagnosis

The adoption of artificial intelligence (AI) systems for plant-borne illness diagnostics
has caused a notable revolution in the agricultural sector. The accuracy, effective-
ness, and comprehension of the illness detection models are directly influenced by
the choice of the appropriate AI algorithm. Convolutional neural networks (CNNs)
and Random Forests have emerged as popular options among the several AI
algorithms available. In the context of developing artificial intelligence models for
plant pathology diagnostics, this section examines the concerns, benefits, and limits
of these algorithms (Simonyan and Zisserman 2014).

2.12.2 Convolutional Neural Networks (CNN)

CNNs are a subset of deep learning algorithms that have completely changed the
field of image analysis. They are naturally suited for plant disease diagnosis based on
leaf photos since they excel at tasks requiring visual data. They imitate the human
visual system by extracting characteristics from pictures in a hierarchical manner.
With the use of this feature extraction capacity, they can identify subtle patterns and
textures in leaves that point to the presence of illness (He et al. 2016). They
automatically detect spatial hierarchies in pictures by identifying local characteristics
and how they are combined to form more complex structures. This is crucial to
distinguish between healthy and unhealthy plant tissues. As a starting point, CNNs
may use pre-trained models on big datasets like ImageNet. Even with few photos of
plant diseases, this transfer learning speeds up model convergence. CNNs have the
ability to identify illnesses at different dimensions of leaves and resolutions because
of their scale-invariant characteristics. Once these are learned, CNN-based models
can identify diseases in real time, allowing for quick actions to stop the spread of the
disease.

2.12.3 Random Forests

Random Forests is an ensemble learning method that combines multiple decision

trees to make predictions. It has been widely used in various domains, including
agriculture, due to its interpretability and robustness. Itprovides insights into feature
importance, indicating which visual attributes contribute most to disease detection.
This information can aid experts in understanding disease symptoms. By aggregat-
ing predictions from multiple decision trees, it reduces the risk of overfitting and
2 Prediction of Plant Disease Using Artificial Intelligence 37

improves generalization. It can handle missing data effectively, a crucial advantage

in cases where datasets are not uniform.These algorithms can capture non-linear
relationships between image features and disease presence, enabling them to identify
complex patterns. It offers interpretable results, making it easier for domain experts
to comprehend the decision-making process, and allowing them to recognize intri-
cate patterns (Zhou et al. 2016).

2.12.4 Factors Affecting Algorithm Selection

The following factors strongly impair the selection of algorithm (Singh and
Shrivastava 2018).

2.12.4.1 Data Accessibility

Due to their powerful feature extraction abilities, CNNs can be useful when a
significant amount of labeled data is accessible.

2.12.4.2 Interpretability

Random Forests offer insights into feature significance and model decision-making
if interpretability is a goal.

2.12.4.3 Computing Power

CNN training requires a large amount of computing power, which not all users may
be able to afford. Random Forests use less computing power.

2.12.4.4 Real-time Requirements

CNN’s prediction accuracy might be a determining factor if real-time illness diag-

nosis is required (Singh and Shrivastava 2018).
The choice of an AI system for diagnosing plant diseases relies on the applica-
tion’s goals, resources, and unique needs. When it comes to image-based analysis,
CNNs stand out because of their outstanding precision and real-time capabilities,
whereas Random Forests offer resilience and interpretability. The eventual decision
may include a trade-off between precision, comprehension, and computational
needs. Advancements in hybrid models and algorithm design may help to further
hone disease diagnostic techniques as the area of AI develops, enabling sustainable
38 M. R. Tammina et al.

and effective agriculture. The following figure describes the fundamental structure of
plant disease identification (Fig. 2.1).
Building precise plant disease detection algorithms requires a well-annotated
dataset as a base. This entails taking pictures of plants, classifying them as healthy
or sick, and dividing the dataset into subgroups for training, validation, and testing.
There are two types of commonly used data: image based and sensor based. The
training data prepares the model for use, the validation data helps adjust
hyperparameters, and the testing data assesses the model’s efficacy on samples
that have not yet been seen.

2.13 Image-based Plant Disease Diagnosis

PlantVillage is one of the renowned disease detection methods based on images.

This tool, created by Penn State University researchers, uses AI and crowdsourcing
to identify plant illnesses. A mobile app that is part of the system enables farmers to
post pictures of their crops. The photos are analyzed by AI algorithms to spot
probable ailments and provide therapy suggestions. Farmers in many locations
have benefited from PlantVillage’s rapid disease identification and control assis-
tance, which have boosted crop yields and decreased losses.
Jung et al. (2023) developed a plant disease detection model using deep learning
methods using 8121 healthy images of different kinds of crops. A total of 31,061
images of infected leaves included images of crops with bacterial spot either with

Fig. 2.1 Fundamental structure of plant disease identification

2 Prediction of Plant Disease Using Artificial Intelligence 39

tomato mosaic virus or Alternaria sp., and potatoes infected by Phytophthora

infestans were collected in this study. A total of 18,445 and 6481 photos of sick
and healthy leaves, respectively, were kept after image filtering for further analysis
to create a systematic AI detection model for plant disease. The image background
was removed instead of annotating for training set in order to construct models based
on the leaf’s form or an image’s lesion (Fig. 2.2).

2.13.1 Data Pre-processing and Training AI Models

Using stratified random sampling, the entire dataset was split 80:20, with training
and test data, respectively. Using the training set, the model was built and validated.
The model that had the best accuracy was chosen from CNN models that were
trained earlier with high-precision scores in classification (Jung et al. 2023). It was
important to employ more data in order to enhance the illness detection model’s
effectiveness. In order to gain extra picture data, data augmentation was done by
modifying the training dataset that was already available. To create an ideal machine
learning classification system independent of the angle, the original photos were
resized to 224 pixels by rotating. The gathered data has a huge image size and the
directions of the leaf midrib are all distinct. As data augmentation techniques,
brightness or color changes were not selected since they likely interfere with the
characteristics of disease or crop lesions. The sum was significant in step 1, since all
data were used. In order to be utilized as analysis data for computing power, the

Bell Pepper Potato Tomato

Healthy

Early Blight Early Blight Bacterial Spot

Disease

Bacterial Spot

Tomato Mosaic
Late Blight Late Blight Virus

Fig. 2.2 Image selection (Source: Jung et al. 2023)

40 M. R. Tammina et al.

picture data was amplified 18 times and rotated 20 degrees. Each image was rotated
10 degrees for steps 2 and 3 in order to multiply the data 36 times.

2.13.2 Systematic Algorithm for Image-based Disease

Diagnosis

The systematic algorithm for image-based disease diagnosis in plants has three key
steps (Fig. 2.3).
In order to simulate human detection, a systematic detection model was built with
successive submodels to identify crops, illness prevalence, and disease categoriza-
tion. The CNN analysis method, one of the deep learning analysis approaches suited
for image analysis, was used to build this model. Five pre-trained CNN models were
used to fine-tune each submodel: AlexNet, ResNet50, GoogLeNet, VGG19, and
EfficientNet. These performed well in the classification of plants and ranked first or
second in the ImageNet Large Scale Visual Recognition Challenge. In step 1, five
pre-trained CNN models were used to create a model to identify and categorize
cultivars using pictures of a whole leaf from three crops in the Solanaceae family,
independent of the presence or absence of disease. Step 2 included dividing the
number of healthy and damaged leaves into two groups. Step 3 involved taking
photos of infected leaves from step 2 to classify disease types using classification
models for specific crop diseases. Only photos of a single disease type were available
for the bell pepper, making it impossible to create a classification model that could
discriminate between different intraspecies illnesses.

2.14 Model Evaluation

Calculations were made to determine accuracy, precision, recall, and F1-score in

order to assess the classification model’s performance. Values for true positive (TP),
true negative (TN), false positive (FP), and false negative (FN) are produced after

Fig. 2.3 Algorithm of

image-based disease
detection
2 Prediction of Plant Disease Using Artificial Intelligence 41

generating a confusion matrix by comparing the model test result with the real
condition. Li and Wang demonstrated customized DenseNet-77, an artificial neural
network program for plant disease detection and classification that effectively
identified and classified the infection among tomato, apple, grapes, rice, potato,
etc. The model evaluation was determined by estimating accuracy, intersection of
union, recall, and precision. The mean accurate precision and mean intersection of
union scores for the CenterNet method were observed to be 0.99 and 0.93. The
scores of the class-wise plant disease prediction are shown in Fig. 2.4.
The intersection of union (IOU) value was higher for the DenseNet-77 than
YOLO or RetinaNet models. Parul et al. conducted a performance assessment
between two types of CNN deep learning plant disease detection models using
image-based classification with images from the PlantVillage database similar to
Jung et al. (2023).

100

99.95

99.9

99.85

99.8
Accuracy

99.75

99.7

99.65

99.6

99.55
w us
m Mo et ites
Ye ic_ pot
C ck b

E l_ h
Pe H ea eal s

To ept o L _B y
Pe ll B ach al S g

m a _B t
o m La ea t

To ato _le _M t

To to arg _M t
M Po err He hy

f
M rthe aiz n_ ew

w
er ea y
y

G G y_ f_B hy

ac ua f_B thy

ta Be l_S hy

ta at B y

ua S ber e_B lthy

ra ra e e y

m rry rry ild y

m ct _S thy
ue le R t
rry C rry eal t

To ma Spi af_ ld
G Lea ea t

Bl Bl _sp t
k k t
ra ap p ea t

_B P act ngb t

Po er_ eria alt t

as L e t

Po bea He ht

To Tom arly Spo

H h

m ria a ligh

ea
Bl App ar_ Ro
be H us

Ea He ot
ze _ H s

pe pe af h
ac ac o
O Gr Gra _M Ro

B lo h

pp act H po

R to o H ligh
le

S at te lth
e la ca

o ia rc
e e eri in
ze d H h
om y_M lth

Po Pot rly_ lth

St St wd n H alth
To be be y_M lth
at L H e

m T ato to lig

m o er Sp

llo Vir
o
ai rn e Ru

ra ra le lig

h ng lig

sh oy ry lig
t
ai w y alt

lt

to ll p
ng e e H s

at sa _S
at er co
o M mo ild

To Ba eaf eal

_L
pl B S

p at a

w w r a
ed _

e
Ap ple ple

f
e

d
Ap Ap

a T
ra
h

o
o
C

t
o
e

at o
N
he

er
ze

Sq
C

pp
ai

To
Pe
M

Classes

Fig. 2.4 Model evaluation of DenseNet-77 for detection of plant diseases

42 M. R. Tammina et al.
However, DenseNet-77 has 99% accuracy scores whereas the S-CNN model has
98.6% accuracy. The results given in Fig. 2.4 show that comparison of confidence
scores among F-CNN and S-CNN neural network models reveals the higher confi-
dence scores of S-CNN models over F-CNN. The mean IOU score of S-CNN
models is 0.93, which is closer to the DenseNet-77 model. This infers that CNN
models are better for image-based plant disease detection than other machine
learning models.

2.15 Sensor Data-based Disease Diagnosis

The research study conducted by Molin and Kvarnheden (2010) presents a novel
method for controlling apple scab disease using a wireless sensor network (WSN)
and a decision support system.The goal of the project is to create a decision support
system that forecasts apple scab primary infection periods using wireless sensor
networks, allowing for prompt and precise disease control tactics. To track environ-
mental factors important for the emergence of apple scab, the researchers set up a
wireless sensor network in an apple orchard. This network gathered information on
leaf moisture and temperature, two important variables affecting the spread of
disease. The gathered information was subsequently included in a decision support
system that offered forecasts for the times of initial infection. This study illustrates
the effective use of wireless sensor networks and decision support systems in the
management of apple scab disease. The technique enables farmers to make educated
decisions and apply disease control measures at important time points by combining
real-time data collecting and predictive modeling. This technologically advanced
solution improves disease control tactics, optimizes resource consumption, and adds
to the long-term viability of apple production.The study highlights the potential of
technology, such as wireless sensor networks and decision support systems, to
transform disease control techniques in agriculture. It shows how data-driven tech-
niques may help farmers tackle illnesses more effectively, ensuring food security and
agricultural sustainability in the long run.

2.16 Future Directions and Innovations

The field of plant disease diagnostics has a great deal of potential to be transformed
by the quick development of artificial neural networks (AI) technology. Emerging
trends are influencing the future environment of AI-driven illness diagnosis as
innovation continues to advance. This section examines these developments, the
incorporation of AI with robots and autonomous systems, and the significant influ-
ence AI is expected to have on sustainable agricultural methods and precision
agriculture (Guidi and Salgado 2018).
2 Prediction of Plant Disease Using Artificial Intelligence 43

2.16.1 Explicit AI for Interpretability

There is an increasing need for openness and interpretability as AI models become

more complicated. Researchers are concentrating on creating methods that shed light
on how AI models make decisions. Regions in pictures that are most important for
the model’s predictions are highlighted using explainable AI techniques like atten-
tion processes and saliency maps. In addition to increasing user confidence, this
interpretable AI also makes it possible for domain specialists to comprehend the
justifications for illness classifications. Therefore, explainable AI eliminates the
disconnect between machine learning algorithms and human knowledge, increasing
the technology’s usability and influence (Ge and Li 2020).

2.16.2 Fusion of Multi-modal Data

Numerous elements, such as the environment, genetic data, and microbial interac-
tions, have an impact on plant health. Integrating data from numerous sources to
provide a full picture of plant health is one of the newest developments. AI models
can offer a more comprehensive picture of disease dynamics by fusing visual data
with environmental information from monitoring devices and genetic data. The
accuracy and resilience of illness detection models are improved by the merging
of multi-modal data, allowing for more accurate predictions and treatment advice.

2.16.3 Limited Transfer Learning with Data

Large-scale data labeling can be time- and resource-intensive when used to train AI
models. The use of pre-trained models on larger datasets and their fine-tuning for
plant disease diagnosis are emerging developments in transfer learning. Because less
detailed tagged plant disease photos are required with this method, AI-driven
diagnosis is more readily available to small-scale farmers and areas with scarce
data resources. Transfer learning expedites the convergence of models and equips
local communities to reap the rewards of AI technology (Tadesse et al. 2020).

2.16.4 Active Learning Techniques

Emerging trends call for the use of active learning techniques to enhance the learning
process. By training the model’s focus on the difficult situations that most contribute
to its learning, active learning entails choosing the most instructive examples for
44 M. R. Tammina et al.

labeling. By streamlining the training procedure, this iterative method decreases the
overall annotation work needed to create reliable disease detection models (Sankaran
et al. 2010). AI models are more effective and flexible to changing illness patterns
thanks to active learning methodologies.

2.16.5 Distributed Learning for Decentralized Data

Federated learning seems to be a potential development in situations where confi-

dentiality of information is an issue. This strategy keeps data local and enables the
training of AI models across several devices or servers (Yang et al. 2019). Data
privacy is maintained since each device adds to the algorithm’s learning without
transferring raw data. Federated learning is especially important for diagnosing plant
diseases because it allows farmers to work together to create reliable models without
jeopardizing private data (Torres-Sánchez et al. 2018).
The future of agriculture will be shaped by the growing trends in AI for diagnos-
ing plant diseases, which will improve disease diagnostic precision, accessibility,
and sustainability. Transfer learning democratizes AI adoption, explainable AI
improves transparency, and multi-modal data fusion offers a full picture of plant
health. Optimizing model performance and addressing data privacy issues are both
goals of active learning and federated learning systems. The potential of AI-driven
disease detection in agriculture is enormous as these developments converge, offer-
ing more robust crops, fewer losses, and a changed agricultural environment.

2.17 Integration of AI for Plant Disease Diagnosis Using

Robotics, Drones, and Automated Farm Equipment

The landscape of plant disease detection is changing, as are agricultural practices, as

a result of the combination of artificial intelligence (AI) and robots, drones, and
autonomous farm equipment. Real-time, effective, and accurate solutions for disease
diagnosis and management are provided by this confluence of technology. This
section explains how AI-driven technologies might improve plant disease detection
by fusing them with robots, drones, and autonomous machinery, emphasizing their
advantages.

2.17.1 Robotics for Precision Plant Inspection

Robots powered by AI that have sensors and cameras can move around fields and
scan plants for disease signs with unmatched precision. These autonomously moving
2 Prediction of Plant Disease Using Artificial Intelligence 45

robots cover enormous regions while collecting data and high-resolution photos for
further analysis by AI systems. Real-time illness detection is enabled through the
integration of AI and robotics.

2.17.2 Aerial Surveillance Using Drones

A bird’s-eye view of agricultural fields may be obtained using drones that are fitted
with specially designed cameras and sensors. The picture is processed by AI
algorithms to find stress patterns, nutritional shortages, and illness signs. Drones
quickly cover broad regions, enabling early detection and specialized solutions.
Drones and AI working together improve disease surveillance effectiveness while
reducing resource waste (Torres-Sánchez et al. 2018).

2.17.3 Autonomous Farm Machinery for Personalized Care

Sprayers and cultivators powered by AI can accurately target locations where

disease outbreaks are occurring. These devices use treatments only when necessary,
by evaluating information gathered by sensors and imaging systems, which mini-
mizes the use of chemicals and their negative effects on the environment. This
connection supports sustainable agricultural methods and improves resource effi-
ciency (Torres-Sánchez et al. 2018).

2.17.4 Analyzing Real-time Data

Robotic systems, drones, and autonomous equipment continuously process data

using AI algorithms. Instantaneous insights about disease prevalence, stressors,
and plant health are provided by the analytics. Farmers are able to make quick
decisions thanks to this real-time information, which successfully directs disease
management tactics.

2.17.5 Improved Disease Surveillance

Continuous disease monitoring is made possible throughout the growing season by

the integration of AI with robots, drones, and autonomous equipment. This preven-
tative strategy helps with early diagnosis, enables prompt actions, and lessens the
requirement for heavy pesticide applications.An important step toward more effec-
tive and long-lasting plant disease diagnostics is the integration of AI with robots,
46 M. R. Tammina et al.

drones, and autonomous agricultural equipment. These innovations enable real-time

monitoring of fields, fine-grained plant inspection, and targeted treatment. Agricul-
ture is advancing toward a day where disease control is accurate, timely, and
ecologically responsible by using the potential of AI-driven solutions.

2.18 AI Revolutionizing Sustainable Farming Methods

and Precision Agriculture

Particularly in the domain of precision agriculture and sustainable agricultural

methods, the application of artificial intelligence (AI) to agriculture has the potential
to revolutionize the industry. AI-driven technologies have the potential to funda-
mentally alter how farmers manage their crops, make the most use of their resources,
and protect the environment. This section explores how AI has the potential to
transform precision agriculture and advance sustainability.With the help of AI,
farmers can apply resources like water, fertilizer, and pesticides more effectively.
Artificial intelligence (AI) algorithms develop specialized irrigation schedules and
fertilizer application strategies by examining information collected by sensors,
satellites, and drones. Through precise resource management, waste is reduced,
resources are conserved, and agricultural production is increased.
Early warning systems for plant epidemics are provided by AI-driven disease
detection algorithms, IoT devices, and drones. With the help of these technologies,
quick action may be taken to prevent loss of crops and the requirement for heavy
pesticide use. AI improves disease prediction accuracy, which supports resilient
farming techniques. Huge datasets are analyzed by AI algorithms to produce insights
that may be used to make decisions. Planting timetables, crop rotation plans, and the
best times to harvest are all influenced by these observations. Data-driven decisions
increase yields, reduce losses, and support environmentally friendly agricultural
methods. AI-based image analysis systems are able to recognize and distinguish
between weeds and crops. This makes it possible to apply herbicides in a targeted
manner, using fewer chemicals and having a less negative environmental impact.
Precision weed control encourages farming that is sustainable and protects crop
health. AI-enabled sensors keep an eye on the moisture content of the soil, as well as
aspects of the soil including its moisture content, pH, and nutrient levels. By
analyzing this data, farmers may better target their fertilization programs to the
unique requirements of their soils, avoiding overfertilization and leaching. This
strategy minimizes environmental contamination while preserving the health of the
soil. Based on the surrounding environment, AI systems can forecast insect out-
breaks. With the use of this knowledge, farmers may manage pests by taking
preventative steps and using natural predators rather than artificial pesticides, pre-
serving the ecological balance. AI has enormous potential to transform sustainable
agricultural methods and precision agriculture. With the use of data-driven insights
provided by AI-driven technology, farmers can manage resources effectively and
2 Prediction of Plant Disease Using Artificial Intelligence 47

identify diseases and pests. In addition to greater output and fewer losses, this change
offers a more sustainable method of farming that reduces the influence on the
environment. AI’s integration with agriculture provides a route to a future where
global food production is more robust, effective, and ecologically responsible.

2.19 Conclusion

AI-based plant disease detection is more than just a technological achievement; it is a

driving force behind agricultural revolution. It enables us to rethink how we
approach disease concerns, enables farmers to make educated decisions, and culti-
vates a more sustainable and resilient agricultural future. We can enhance AI’s
influence, safeguard our food systems, and support the emergence of a wealthier
agricultural world by embracing its promise and encouraging collaboration. The
identification of diseases constitutes a quantum leap ahead. The trip through the
investigation of AI’s function in this critical sector brings us to a resounding
conclusion: AI-based plant disease detection has the potential to transform how we
perceive, manage, and safeguard our agricultural systems. AI’s capacity to evaluate
massive volumes of data—from photos to sensor readings—reveals hidden patterns
and connections that the human eye may overlook. AI learns from previous data and
improves its accuracy over time using machine learning techniques. This, in turn,
improves disease detection and forecasting, resulting in more proactive agricultural
management. One of the most important characteristics of AI-based plant disease
detection is its capacity to detect diseases early.

References

Agrios GN (2005) Plant pathology, 5th edn. Academic Press

Das A, Pradhan B, Jha AK (2020) Integrating artificial intelligence with IoT in precision agriculture
for sustainable crop production: a review. Comput Electron Agric 173:105370
Elad Y, Pertot I (2014) Climate change impacts on plant pathogens and plant diseases. J Crop
Improv 28(1):99–139
Ge D, Li J (2020) A review on applications of deep learning in plant disease detection. Comput
Electron Agric 177:105612
Guidi G, Salgado R (2018) Robotics and artificial intelligence in agriculture: current applications
and future challenges. Agroecol Sustain Food Syst 42(7):702–722
He K, Zhang X, Ren S, Sun J (2016) Deep residual learning for image recognition. In Proceedings
of the IEEE conference on computer vision and pattern recognition
Hughes G, Salathé M (2015) An open access repository of images on plant health to enable the
development of mobile disease diagnostics. arXiv preprint arXiv:1512.03477
Jones DG (2009) Plant pathogens and principles of plant pathology. In: Plant pathology. Wiley-
Blackwell, pp. 1–17
Jung M, Song JS, Shin AY (2023) Construction of deep learning-based disease detection model in
plants. Sci Rep 13:7331. https://doi.org/10.1038/s41598-023-34549-2
48 M. R. Tammina et al.

Kamilaris A, Kartakoullis A (2021) Applications of machine learning in precision agriculture: a

review. Precis Agric 22(3):397–425
Kamilaris A, Prenafeta-Boldú FX (2018a) Deep learning in agriculture: a survey. Comput Electron
Agric 147:70–79
Kamilaris A, Prenafeta-Boldú FX (2018b) Deep learning in agriculture: a survey. Comput Electron
Agric 147:70–90
Kamilaris A, Kartakoullis A, Prenafeta-Boldú FX (2021) A review on the practice of plant disease
detection using convolutional neural networks. Inform Process Agric 8(1):11–28
LeCun Y, Bengio Y, Hinton G (2015) Deep learning. Nature 521(7553):436–444
Lucas JA (2008) Plant pathology and plant diseases. In: Encyclopedia of life sciences. John Wiley
& Sons, pp. 1–8
Mohanty SP, Hughes DP, Salathé M (2016) Using deep learning for image-based plant disease
detection. Front Plant Sci 7:1419
Molin J, Kvarnheden A (2010) Decision support system for forecasting primary infection periods of
apple scab based on wireless sensor network. Comput Electron Agric 70(1):77–84
Picon A, Onelli E, Azzarello E, Giordano C, Masi E, Moscatiello R et al (2020) The role of data
pre-processing in plant image analysis: a case study on image-based plant phenotyping. Front
Plant Sci 11:1–17
Sankaran S, Mishra A, Ehsani R, Davis C (2010) A review of advanced techniques for detecting
plant diseases. Comput Electron Agric 72(1):1–13
Savary S, Willocquet L, Pethybridge SJ, Esker P, McRoberts N, Nelson A (2019) The global
burden of pathogens and pests on major food crops. Nat Ecol Evol 3(3):430–439
Schumann GL, D’Arcy CJ (2010) Essential plant pathology, 2nd edn. American Phytopathological
Society
Simonyan K, Zisserman A (2014) Very deep convolutional networks for large-scale image recog-
nition. arXiv preprint arXiv:1409.1556
Singh D, Shrivastava S (2018) An overview of convolutional neural networks: architectures,
applications, and challenges. ArXiv Preprint arXiv:1710.09829
Tadesse T, Mwebaze E, Vossen G (2020) Deep learning techniques for plant disease detection and
diagnosis. In: Deep learning and convolutional neural networks for medical image computing.
Springer, pp. 315–339
Torres-Sánchez J, López-Granados F, De Castro AI (2018) Assessing the accuracy of mosaics from
unmanned aerial vehicle (UAV) imagery for precision agriculture purposes in wheat. Precis
Agric 19(5):770–786
Windstam ST, Schmale DG (2018) The complexity of diagnosing plant diseases. In: Plant disease
diagnosis. Springer, pp. 1–18
Yang Q, Liu Y, Chen T, Tong Y (2019) Federated machine learning: concept and applications.
ACM Trans Intell Syst Technol (TIST) 10(2):1–19
Zhou B, Khosla A, Lapedriza À, Oliva A, Torralba A (2016) Learning deep features for discrim-
inative localization. In Proceedings of the IEEE conference on computer vision and pattern
recognition
Chapter 3
Computer Vision-based Remote Care
of Microbiological Data Analysis

Pritesh Kumar Jain and Sandeep Kumar Jain

Abstract The field of remote care of microbiological data analysis is rapidly

evolving, utilizing computer vision to automate various tasks related to microorgan-
ism identification, classification, quantification, tracking, and presence detection in
images. By employing this technology, healthcare providers can benefit from an
improved and more efficient means to analyze microbiological data, leading to
enhanced diagnosis, treatment, and prevention of infectious diseases.
This chapter delves into the challenges and prospects associated with utilizing
computer vision in remote care of microbiological data analysis. Initially, an over-
view of the fundamental principles of computer vision is presented, along with its
applicability to microbiological data analysis. Subsequently, the challenges associ-
ated with employing computer vision in this context are discussed, encompassing
factors such as image variability, microorganism complexity, and the scarcity of
extensive datasets. Furthermore, the chapter explores the potential of computer
vision in enhancing the diagnosis, treatment, and prevention of infectious diseases.
To conclude, the chapter examines the future of computer vision-based remote
care of microbiological data analysis. It contends that this technology holds immense
potential in revolutionizing the diagnostic and treatment approaches utilized by
healthcare providers in relation to infectious diseases.

Keywords Microbial · Computer vision · Healthcare · Diseases · Microorganism

3.1 Introduction

Microbiological data analysis plays a crucial role in comprehending the world of

microorganisms, spanning various domains such as health, environment, and indus-
try. It unveils the intricate microbial diversity, behaviors, and functions. Within the

P. K. Jain (✉) · S. K. Jain

Department of Computer Science and Engineering, Shri Vaishnav Vidyapeeth
Vishwavidyalaya, Indore, Madhya Pradesh, India
e-mail: [email protected]; [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 49
A. Khamparia et al. (eds.), Microbial Data Intelligence and Computational
Techniques for Sustainable Computing, Microorganisms for Sustainability 47,
https://doi.org/10.1007/978-981-99-9621-6_3
50 P. K. Jain and S. K. Jain

healthcare sector, it aids in the diagnosis of diseases and tracking of epidemics, and
contributes to the development of therapeutic interventions. Furthermore, environ-
mental monitoring employs this analysis to assess the quality of air, water, and soil,
thereby facilitating the detection of pollution and conservation efforts. The realm of
food safety greatly relies on this analysis for the detection of hazards and ensuring
quality control. Industries also make use of microbiological data analysis in areas
such as bioproduction, agriculture, and bioremediation. Additionally, research and
predictive modeling provide valuable insights into microbial genetics and behavior,
ultimately contributing to product safety and playing a role in the field of biological
engineering. As technology evolves, microbiological data analysis continues to drive
scientific exploration.

3.2 The Role of Computer Vision

The remote management of microbiological data analysis, commonly known as

digital microbiology or microbiome analysis, is undeniably a swiftly developing
domain that utilizes computer vision and other advanced technologies to mechanize
various facets of microbiological research (Zhang et al. 2022). This realm holds the
potential to substantially augment the swiftness and precision of microbiological
data analysis, rendering it more accessible and efficient. The following are a few
pivotal facets of this realm:

3.2.1 Automatization of Microorganism Identification

Computer vision algorithms (Zhang et al. 2022) can be educated to recognize and
discern diverse categories of microorganisms in images, comprising bacteria,
viruses, fungi, and protozoa. This mechanization has the capacity to economize a
significant amount of time in manual identification procedures.

3.2.2 Classification and Taxonomy

Automated systems have the capability to categorize microorganisms into distinct

taxonomic groups based on their morphological characteristics or genetic indicators.
This can facilitate the comprehension of the multitude of microorganisms that exist
within a particular specimen.
3 Computer Vision-based Remote Care of Microbiological Data Analysis 51

3.2.3 Quantification

Computer vision can be employed to quantify the abundance of microorganisms

within a given sample. This proves to be particularly advantageous in the realm of
ecological research, as well as in the monitoring of microbial populations in diverse
environments.

3.2.4 Tracking and Behavior Analysis

The tracking and analysis of the movement and behavior of microorganisms over a
period of time is of utmost importance in the field of microbiology. Computer vision
is capable of automating the tracking of individual cells or populations within a
dynamic setting.

3.2.5 Presence/Absence Detection

Automated systems possess the ability to rapidly ascertain the presence or absence of
specific microorganisms within a sample. This proves to be highly valuable in the
domains of medical diagnostics, environmental surveillance, and food safety testing.

3.2.6 High-throughput Screening

One of the primary merits of employing computer vision in the analysis of micro-
biological data lies in its capacity to efficiently process a large quantity of samples
within a short span of time, thereby enabling high-throughput screening and
analysis.

3.2.7 Integration with Other Technologies

The integration of computer vision (Singh et al. 2020; Chen et al. 2023) with other
technologies, such as genomics, metagenomics, and artificial intelligence, is a
common practice in the field of digital microbiology. This approach aims to offer
a comprehensive comprehension of microbial communities and their functionalities.
52 P. K. Jain and S. K. Jain

3.2.8 Remote Monitoring

The notion of “remote care” denotes the capacity of this technology to be employed
in distant locations, such as field studies or areas with limited laboratory access.
Consequently, it becomes feasible to monitor microbial ecosystems in real time.

3.2.9 Applications

Diverse domains benefit from the applications of this technology, encompassing

healthcare (diagnosis and treatment monitoring), agriculture (soil and plant
microbiomes), environmental science (water quality monitoring), and food safety
(contaminant detection) (Dange et al. 2023).
Challenges encountered within this field encompass the assurance of accuracy
and reliability of computer vision models, management of varied and intricate
microbial communities, as well as the ethical and privacy concerns associated with
data collection and analysis. However, with technological advancements and the
availability of more data, the remote analysis of microbiological data holds tremen-
dous potential in advancing our comprehension of microorganisms and their eco-
logical functions.

3.3 Steps Required to Implement

There exists a plethora of tools and software that can be employed for the purpose of
implementing computer vision in the context of remote care for the analysis of
microbiological data.
Some of the most popular tools include (Rao et al. 2022):
OpenCV: OpenCV is a library of computer vision algorithms that is both free and
open-source. It is accessible on a multitude of platforms, including Windows,
macOS, and Linux.
TensorFlow: TensorFlow is a renowned open-source machine learning library. It
has the potential to be utilized for the training and deployment of computer vision
models.
PyTorch: PyTorch is another widely recognized open-source machine learning
library. It bears similarities to TensorFlow, yet it has been engineered to possess
enhanced flexibility and user-friendliness.
MATLAB: MATLAB is a commercial software package that is frequently
employed for scientific computing. It encompasses an array of functions tailored to
computer vision.
ImageJ: ImageJ is an image processing software that is both free and open-source.
It serves the purpose of loading, viewing, and analyzing images.
3 Computer Vision-based Remote Care of Microbiological Data Analysis 53

The selection of programming language for the execution of computer vision in

order to provide remote care for microbiological data analysis would be contingent
on the particular tool or software being utilized. Nonetheless, Python has emerged as
a prevalent language for computer vision and is compatible with the majority of
aforementioned tools and software (Hu et al. 2021).

3.4 Microbiological Data

Microbiological data is data on microorganisms, such as bacteria, viruses, fungi, and

parasites. This data can be collected from a variety of sources, such as clinical
samples (e.g., blood, urine, stool), environmental samples (e.g., water, soil, air), and
food samples.
Microbiological data can be collected using a variety of methods (Wang et al.
2020; Li et al. 2019), including:
Culture: Microorganisms are grown in a laboratory medium to allow them to
multiply. This allows for the identification and quantification of microorganisms.
Serology: Antibodies to specific microorganisms are detected in the blood or
other bodily fluids. This can be used to diagnose infections and to track the immune
response to infection.
Molecular diagnostics: Nucleic acids (DNA and RNA) from microorganisms are
detected and analyzed. This can be used to identify microorganisms, to determine
their resistance to antibiotics, and to track the spread of disease.
Microbiological data is used in a variety of ways, including:
Diagnosis and treatment of infections: Microbiological data is used to diagnose
infections and to determine the best course of treatment. For example, culture data
can be used to identify the specific bacteria causing an infection, which can then be
treated with the appropriate antibiotic.
Public health surveillance: Microbiological data is used to track the spread of
disease and to identify outbreaks. For example, data on the types of bacteria causing
pneumonia in hospitals can be used to identify and control hospital-acquired
infections.
Food safety: Microbiological data is used to ensure the safety of food. For
example, food companies test their products for harmful bacteria, such as Salmonella
and E. coli.
Research: Microbiological data is used to study the biology of microorganisms
and to develop new treatments and vaccines. For example, researchers are using
microbiological data to develop new antibiotics that are effective against drug-
resistant bacteria.
Microbiological data is essential for the diagnosis and treatment of infections,
public health surveillance, food safety, and research.
Outlined below are several steps involved in the implementation of computer
vision (Oh et al. 2018; Bohr and Memarzadeh 2020) for remote care of microbio-
logical data analysis:
54 P. K. Jain and S. K. Jain

1. Collect a dataset of images of microorganisms: Assemble a dataset comprising a

wide range of images of microorganisms so as to ensure that the computer vision
model can effectively generalize to novel images.
2. Label the images in the dataset: Label the images within the dataset by identifying
and categorizing the microorganisms present in each image.
3. Train a computer vision model on the labeled dataset: Train a computer vision
model on the annotated dataset using various machine learning algorithms,
including supervised learning, unsupervised learning, and reinforcement
learning.
4. Deploy the computer vision model to a remote server: Deploy the computer
vision model to a remote server, thereby enabling real-time analysis of microor-
ganism images.
The implementation of computer vision for remote care of microbiological data
analysis is a multifaceted endeavor. Nonetheless, the advantages offered by this
technology are substantial and it is poised to witness greater adoption in the future.

3.5 Different Algorithms to Implement Computer Vision

Certainly, a variety of algorithms exist that can be employed to execute computer

vision in the context of remote care for microbiological data analysis.
Several of the most widely utilized algorithms (Tripathi and Maktedar 2020; Tian
et al. 2020) encompass:
Convolutional neural networks (CNNs):
CNNs are a type of deep learning algorithm that is well-suited for image analysis
tasks. They are able to learn the spatial relationships between features in an image,
which makes them effective at identifying and classifying microorganisms.
Support vector machines (SVMs):
SVMs are another type of machine learning algorithm that can be used for image
analysis. They are able to learn the boundaries between different classes of images,
which makes them effective at classifying microorganisms.
Random forests:
Random forests are a type of ensemble learning algorithm that combine the
predictions of multiple decision trees. This makes them more robust to noise and
outliers than single decision trees.
Gaussian mixture models (GMMs):
GMMs are a type of probabilistic model that can be used to segment images.
They are able to identify clusters of pixels that belong to the same object, which
makes them effective at identifying microorganisms.
K-means clustering:
K-means clustering is a simple unsupervised learning algorithm that can be used
to segment images. It identifies k clusters of pixels in an image, and assigns each
pixel to the cluster that it is closest to.
3 Computer Vision-based Remote Care of Microbiological Data Analysis 55

Table 3.1 Computer vision can vary depending on the specific algorithm and application
Parameter Description
Image resolution The number of pixels per unit of length in an image. Higher-resolution
images provide more detail, which can be helpful for identifying and
classifying microorganisms
Image size The total number of pixels in an image. Larger images provide more data
for the algorithm to learn from, which can improve the accuracy
Image contrast The difference between the brightest and darkest pixels in an image.
Images with high contrast will make it easier to identify and classify
microorganisms
Image noise The random variation of pixel values in an image. Images with low noise
will be easier for the algorithm to process
Image background The area of the image that does not contain microorganisms. Images with
a simple background will make it easier for the algorithm to identify and
classify microorganisms
Image pre-processing The steps that are performed on an image before it can be analyzed. This
may include steps such as resizing, cropping, and denoising
Microorganism type The type of microorganism being analyzed. Different types of microor-
ganisms may require different parameters
Microorganism The number of microorganisms per unit area in the image. Images with a
concentration high concentration of microorganisms may require different parameters
than images with a low concentration
Microorganism The shape and size of the microorganisms. Different morphological
morphology features may require different parameters
Microorganism The environment in which the microorganisms are found. Different
environment environments may require different parameters

The selection of an algorithm will rely on the particular problem that is being
addressed. For instance, CNNs are frequently employed for tasks such as identifying
objects and dividing them into segments, whereas SVMs are frequently utilized for
tasks such as categorization.
Various metrics, such as accuracy, precision, and recall, can be employed to
compare the performance of different algorithms. Accuracy assesses the proportion
of images that are correctly classified, precision evaluates the proportion of images
that are classified as positive and are actually positive, and recall determines the
proportion of positive images that are accurately classified (Table 3.1).

3.6 Challenges and Future Directions

The utilization of computer vision-based remote care for the analysis of microbio-
logical data encounters numerous formidable limitations and obstacles. Microbio-
logical data possesses inherent complexity, consisting of intricate structures and
dynamic processes at the microscopic level. The accurate interpretation of this data
56 P. K. Jain and S. K. Jain

through computer vision algorithms necessitates the advancement of sophisticated

techniques for the processing of images.
Microorganisms exhibit a broad range of morphological diversity, encompassing
various shapes, sizes, and characteristics. The training of computer vision models to
reliably identify and differentiate between different microbial species and strains can
be challenging due to this inherent variability.
Furthermore, the quality and resolution of microbial images can present prob-
lems. In remote or field settings, the acquisition of high-quality images of microor-
ganisms may be hindered by factors such as suboptimal lighting conditions, variable
magnification, and varying quality of imaging equipment.
In addition, sample preparation and staining techniques are crucial in the field of
microbiology, but they can introduce further complexities to the analysis based on
computer vision. Staining protocols have the potential to alter microbial morphology
or introduce artifacts, thereby rendering accurate image analysis more challenging.
Moreover, the dynamic nature of microbiological processes, such as microbial
growth and movement, can pose difficulties for real-time remote analysis. The
continuous monitoring and tracking of microorganisms over time may necessitate
advanced computer vision models capable of handling dynamic data streams.

3.7 Conclusion

In conclusion, while computer vision holds tremendous potential for the remote
analysis of microbiological data, these challenges underscore the necessity for
ongoing research and development to overcome the intricacies of this field and
fully exploit the capabilities of computer vision in remote care and analysis of
microbiological data.

References

Bohr A Memarzadeh K (2020) The rise of artificial intelligence in healthcare applications. In:
Artificial intelligence in healthcare. Academic Press, pp 25–60
Dange BJ et al (2023) Grape vision: a CNN-based system for yield component analysis of grape
clusters. Int J Intell Syst Appl Eng 11(9s):239–244
Hu et al (2021) Computer vision for medical imaging: a comprehensive guide. Springer
Li et al (2019) Computer vision for remote sensing: a comprehensive guide. Springer
Oh et al (2018) Telemedicine and telehealth systems: a comprehensive guide. Springer
Rao et al (2022) Microbiological data analysis: a comprehensive guide. Springer
Tian H et al (2020) Computer vision technology in agricultural automation—a review. Inform
Process Agric 7(1):1–19
3 Computer Vision-based Remote Care of Microbiological Data Analysis 57

Tripathi MK, Maktedar DD (2020) A role of computer vision in fruits and vegetables among
various horticulture products of agriculture fields: a survey. Inform Process Agric 7(2):183–203
Wang et al (2020) Deep learning for microbiological data analysis. Academic Press
Zhang, J., Li, C., Rahaman, M. M., Yao, Y., Ma, P., Zhang, J., ... & Grzegorzek, M. (2022). A
comprehensive review of image analysis methods for microorganism counting: from classical
image processing to deep learning approaches. Artificial Intelligence Review, 1-70.
Chapter 4
A Comparative Study of Various Machine
Learning (ML) Approaches for Fake News
Detection in Web-based Applications

Mahabub Hasan Mahalat, Sushree Bibhuprada B. Priyadarshini,

Sandip Swain, Shobhit Sahoo, Atish Mohapatra, and Mangaldeep Das

Abstract The development of fake news detection and intervention methods is a

result of the exponential spread of false news and its impact on justice, democracy,
and public confidence. People’s trust in the government, news media, news stories,
social media, etc. has decreased as a result of widespread fake news. Hence, fake
news has become a crucial problem in our society. In this chapter we have presented
a comparative examination of detecting fake news. We have defined the negative
behavior of bogus news and introduced detection methods. Many of such methods
rely on identifying content and context aspects that suggest misrepresentation. We
also looked at existing datasets that have been employed to categorize bogus news.
Finally, we have suggested possible research avenues by applying various machine
learning-based classification algorithms and analyzed bogus news in terms of vari-
ous performance parameters like accuracy, precision, recall, and F1 score.

Keywords Accuracy · F1 score · Precision · Recall

4.1 Introduction

Social media has become a significant part of daily life today. Due to the exponential
growth in users, it has become an indispensable requirement. In this context,
Instagram is one of the most popular social media platforms with over 1 trillion
active users, making it one of the most used social media platforms. The major
benefit of online social media is the ability for people to communicate with one
another effortlessly. This has opened up a fresh avenue for prospective attacks like
identity theft and disinformation. According to a recent poll, there are more social
media accounts established than there are active users. This has made it more

M. H. Mahalat · S. B. B. Priyadarshini (✉) · S. Swain · S. Sahoo · A. Mohapatra · M. Das

Computer Science & Information Technology, Institute of Technical Education and Research,
Siksha ‘O’ Anusandhan Deemed to be University, Bhubaneswar, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 59
A. Khamparia et al. (eds.), Microbial Data Intelligence and Computational
Techniques for Sustainable Computing, Microorganisms for Sustainability 47,
https://doi.org/10.1007/978-981-99-9621-6_4
60 M. H. Mahalat et al.

Fig. 4.1 Major features

of data

challenging for social media companies to spot bogus accounts. Due to the preva-
lence of fraudulent information and marketing on social media, it is important to spot
these fake accounts. The truth and intent of every remark are frequently intractable
for computers alone, necessitating a collaborative effort between humans and tech-
nology (Radianti et al. 2016; Alkhodair et al. 2020).
Three major features associated with data are volume, veracity, and velocity, as
illustrated in Fig. 4.1. Hence, identifying any new data is of paramount importance.
Fake news has been circulating for several years and it will continue to do so; hence
required procedures must be applied for eliminating such fake news. Articles,
current events, posts, statements, assertions, and any other types of information
linked to public interests and organizations are all examples of news.
Such news explanations raise social and political concerns. This concept is
compatible with the majority of existing fake news studies and datasets. Although
fake news is not a new phenomenon, questions such as “Why has it become a global
topic of interest?” and “Why has it been increasingly attracting people’s attention”
are pertinent at this time. Furthermore, a large volume of false news is manufactured
and distributed via the Internet, posing a potential threat to social communities and
having a significant detrimental influence on Internet activities such as online
shopping and social networking. During election campaigns, fake news has been
blamed for increasing political polarization and partisan confrontation (Yi et al.
2003; Tapaswi et al. 2012).
Consider the COVID-19 pandemic that affected the entire world. Doctors and
scientists were battling to find a cure, despite the fact that there were numerous
therapies available on the Internet via social media, news articles, and so
on. According to these articles, taking certain medications would cure COVID-19.
People truly believed the “facts” and took the drugs without consulting a doctor,
resulting in the spread of other disorders. We don’t understand the context and jump
to conclusions without understanding if what we have read is true or untrue.
Figure 4.2 shows a scenario of news emerging from various sources (Radianti
et al. 2016; Ranjan et al. 2003; MonaDiab et al. 2004; Rouse 2018; Dua and Du
2016; Huang n.d.; Researchgate.net 2018; Researchgate.net 2014; Sirikulviriya and
Sinthupinyo 2011; Kevric et al. 2017; Parikh and Atrey 2018; Granik and Mesyura
4 A Comparative Study of Various Machine Learning (ML) Approaches for. . . 61

Fig. 4.2 News coming from various sources

2017; Gilda 2017; Jain and Kasbe 2018; Qin et al. 2018; Khanam and Ahsan 2017;
Perez-Rosas et al. 2017; Aphiwongsophon et al. 2018; Kaur et al. 2019; Looijenga
2018; Anonymous 2018; Traore et al. 2017; Khanam and Ahsan 2018; Sharma et al.
2019; Shu et al. 2017; Khanam and Agarwal 2015; Zhang et al. 2020; Ludwig and
Creation 2020; da Silva et al. 2019a, 2019b; Bovet and Makse 2019; Lazer et al.
2018; Raza and Ding 2022a, 2022b; Jain et al. 2019a, 2019b; Fan 2017; Shu et al.
2019; Rubin 2017; Kogan et al. 2019; Vosoughi et al. 2018).
This broad breadth of false news identification necessitates the use of highly
developed computational quantification and visualization technologies. Previously,
utilized algorithms such as naive Bayes, support vector machine, and random forests
were ineffective in detecting bogus news. This study compares existing news
identification approaches and enhances the effectiveness of previously used algo-
rithms to correctly recognize whether the news is false or authentic (Yi et al. 2003;
Tapaswi et al. 2012; Ranjan et al. 2003; MonaDiab et al. 2004; Rouse 2018;
Antweiler and Frank 2005; Allcott and Gentzkow 2017; Ahern and Sosyura 2014;
Mosseri 2016; Meyer 2017; Rashkin et al. 2017; Singh et al. 2017; Tacchini et al.
2017).
The most important reason for this study is that false news can be manufactured
and distributed more quickly and cheaply than traditional news media such as
newspapers and television. In today’s society, social media is an integral part of
daily life. Social media users are rapidly increasing on Instagram, one of the most
popular social media platforms, which has over a trillion active users (Rashkin et al.
2017; Singh et al. 2017; Tacchini et al. 2017; Akoglu et al. 2013; Rubin et al. 2016).
This has, in some ways, heightened awareness of potential attacks such as imper-
sonation and misinformation. The suggested study would take the fake and real news
datasets as input and determine whether the supplied input is true or false using
various types of machine learning algorithms. The input may take the shape of news
62 M. H. Mahalat et al.

items, related side information, and social situations. Given the input, we obtain one
of two labels: 0 for “real” and 1 for “fake.” The main goal is to create a model that
can determine whether the news provided as input is accurate or false (Rouse 2018;
Dua and Du 2016; Huang n.d.).
Various classifiers are normally used for classification purposes, as follows.

4.1.1 Logistic Regression

This is a classification procedure that is used to assess the likelihood of an event

occurring as a linear combination of a set of input features. Because it deals with
probability approximation, the model is best suited for binary classification. Assume
the logistic regression model calculates p for a linear combination of independent
variables in order to establish the actual class label (Radianti et al. 2016; Alkhodair
et al. 2020; Yi et al. 2003; Tapaswi et al. 2012; Ranjan et al. 2003; MonaDiab et al.
2004; Rouse 2018; Dua and Du 2016). The calculated expression can then be
written as:

1
pð x Þ = ð4:1Þ
1þ e - ðβ0 þβ1 xÞ

4.1.2 Decision Tree Classifier

This represents a supervised machine learning technique that uses a set of rules to
make judgments in the same way that humans do. Every internal node of the decision
tree assigns a condition or “test” to an attribute, and the connection is created based
on the test conditions and the outcome. Finally, after computing all characteristics,
the sheet node has a class label.

4.1.3 Random Forest Classifier

These are based on the idea of generating numerous decision tree algorithms, each of
which generates a unique output. The Random Forest Classifier uses a variety of
decision trees on different subsets of the input data to improve the dataset’s predic-
tive accuracy. The random forest employs the outcomes predicted by several deci-
sion trees. When applied to connected trees, the result will more or less resemble a
single decision tree. Furthermore, uncorrelated decision trees can be generated using
feature randomness and bootstrapping (Tapaswi et al. 2012; Ranjan et al. 2003;
MonaDiab et al. 2004; Rouse 2018; Dua and Du 2016; Huang n.d.; Researchgate.net
4 A Comparative Study of Various Machine Learning (ML) Approaches for. . . 63

2018; Researchgate.net 2014; Sirikulviriya and Sinthupinyo 2011; Kevric et al.

2017; Parikh and Atrey 2018; Granik and Mesyura 2017; Gilda 2017; Jain and
Kasbe 2018; Qin et al. 2018; Khanam and Ahsan 2017; Perez-Rosas et al. 2017;
Aphiwongsophon et al. 2018; Kaur et al. 2019; Looijenga 2018; Anonymous 2018;
Traore et al. 2017; Khanam and Ahsan 2018; Sharma et al. 2019; Shu et al. 2017;
Khanam and Agarwal 2015; Zhang et al. 2020; Ludwig and Creation 2020; da Silva
et al. 2019a, 2019b; Bovet and Makse 2019; Lazer et al. 2018; Raza and Ding
2022a).

4.1.4 Linear Support Vector Classifier (SVC)

This method is based on arranging each data item as a point in a range of dimensions
n (the number of accessible attributes), and the value of a certain property is the
number of given coordinates. The SVM algorithm uses coordinates that correspond
to the values of each feature in a set of n features to plot a data item in n-dimensional
space. The hyperplane that was developed to split the two classes is used to classify
the data. The goal of the SVM algorithm is to find the best line or decision boundary
that can divide n-dimensional space into classes, allowing us to classify new data
points fast in the future. This ideal decision is referred to as a hyperplane (Alkhodair
et al. 2020; Yi et al. 2003; Tapaswi et al. 2012; Ranjan et al. 2003; MonaDiab et al.
2004; Rouse 2018; Dua and Du 2016).

4.1.5 Multinomial Naive Bayes (NB)

This algorithm, which is employed in a variety of machine learning applications, is

based on the Bayes’ theorem and assumes that it is free of predictors. Simply
expressed, naive Bayes assumes that one function in a category has no bearing on
another. For example, an apple is categorized as such if it is red in color, has swirls,
and has a diameter of less than 3 inches. Whether these functions are dependent on
one other or on distinct functions, and even if they are dependent on each other or on
other functions, naive Bayes assumes that all of these functions have a different
proof of the apples (Researchgate.net 2018; Researchgate.net 2014; Sirikulviriya
and Sinthupinyo 2011; Kevric et al. 2017; Parikh and Atrey 2018; Granik and
Mesyura 2017; Gilda 2017).
64 M. H. Mahalat et al.

4.1.5.1 Naive Bayes Equation

ðmjnÞ = ðmjnÞ=ðmÞ ðnÞ ð4:2Þ

ðmjnÞ = ðm1jnÞ × ðm2jnÞ × . . . × ðm2jnÞ × PðzÞ ð4:3Þ

where P(m|n) is the posterior probability, P(m|n) is the likelihood, P(m) is the class
prior probability, and P(m) is the predictor prior probability.
The remainder of the chapter is organized as follows. The next section discusses
the various threads of related work done in this context. Subsequently, Sect. 4.3 goes
into more detail on this work. Section 4.4 details the performance evaluation of the
various methods. Section 4.5 concludes.

4.2 Related Work

The study required to establish whether the provided content is real was done in
(Parikh and Atrey 2018; Granik and Mesyura 2017; Gilda 2017; Jain and Kasbe
2018) by identifying various news sources and conducting the relevant analysis. The
studies made use of models created around speech features and predictive algorithms
that don’t mesh with other current techniques. Similar to this, the works shown in
(Granik and Mesyura 2017; Gilda 2017; Jain and Kasbe 2018) employ a classifier
based on naive Bayes to recognize fake news. This method produced an accuracy of
74% when evaluated with different social media records and was employed as a
software platform. The punctuation errors were disregarded, which reduced the
paper’s accuracy. Analogous to this, the studies described in (Jain and Kasbe
2018; Qin et al. 2018) computed a number of machine learning (ML) methods and
investigated predictive accuracy.
The accuracy of several predictive models, including bounded decision trees,
gradient improvement, and support vector machines, varied. Additionally, (Qin et al.
2018; Khanam and Ahsan 2017; Perez-Rosas et al. 2017; Aphiwongsophon et al.
2018; Kaur et al. 2019; Looijenga 2018) uses the naive Bayes classifier to describe
and demonstrate how to incorporate bogus information identification into various
social media sites. Through Facebook, Twitter, and other social networking sites,
they were able to obtain news. Qin et al. (2018), Khanam and Ahsan (2017), and
Perez-Rosas et al. (2017)s discuss locating erroneous information in the present
moment. They draw on a novelty-based attribute and uses Kaggle for its data.
The mean accuracy of this pattern is 74.5%. Low-quality websites and advertising
contribute to a lower resolution. We apply the methods in (Aphiwongsophon et al.
2018; Kaur et al. 2019) to locate Twitter spam senders. In this situation, models like
the decision tree, clustering, and naive Bayes algorithms are used. Spam and
fraudster identification accuracy averages 70 and 71.2%, respectively. Only a
4 A Comparative Study of Various Machine Learning (ML) Approaches for. . . 65

modest level of intermediate precision has been obtained by the models used to
differentiate between spam and non-spam.
The methodology in (Looijenga 2018; Anonymous 2018) looked at a variety of
ways to spot fake news. Only 76% of the language model is accurate. Improved
accuracy can be obtained by using a predictive model (Looijenga 2018; Anonymous
2018). used machine learning approaches to detect fake news. Three frequently
utilized methods are used in their research: support vector machines (SVM), neural
networks, and naive Bayes. The discovery of fake news detection as a predictive
analysis application was made in (Traore et al. 2017). To find fraudulent messages,
the three phases of processing, feature extraction, and classification are used. The
hybrid classification method used in this study was developed to reveal false
information. The classification procedure combines k-nearest neighbors algorithm
and random forests. The application of the proposed model is evaluated for precision
and recall. The outcomes were enhanced by up to 8% when a mixed false message
detection model was used (Khanam and Ahsan 2018; Sharma et al. 2019; Shu et al.
2017; Khanam and Agarwal 2015; Zhang et al. 2020; Ludwig and Creation 2020; da
Silva et al. 2019a, 2019b; Bovet and Makse 2019; Lazer et al. 2018; Raza and Ding
2022a, 2022b; Jain et al. 2019a, 2019b; Fan 2017; Shu et al. 2019; Rubin 2017;
Kogan et al. 2019; Vosoughi et al. 2018; Antweiler and Frank 2005; Allcott and
Gentzkow 2017; Ahern and Sosyura 2014).
The work depicted in (Shu et al. 2017; Raza and Ding 2022a; Jain et al. 2019a)
discussed fake news detection based on news content and social contexts, where a
deep neural framework is used for fake news detection. The problem statement is
identified into two unique challenges: (i) early detection of fake news and
(ii) shortage of label. A unique transformer model is designed for the detection of
fake news. The encoder and decoder blocks are used for the early detection of fake
news. An effective supervision labeling scheme is used to resolve the label shortage
issue. However, the paper doesn’t touch upon enough algorithms, for example the
classifier algorithms like logistic regression, support vector machine, etc., to carry
out fake news detection.
Similar to this, some studies (Shu et al. 2017; Jain et al. 2019a) proposed a
machine learning-based smart news system that uses the naive Bayes classifier,
SVM, and natural language processing to identify bogus news. Based on the
model used, the system in question can identify bogus news. Additionally, it offers
some pertinent news on the subject, which is quite beneficial for any user. The
prototype’s effectiveness and precision can, however, be improved to a considerable
extent. It currently has a 93.5% accuracy rate.

4.3 Work Done

Figure 4.3 depicts a high-level overview of our suggested design. We begin with
news as input in the form of a Kaggle dataset. The news data are preprocessed. Then
feature selection is carried out. Data is split into training and testing data. Later on we
66 M. H. Mahalat et al.

Fig. 4.3 Model for classification of data

Fig. 4.4 Different methods applied on datasets

employ the learning model to perform classification. The news items, social media
contexts, and side information are routed to the appropriate areas for data
preprocessing. Then, from our preprocessed dataset, we extract various language
features and choose the ones that are required. Then we divide our dataset into train
and test datasets, with 80% of the dataset used for training and 20% used for testing.
The train dataset is then utilized to train our model using different algorithms as
shown in Fig. 4.4. The model is then evaluated after it has been trained. If there is a
need to improve performance and accuracy, hyperparameters or model tuning can be
used to help. The training is then evaluated on the test dataset, which uses multiple
methods to determine whether the news is false or true.
4 A Comparative Study of Various Machine Learning (ML) Approaches for. . . 67

Fig. 4.5 A scenario of sequence of steps in proposed method

We obtained 20,800 distinct news data points from Kaggle for this study. When
developing false news detection systems, the primary goal is to achieve the best
feasible result by improving performance and accuracy. The numerous reasons why
false news exists, its characteristics and nature, who spreads it, and how fake news
spreads can be discovered through descriptive and in-depth examination. We used a
comparative analysis strategy to create our model with several algorithms, yielding
an accuracy of 96%. The project can be used to provide an ideal model that
individuals can employ to determine whether or not the news they hear or see is
fake. Figure 4.5 shows the sequence of steps in the proposed method.
The following are the steps involved in our proposed strategies:
Step 1.
We collected the news dataset from Kaggle which has 20,800 data points
spread across five columns: - id, title, author, text, and label (as depicted in
Fig. 4.6).
Step 2.
We imported and downloaded the necessary libraries and modules like
pandas, numpy, matplotlib, nltk, and sklearn, required to construct our model.
Step 3.
We retrieved and assigned the collected news data from the comma-separated
values (CSV) file to a variable as illustrated in Fig. 4.7.
68 M. H. Mahalat et al.

Fig. 4.6 Spreading of data points across five columns collected from Kaggle

Fig. 4.7 Collected news data from CSV to variable

Fig. 4.8 Performing Null imputation on the dataset

Step 4.
We performed null imputation on the dataset that replaces null values present
in the dataset with empty space as shown in Fig. 4.8. Here 116 is the number of
null rows. Since the number of null rows is lower compared to that of the dataset,
we need to remove these rows.
Step 5.
Next we performed data preprocessing on the dataset, which included remov-
ing special characters, expanding contractions, converting to lowercase, word-
tokenization, and removing stopwords as shown in Fig. 4.9a, Fig. 4.9b, and
Fig. 4.9c respectively. The defined function preprocess_text removes special
characters, expands the contractions (e.g., if the dataset contains text like “I’ll”
then it will be converted to “I will”), and converting to lowercase. Here
word_tokenize on text_cols is performed, which splits a given sentence into a
list of words (e.g., “Hello, how are you?” will be converted to [“Hello,” “,”,
“how,” “are,” “you,” “?”]). Here stopwords are removed from the text_cols
(stopwords are the most commonly used words, e.g., “a,” “an,” “the,” “in,” etc.).
4 A Comparative Study of Various Machine Learning (ML) Approaches for. . . 69

Fig. 4.9 Data preprocessing: (a) removal of special characters, (b) word tokenization, (c) removal
of stopwords

Step 6.
After data preprocessing, the dataset is then considered for model training as
illustrated in Fig. 4.10a. Here label is taken as the target variable which will be
detected by the model. Here Feature Extraction is done on the dataset as illus-
trated in Fig. 4.10b. In order to use text data for predictive modeling, the text must
be parsed to remove certain words, which is called tokenization. The words in the
dataset may be encoded as integers or floating-point values to be provided as
input in model training algorithms. Count Vectorizer counts the number of unique
words, limits vocabulary size, etc. The tf-idf transformer then assigns a unique
value to each word as shown in Fig. 4.10c. This feature can be taken for model
training, which will be easier for the model to detect perfectly.

4.4 Result Discussion

The hardware and software requirements for our research are listed in Table 4.1. We
have split our dataset into training and testing dataset using train_test_split library
where 80% of the dataset is taken for training and the remaining 20% is taken for
testing. Here the function get_perf_metrics performs model training on the training
dataset and testing on the test dataset on the basis of the algorithms passed as
70 M. H. Mahalat et al.

Fig. 4.10 Steps after data preprocessing: (a). model training, (b). feature extraction, (c) tf-idf
transformer’s use

Table 4.1 Hardware and software required for implementation

Hardware requirements Software requirements
Processor Intel(R) Core (TM) i5-8250 CPU @ 1.60 GHz Python
1.80 GHz ReactJs
Flask
Visual Studio Code
RAM 8.00 GB (7.89 GB usable) Jupyter notebook
System type ×64-based processor 64-bit operating
system
Graphics NVIDIA GeForce 940MX
card

parameters as shown in Fig. 4.11. In this context, training of the model is done using
various algorithms like logistic regression, decision tree classifier, Linear SVC,
multinomialNB, and Random Forest Classifier with hypertuning as illustrated in
Fig. 4.12. We have assessed the performance in terms of various performance
parameters such as accuracy, precision, recall, and F1 score.
From the applied algorithms, Linear SVC gives us a better accuracy of 96.1566%
among all the approaches, as illustrated in Table 4.2, thereby ensuring its greater
ability to predict the correctness as compared to other strategies. Table 4.2 also
shows the attained precision, recall, F1 score, and training duration of various
classification models. It can be seen that precision is attained at 99.7475% for
multinomialNB, which is the highest value among all the chosen classifiers. More-
over, Linear SVC attained maximal recall value at 97.3774% and maximal F1 score
4 A Comparative Study of Various Machine Learning (ML) Approaches for. . . 71

Fig. 4.11 Model training and testing

Fig. 4.12 Various algorithms used for training of model

at 96.1861%. The training time is the least at 0.72 s for multinomialNB. The
accuracy levels of various strategies are as depicted in Fig. 4.13. Fig. 4.14 shows a
screenshot of an actual fake news item.

4.5 Conclusions and Future Direction

In this research, we have applied many classifiers and divided our model into two
phases: characterization and detection. We have also created a web-based applica-
tion for our suggested model that checks whether the news is fake or true as we enter
the news body. We have used 80% of the dataset for training, and 20% was used for
testing. Numerous classification models were employed to obtain accuracy, preci-
sion, recall, F1 score, and training time. It has been found that Linear SVC provides
us with a greater accuracy of 96% whereas multinomialNB achieves precision at
99.74%, which is the maximal value of all the classifiers applied. Additionally,
Linear SVC achieved a maximum F1 score of 96.18% and a maximum recall value
of 97.37% among all the other classification strategies. Further, for multinomialNB,
the training time is attained as the minimal at 0.72 s.
Although all the models perform well, our suggested model possesses a few
downsides, such as the fact that it can only detect news in English. Multi-linguistic
traits are not supported by our model and we have not trained it to understand more
than one language.
 72

Table 4.2 Results attained through various algorithms

Serial No. Model used Accuracy_Training_Set Accuracy_Test_Set Precision Recall F1 score Training time (s)
1 Logistic regression 0.983441 0.944888 0.929611 0.962118 0.945585 19.38
2 Decision tree classifier 0.999275 0.898235 0.890744 0.906751 0.898676 434.05
3 Linear SVC 0.999940 0.961566 0.950237 0.973774 0.961861 5.68
4 MultinomialNB 0.899257 0.788011 0.997475 0.575522 0.729905 0.72
5 Random Forest Classifier 0.748776 0.733382 0.732264 0.731909 0.732086 9.31
M. H. Mahalat et al.
4 A Comparative Study of Various Machine Learning (ML) Approaches for. . . 73

Fig. 4.13 Accuracy bar plot

Fig. 4.14 Fake news detection

In addition, our model does not examine the context of news throughout the
Internet. It only determines whether the news is true or not. It only notes the news
based on the words, i.e., the news body’s content. We hope to create a future system
that can support more than one language. We would also like to apply this method-
ology to social media apps, which are the leading source of bogus news. Our plan is
to enable the model to accurately detect the news, eliminating the need for a
74 M. H. Mahalat et al.

background cross-check. Finally, instead of writing long headlines, we hope to

create a voice-based input capability for our proposed model that would allow us
to determine if a piece of news is legitimate or fake by simply stating it out loud. We
are also open to suggestions for the types of benchmarks that should be used to
assess neural fake news detectors as a direction toward our future research.

Acknowledgments The authors are deeply grateful to the Department of Computer Science and
Information Technology of the Institute of Technical Education and Research, Siksha ‘O’
Anusandhan Deemed to be University for providing the required support for making this investi-
gation a success.

Conflict of Interest

There are no financial or non-financial conflicts of interest.

Funding There is no funding to declare.

References

Ahern KR, Sosyura D (2014) Who writes the news? corporate press releases during merger
negotiations. J Finance 69(1):241–291
Akoglu L, Chandy R, Faloutsos C (2013) Opinion fraud detection in online reviews by network
effects. In: ICWSM
Alkhodair SA, Ding SHH, Fung BCM, Liu J (2020) Detecting breaking news rumors of emerging
topics in social media. Inf Process Manag 57:102018
Allcott H, Gentzkow M (2017) Social media and fake news in the 2016 election. J Econ Perspect
31(2):211–235
Anonymous (2018) Twente student conference on IT, Jun 6th, 2018, Enschede, The Netherlands.
Netherlands. essay.utwente.nl
Antweiler W, Frank MZ (2005) Is all that talk just noise? The information content of internet stock
message boards. J Finance 59(3):1259–1294
Aphiwongsophon S et al. (2018) Detecting fake news with machine learning method. 2018 15th
international conference on electrical engineering/electronics, computer, telecommunications
and information technology (ECTI-CON). Chiang Rai, Thailand, Thailand. IEEE
Bovet A, Makse HA (2019) Influence of fake news in twitter during the 2016 US presidential
election. Nat Commun 10(1):1–14
Dua S, Du X (2016) Data mining and machine learning in cybersecurity. Auerbach Publications,
New York
Fan C (2017) Classifying fake news. http://www.conniefan.com/2017/03/classifying-fake-news.
Accessed 18 Feb 2018
Gilda S (2017) Evaluating machine learning algorithms for fake news detection. In: 15th student
conference on research and development (SCOReD). IEEE, pp 110–115
Granik M, Mesyura V (2017) Fake news detection using naive Bayes classifier. In: First Ukraine
conference on electrical and computer engineering (UKRCON). Ukraine. IEEE.
Huang T-Q (n.d.). https://www.researchgate.net/figure/Pseudo-code-of-information-gain-
basedrecursive-feature-elimination-procedure-with-SVM_fig2_228366941 2018
Jain A, Kasbe A (2018) Fake news detection. In: 2018 IEEE international Students' conference on
electrical, electronics and computer science (SCEECS). Bhopal, India. IEEE
4 A Comparative Study of Various Machine Learning (ML) Approaches for. . . 75

Jain A, Shakya A, Khatter H, Gupta AK (2019a) A smart news system for fake news detection using
machine learning. In: 2nd International conference on issues and challenges in intelligent
computing techniques (ICICIT)
Jain A, Shakya A, Khatter H, Gupta AK (2019b) A smart system for fake news detection using
machine learning. In: 2019 International conference on issues and challenges in intelligent
computing techniques (ICICT), pp. 1–4, doi: https://doi.org/10.1109/ICICT46931.2019.
8977659
Kaur P et al (2019) Hybrid text classification method for fake news detection. Int J Eng Adv
Technol (IJEAT):2388–2392
Kevric J et al (2017) An effective combining classifier approach using tree algorithms for network
intrusion detection. Neural Comput & Applic 28:1051–1058
Khanam Z, Agarwal S (2015) Map-reduce implementations: survey and performance comparison.
Int J Comput Sci Inform Technol (IJCSIT) 7(4):119
Khanam Z, Ahsan MN (2017) Evaluating the effectiveness of test driven development: advantages
and pitfalls. Int J Appl Eng Res 12:7705–7716
Khanam Z, Ahsan MN (2018) Implementation of the pHash algorithm for face recognition in
secured remote online examination system. Int J Adv Sci Res Eng (IJASRE) 4(11):01
Kogan S, Moskowitz TJ, Niessner M (2019) Fake news: evidence from financial markets. https://
ssrn.com/abstract=3237763
Lazer DMJ et al (2018) The science of fake news. Science 359(6380):1094–1096. https://doi.org/
10.1126/science.aao2998. https://science.sciencemag.org/content/359/6380/1094.summary
Looijenga MS (2018) The detection of fake messages using machine learning
Ludwig K, Creation M (2020) Dissemination and uptake of fake-quotes in lay political discourse on
Facebook and twitter. J Pragmat 157:101–118
Meyer R (2017) The rise of progressive ‘Fake News’. Retrieved from The Atlantic: https://www.
theatlantic.com/technology/archive/2017/02/viva-la-resistancecontent/515532/
MonaDiab et al. (2004) Automatic tagging of Arabic text: from raw text to base phrase chunks.
Proceedings of HLT-NAACL 2004: short papers, Association for Computational Linguistics,
Boston, MA, pp. 149–152
Mosseri A (2016) News feed FYI: Addressing hoaxes and fake news. Retrieved from Facebook
newsroom
Parikh SB, Atrey PK (2018) Media-rich fake news detection: a survey. IEEE conference on
multimedia information. Miami, FL: IEEE
Perez-Rosas V et al. (2017). https://www.researchgate.net/publication/319255985_Automatic_
Detection_of_Fake_News
Qin Y et al (2018) Predicting future rumours. Chin J Electron 27(3):514–520
Radianti J et al. (2016) An overview of public concerns during the recovery period after a major
earthquake: Nepal Twitter analysis. HICSS '16 Proceedings of the 2016 49th Hawaii Interna-
tional Conference on System Sciences (HICSS). Washington, DC, USA. IEEE, pp. 136–145
Ranjan et al. (2003) Part of speech tagging and local word grouping techniques for natural language
parsing in Hindi. In Proceedings of the 1st international conference on natural language
processing (ICON 2003). Semanticscholar
Rashkin H, Choi E, Jang J, Volkova S, Choi Y (2017) Truth of varying shades: analyzing language
in fake news and political fact-checking. In: EMNLP
Raza S, Ding C (2022a) Fake news detection based on news content and social contexts: a
transformer-based approach. Nat Publ Health Emerg Collection 13:335
Raza S, Ding C (2022b) Fake news detection based on news content and social contexts: a
transformer-based approach. Int J Data Sci Analyt 13(4):335–362
Researchgate.net (2014). https://www.researchgate.net/figure/Pseudocode-for-KNNclassification_
fig7_260397165
Researchgate.net (2018). https://www.researchgate.net/figure/Pseudocode-ofnaive-bayes-algo
rithm_fig2_325937073
Rouse M (2018). https://searchenterpriseai.techtarget.com/definition/machine-learning-ML
76 M. H. Mahalat et al.

Rubin V (2017) Deception detection and rumor debunking for social media. Handbook of social
media research methods
Rubin V, Conroy N, Chen Y, Cornwell S (2016) Fake news or truth? Using satirical cues to detect
potentially misleading news. In: NAACLCADD
Sharma K, Qian F, Jiang H, Ruchansky N, Zhang M, Liu Y (2019) Combating fake news: a survey
on identification and mitigation techniques. ACM Trans Intell Syst Technol (TIST) 10(3):1–42
Shu K et al (2017) Fake news detection on social media: a data mining perspective. ACM SIGKDD
Explor Newslett 19(1):22–36
Shu K, Wang S, Liu H (2019) Beyond news contents: the role of social context for fake news
detection. In: Proceedings of the twelfth ACM international conference on web search and data
mining, pp. 312–320
da Silva FCD et al. (2019a) Can machines learn to detect fake news? A survey focused on social
media. https://scholarspace.manoa.hawaii.edu/handle/10125/59713
da Silva FCD, Vieira, R., & Garcia, AC. (2019b). Can machines learn to detect fake news? A survey
focused on social media. In: Proceedings of the 52nd Hawaii International Conference on
System Sciences
Singh V, Dasgupta R, Sonagra D, Raman K, Ghosh I (2017) Automated fake news detection using
linguistic analysis and machine learning. In: SBP-BRiMS
Sirikulviriya N, Sinthupinyo S (2011) Integration of rules from a random forest. International
conference on information and electronics engineering, p 194, 198. Singapore,
semanticscholar.org
Tacchini E, Ballarin G, Della Vedova M, Moret S, de Alfaro L (2017) Some like it hoax: automated
fake news detection in social networks. CoRR, abs/1704.07506
Tapaswi et al. (2012) Treebank based deep grammar acquisition and part-of-speech tagging for
Sanskrit m sentences. Software engineering (CONSEG), on software engineering (CONSEG).
IEEE, pp 1–4
Traore et al. (2017) Detection of online fake news using N-gram analysis and machine learning
techniques. International conference on intelligent, secure, and dependable systems in distrib-
uted and cloud environments. Springer International Publishing, pp. 127–138
Vosoughi S, Roy D, Aral S (2018) The spread of true and false news online. Science 359(6380):
1146–1151
Yi J et al. (2003) Sentiment analyzer: extracting sentiments about a given topic using natural
language processing techniques. In Data mining, 2003. ICDM 2003. Third IEEE International
Conference, pp 427–434. http://citeseerx.ist.psu.edu
Zhang J, Dong B, Yu Philip S (2020) Fakedetector: effective fake news detection with deep
diffusive neural network. In: 2020 IEEE 36th international conference on data engineering
(ICDE). IEEE
Chapter 5
Analytics and Decision-making Model
Using Machine Learning for Internet
of Things-based Greenhouse Precision
Management in Agriculture

Ashay Rokade, Manwinder Singh, Anudeep Goraya, and Balraj Singh

Abstract As wired systems for smart farming are difficult to manage and install,
wireless connectivity is currently taking their place. Smart farming with precision
greenhouse technology is installed to improvise in managing the growth of agricul-
ture and therefore observing different environments in precision agriculture. Numer-
ous systems have been developed for control and remote monitoring of precision
agriculture. But due to limited solutions, monitoring of greenhouse is not yet
competent to deal with the agricultural growth on entirely control systems. For
better farming growth control, smart farming with precision greenhouses must be
applied, necessitating precision agriculture monitoring under various circumstances.
Supervised machine learning techniques are used in intelligent agricultural systems
to provide intelligent information farming systems with predictive data analysis of
sensor parameters. Cloud layer, fog layer, edge layer, and sensor layer are four
important parts of the proposed approach. The data needed for the sensor layer of the
analytical model is collected using Internet of Things-based embedded system
devices in two greenhouse facilities, with sensor parameters as inputs and
corresponding actuators as outputs. Using classification and regression models,
two distinct analytical models for intelligent and accurate farming were built. By
modifying farming circumstances in accordance with plant requirements taken into
account during experimentation, the primary goals of this analytics are to boost
output and offer organic farming. A decision-making and analytics system was built
at the fog layer using the support vector machine and artificial neural network, two

A. Rokade · M. Singh
School of Electronics and Electrical Engineering, Lovely Professional University, Phagwara,
Punjab, India
e-mail: [email protected]
A. Goraya · B. Singh (✉)
School of Computer Science and Engineering, Lovely Professional University, Phagwara,
Punjab, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 77
A. Khamparia et al. (eds.), Microbial Data Intelligence and Computational
Techniques for Sustainable Computing, Microorganisms for Sustainability 47,
https://doi.org/10.1007/978-981-99-9621-6_5
78 A. Rokade et al.

supervised classification-based machine learning methods. MATLAB software’s

statistics and machine learning tools were used to analyze and interpret the exper-
imental outcomes. Accuracy, sensitivity, specificity, and F-score are used to examine
the confusion matrix-based metrics used in the performance evaluation of the
suggested system. Based on the results of the experiments, the suggested method
also proved to be the best at providing actuators with predictions and control.

Keywords Precision agriculture · Intelligent greenhouse · Internet of things ·

Machine learning · Smart farming · Fog computing

5.1 Introduction

Making plant agriculture a creative task, the land and the quality of the plants are
now the crucial daily bounds for either money harvests or food crops. Poor farming
knowledge and information about new techniques is a significant problem in modern
agriculture. Our forefathers in the agricultural sector avoided using specialized
technology for individual plant growth in favor of general natural phenomena. The
introduction of technical machinery into the agricultural sector has made it possible
to cultivate plants in settings far beyond the norm. This has led to the production of
both higher yields and lower manure usage. The widespread use of fertilizers,
defoliant, and water in plant crops is in line with their natural rationality (Chehri
et al. 2020; Subahi and Bouazza 2020). In intensive nursery settings, growers often
use agrochemicals in quantities that exceed the true yield demands, leading to
ecological pollution and waste. When compared to predictions of total creation,
the value of water and agronomics is low since crops are managed with a lot of
induction rather than taking into account target estimations from advanced crop-
checking technology. For instance, a sizable percentage of the working population
relies on the farming sector for their living. In India, this number is particularly high,
at 61% of the working population, respectively (Kour and Arora 2020; Chukkapalli
et al. 2020).
Figure 5.1 shows the industry’s significant export growth over the previous
12 months. In FY22, exports of marine products totaled $7.77 billion, exports of
rice, basmati and non-basmati, totaled $6.98 billion USD, the entire value of buffalo
meat exports was $3.30 billion, exports of sugar totaled $4.60 billion USD, tea
exports were worth US$750.93 million, and coffee exports were worth $1020.80 bil-
lion USD.
Modern farming and horticultural production systems are undergoing significant
technological progress, which has given rise to the terms “agriculture 4.0” and
“Smart Agriculture” (SA) (El-Basioni and El-Kader 2020). Numerous innovative
technologies, such as autonomous agricultural trucks, satellite infrastructure, and
unmanned aerial vehicles (UAVs), will be linked to SA-based future scenarios. In
particular, modern farmers will benefit greatly from adopting both technologies
related to Precision Farming and the Internet of Things (IoT). In reality, a much
more sensible and superior horticultural production framework is needed to
5 Analytics and Decision-making Model Using Machine Learning for Internet. . . 79

45.00

41.25**
40.00

38.54
38.21
35.00

35.09
33.87
32.08
30.00

25.00

20.00

18.91
15.00

10.00

5.00

0.00
FY16 FY17 FY18 FY19 FY20 FY21 FY22*

Fig. 5.1 Agriculture exports from India (US$ billion) (Source: India Brand Equity Foundation)

adequately meet a few impending problems, like rapid increase in population,

atmospheric conditions, and common assets consumption. Thus, investigation and
mechanical advancement can be the solutions to reduce these challenges. The
advancements in agriculture (Alonso et al. 2020; Bandur et al. 2019; Carrasquilla-
Batista and Chacon-Rodriguez 2019; Tan et al. 2020) in the twenty-first cen-
tury helped to overcome the designing challenges with inventions (Prabha et al.
2018; Sadowski and Spachos 2018; Lakshmanna et al. 2022a; Rezk et al. 2021;
Tageldin et al. 2020; Aliar et al. 2022; Sekaran et al. 2020; Singh et al. 2018; Roy
et al. 2020; Hassan et al. 2021a). The advancement related to agriculture 5.0 is based
on the concept that farms are using automated activities and emotionally supporting
networks based on the freedom of individual choice, as outlined by the precision
agriculture standards. As a result, it is likely that the ideas behind agriculture 5.0 will
incorporate the usage of robotics and possibly even some types of artificial intelli-
gence (Hassan et al. 2021b, 2021c, 2021d; Rokade and Singh 2021; Kadu and Singh
2021; Khalaf et al. 2021; Walia et al. 2021, 2022a, 2022b; Ali Al-Samawi and Singh
2022; Belsare and Singh 2022a). Ranches have typically relied on a large number of
sporadic professionals to gather harvests and keep profits high. Since society has
changed from an agricultural culture in which many people lived on homesteads to
one in which many people live in urban settlements, ranches are facing the difficulty
of a manpower shortage. AI-enhanced farming robots are one approach to the
workforce crisis (Hassan et al. 2022a, 2022b, 2022c; Rokade et al. 2022a, 2022b;
80 A. Rokade et al.

Bolla and Singh 2022; Kadu et al. 2022; Marwah et al. 2022; Belsare and Singh
2022b; Singh and Gandam 2022; Singh et al. 2022).

5.2 Related Work

In order to maximize energy utilization and network longevity, K. Lakshmana et al.

(Lakshmanna et al. 2022b) present an enhanced IMD-EACBR technique for wireless
sensor networks (WSN). Based on factors including energy savings, detachment,
node degree, and inter-cluster distance, the IAOAC algorithm chooses an appropri-
ate aim that connects different structures. Several facets of the IMD-EACBR
model’s performance have been investigated. The last step is extensive testing of
the proposed network utilizing all of NS-3.26’s simulation features. Improvements
in packet delivery ratio (PDR), latency, energy consumption, and number of dead
nodes are among the other metrics that stand out from the simulation findings.
Lakshmanna et al. (2022a) created a new privacy-protecting multi-agent system
(MAS) for an industrial IoT (IIoT) environment. In order to choose and build the
right clusters for the IIoT system, they first created an expanded moth swarm
algorithm-based clustering (EMSA-C) method. Additionally, a multi-agent system
is implemented to provide encrypted exchanges between different clusters. The
technique’s potential is investigated via a comprehensive comparative study, with
outcomes evaluated across a range of metrics. The investment required is substantial.
The high cost of implementation is an evident issue for industrial IoT. Secure data
storage and management connection failures are frequent because of the enormous
volume of data generated by IoT devices. The simulation findings show that
BDL-PPDT is superior to current approaches in terms of output. The provided
BDL-PPDT method may only have a 98.15% success rate, but it nonetheless yields
the best attainable result. The BDL-PPDT approach was demonstrated to be superior
to the other existing methods across a number of criteria and is therefore advised
based on the findings of the aforementioned data analysis.
Suma (2021) highlight the challenges and complications that may be encountered
when integrating modern farming practices with older methods of production. The
use of statistical and quantitative techniques can lead to revolutionary changes in our
current agricultural system. The present and upcoming agricultural trends are pro-
vided through the systematic analysis.
In order to predict agricultural productivity and drought, Rezk et al. (2021)
propose an IoT-based smart farming system and a powerful prediction technique
called WPART that is based on ML techniques. The suggested approach is estimated
using five different datasets. The findings showed that the suggested strategy
outperformed the existing methods in categorizing and predicting drought and
agricultural productivity. The outcomes revealed that the suggested strategy was
the most effective at forecasting drought and measuring the yield of crops. The
WPART approach outperforms state-of-the-art, gold standard algorithms with
5 Analytics and Decision-making Model Using Machine Learning for Internet. . . 81

accuracy levels above 90%, across five datasets assessing drought categorization and
crop yield.
Araby et al. (2019) collected various data samples from a variety of crop fields
using a sensor network, transmitted it to an ML algorithm to produce an alarm
message, and then showed it through graphical user interface.
Tageldin et al. (2020) have developed a number of ML strategies for predicting
infection in plants. This study laid the groundwork for utilizing ML to forecast the
presence of CLW in greenhouse crops. Weekly CLW data collection was conducted
in a commercial hydroponic greenhouse for 2 years. Temperature and relative
humidity readings were also taken continuously during the investigation. The
XGBoost algorithm was found to be the most efficient one used throughout this
research. This algorithm has attained accuracy in prediction of 84%. To guarantee a
comprehensive dataset for future outcomes, the authors investigated the effect of
several environmental factors on prediction precision.
In their comprehensive review of smart farming techniques and designs, Ahamed
Aliar et al. (2022) cover all the bases. They also provide an in-depth analysis of
various designs and viable recommendations for fixing the current state of smart
farming.
Remote monitoring of rice paddies using deep learning and the IoT is proposed
by Sethy et al. (2021). For rice leaf disease detection and nitrogen status assessment,
the Vgg16 pre-trained network is being investigated. In this context, transfer learn-
ing and deep feature extraction are used to recognize photos. Support vector
machines (SVM) have been introduced to identify pictures with the deep feature
extraction method. Vgg16’s transfer learning method achieves 79.86 and 84.88%
accuracy, respectively, when used to recognize four distinct leaf diseases and to
forecast nitrogen status. The Vgg16 deep features and the SVM findings have a
97.31 and 99.02% accuracy rate in recognizing four different leaf diseases and
predicting nitrogen status, respectively. Additionally, an IoT and deep learning-
based architecture is proposed for remote paddy field monitoring. The proposed
prototype has advantages over the state of the art since it not only regulates
temperature and humidity, but also monitors the extra two factors: the detection of
nitrogen status and illnesses.
The Smart Agriculture approach proposed by Kaushik et al. (2019) includes
monitoring the agricultural land and can greatly aid farmers in increasing output.
The cloud-stored data, which contains details like temperature, moisture, and humid-
ity that affect disease in an agricultural field, is subjected to a naive Bayes analysis.
An IoT-enabled agricultural monitoring prototype was proposed by Wong Hin
Yong et al. (Siddiquee et al. 2022), which would use a variety of algorithms to
monitor crops for a variety of purposes, including detection, quantification, maturity
testing, and disease. The authors discuss IoT-enabled, intelligent farm monitoring
solutions. Defects in vegetables have also been identified by the use of color
threshold and color segmentation. All methods were designed and implemented
using convolutional neural networks (CNNs), an ML technique. In order to deter-
mine which approach would be best for integrating into this agricultural monitoring
system, MATLAB simulations have been used to compare traditional methods with
82 A. Rokade et al.

CNNs. This study found that CNNs outperformed other approaches and existing
algorithms with an accuracy of 90% or higher, making it the preferred option.
Using the IoT to carry out a variety of outdoor tasks, Ferehan et al. (2022) created
and launched a revolutionary wireless mobile robot. Agriculture, transportation, and
water distribution can all benefit from this study’s findings. The IoT and remote
sensor system are used to create sophisticated agricultural frameworks in many
regions of the world. Singh et al. (Singh and Verma 2022a, 2022b) discussed the
various applications of machine learning (ML) and data processing using a big data
platform. In this regard, one of the offshoots of intelligence is the practice of
exactness. For a number of agricultural applications, researchers have created a
wide range of checking and robotization frameworks. Data collection and sharing
among ranch-based IoT devices will be simple with the help of WSN.

5.3 Development of IoT-based Smart Farming System

The proposed Smart Farming System for Data Analytics (SFSDA) Using ML
Enabled Internet of Things will be divided into four layers for better understanding.
The four basic architectural layers for a precisely regulated greenhouse management
system are the cloud layer for data storage, fog layer for data processing, edge layer
for data conversion and sampling, and sensor/device layer for data collection. In
order to gather data on the greenhouse environment, including temperature, humid-
ity, CO2, soil moisture, and light intensity, a hardware prototype has been
constructed. Figure 5.2 depicts the suggested paradigm for an intelligent farming
system to be used in the greenhouse.

Fig. 5.2 Smart farming infrastructure for greenhouse management

5 Analytics and Decision-making Model Using Machine Learning for Internet. . . 83

Fig. 5.3 Experimental test bench

The sensor layer is where all the field environment-related sensors and actuators
are situated. The edge layer typically comprises a controller unit to which various
sensors and actuators are connected in order to collect data for transmission to the
fog layer. By leveraging edge-level data, the fog layer can build analytics and
decision-making models and send actuator signals to the next layer. Finally, the
upper layer provides a user interface dashboard which displays a graphical repre-
sentation of sensor and actuator data. The proposed framework is unique in that it
uses the IoT to aid farmers with greenhouse management. It is all done remotely so
that farmers may monitor and adjust factors like soil moisture, CO2, light, and
temperature from afar.
The results of prototype tests utilizing the suggested experimental design, which
took into consideration two harvests produced under various conditions, are shown
in Fig. 5.3. The three main testing phases involved constructing core model-
embedded systems for plant growth and feeding, automating actuators, and devel-
oping a sensor net for intelligent greenhouse monitoring. The suggested solution
used an embedded system to precisely measure the soil moisture, temperature, CO2,
and plant light—all essential elements in a greenhouse’s effective operation.
The climate conditions in the study region are tropical wet and dry, with hot, dry
summers and mild to cool winters. Winter lasts from November to March, rainy
season from July to October, and the summer season from March to June. The test
field region, which is 0 feet (0 m) above sea level, experiences tropical wet and dry
seasons. The district’s average annual temperature of 30.63 °C (87.13 °F) is 4.66%
higher than the national average for India. The test field area normally has 103.26
wet days (28.29% of the time) annually and receives about 120.15 millimeters
(4.73 inches) of precipitation on average. The maximum temperature observed is
47 ° C and the lowest is 24.03 ° C.
84 A. Rokade et al.

Fig. 5.4 Sensors reading in rainy season for greenhouse monitoring

Similarly, the data collection is processed for the rainy season for all the five
parameters inside the greenhouse comparing reference data of gerbera and broccoli
with actual data in Fig. 5.4.
5 Analytics and Decision-making Model Using Machine Learning for Internet. . . 85

As shown in the graph, the temperature of the greenhouse is quite low at night and
high in the day time. It will require much more data monitoring and parameter
controlling over the season as the temperature requirement for both gerbera and
broccoli is low for both day and night time.
The graphs conclude that winter is the best season for cultivation of gerbera and
broccoli. The research works for the cultivation of crops in any season using the
greenhouse management system and by applying ML algorithms to control the
atmosphere of the greenhouse for smooth development of the crop.
The proposed approach was established at the third layer of our proposed model,
i.e., the fog layer for a data analytics system employing various ML methods. The
SVM and multilayer perceptron machine learning algorithms were chosen because
of their theoretical and implementation advantages which suit the system dataset
perfectly to obtain the precise output. The classification and regression techniques
were adopted and analyzed to acquire the expected results. The actual data and
expected data are compared for both techniques. The system is developed to control
the various devices like pump, fan, and light within the greenhouse, making the
greenhouse a smart one.
The training and testing of data is done in this chapter using the MATLAB 2021
version. The accuracy, sensitivity, specificity, latency, F-score, and least root square
mean (LSME) are calculated and observed.

5.4 Experimental Results and Discussion

A confusion matrix is used to gauge the system’s effectiveness. While root-mean-

square-error (RMSE) is computed for regression models using the confusion matrix
parameters, accuracy, sensitivity, specificity, and F-score are computed for classifi-
cation models. Sensitivity quantifies a model’s ability to recognize positive class
members accurately, while specificity gauges how well it can recognize negative
class members. In some applications, one statistic could be more important than
another. The accuracy of a model is assessed using the F score, a machine learning
evaluation statistic. It takes into consideration a model’s recall and precision scores.
The accuracy statistic shows how often a model predicts correctly over the entire
dataset. When your class is divided unevenly, the F grade is typically more helpful
than accuracy. When false positive and false negative costs are roughly similar,
accuracy performs best. In cases where the costs of false positives and false
negatives differ greatly, it is advisable to incorporate both precision and recall.
The recommended approach makes use of an embedded system to conduct a
reliable analysis of CO2, soil moisture, temperature, and plant light in greenhouse
operations. As the measurements are taken under various environmental conditions,
they are all monitored in real time on a personal computer through serial transmis-
sion. As data is published from nodes to Adafruit’s broker, it can be seen in the
Adafruit IO Cloud dashboard for remote monitoring of all these sensors. The user
may then subscribe to this information in order to obtain updates as they occur.
86 A. Rokade et al.

Table 5.1 Performance evaluation of classification model

Accuracy (%) Sensitivity (%) Specificity (%) F-score (%)
Attributes SVM MLP SVM MLP SVM MLP SVM MLP
Pump 86.66 96.66 81.81 100 100 87.5 95.74 96.49
Light 96.66 100 100 100 90.90 100 95.95 100
Fan 90 96.66 94.11 94.11 84.61 100 89.88 98.76

Table 5.2 Performance evaluation of regression model

PPV (%) NPV (%) FNR (%) FPR (%)
Attributes SVM MLP SVM MLP SVM MLP SVM MLP
Pump 100 95.65 66.66 100 18.18 0 0 12.5
Light 95 100 100 100 0 0 9.09 0
Fan 88.88 100 91.66 92.85 5.88 5.88 15.38 0

Table 5.3 The result of comparative analysis

Accuracy Sensitivity Specificity F-score Latency
Ref. (%) (%) (%) (%) (s) RMSE
Lakshmanna et al. NI 93.59 94.63 96.30 NI NI
(2022a)
Rezk et al. (2021) NI NI NI NI NI 0.2431
Tageldin et al. 87.35 87.33 95.76 87.17 224.6 NI
(2020)
Aliar et al. (2022) 90 NI NI NI NI NI
Sekaran et al. (2020) NI NI NI NI NI 0.02726
Proposed work 97.77 98 98.83 98.41 6.49 0.0615
NI: Not investigated in research

Table 5.1 shows the categorization model based on a confusion matrix. The MLP
method outperforms the SVM in every metric taken into account, according to the
data in Table 5.1.
Similarly, the other performance evaluation parameters are evaluated from con-
fusion matrix attributes, PPV, NPV, FNR, and FPR as shown in Table 5.2. It has
been found that the suggested system can effectively classify with greater PPV and
NPV values as well as with minimal FNR and FPR ratio values.
The final section of Table 5.3 presents a comparison to the most recent state-of-
the-art techniques. The results show that the proposed classification and regression
model for intelligent and precision smart farming in greenhouses produces better
results when accuracy, sensitivity, and specificity of the classification model are
compared with the root-mean-square-error (RMSE) of the regression model.
The results show that the proposed system works with high precision by
implementing the proper resource utilization and comparing system parameters
like accuracy, sensitivity, specificity, F-score, latency, and RMSE. The average
values of the MLP classification and SVM regression algorithms were used to
extract the findings from the suggested work in order to produce superior outcomes.
5 Analytics and Decision-making Model Using Machine Learning for Internet. . . 87

Table 5.4 Comparative analysis of various existing studies

Technique(s)/models System
Ref. used Gap/feature accuracy
Tan et al. (2020) Rasbery Pi & Open Only focuses on automating pH 92%
Cv control
Prabha et al. (2018) Bayesian network and Less data to analyze in the 66%
ML datasets
Sadowski and Spachos ANN and fuzzy logic It is an expensive system 89%
(2018)
Proposed work SVM and MLP Low-cost system 97.77%

A comparison of different models that have already been developed is shown in

Table 5.4.
In Table 5.4 it is shown that researchers have taken the initiative toward precision
agriculture and the proposed model considered numerous factors, e.g., crop sensors
and technology, to make an effective smart greenhouse management system which
uses machine learning-enabled Internet of Things. This proposed system can be
easily used and handled by farmers for crop wellness.

5.5 Conclusion

This study proposes a method through which an intelligent greenhouse’s automation

is upgraded by using IoT technologies and concepts. A user can now control and
monitor data transfer between a device and a fog layer, and vice versa, using real-
time sensor data. In order to improve agricultural output, the IoT concept is applied
to the system by centralizing data storage and processing in a reliable cloud. The
precision in data rectifies the proper resource utilization. Use of IoT lowers mainte-
nance expenses.
The proposed system will correctly monitor and adjust greenhouse characteristics
such soil moisture, carbon dioxide levels, temperature, humidity and light to assist
farmers in boosting production. The model is validated by using data from actual
greenhouses to determine the optimum soil moisture, carbon dioxide, temperature,
humidity and light for producing broccoli and gerbera. The proposed monitoring
system will be used for any crop which can be cultivated inside the greenhouse,
which results in disease-free and large production of the crop.
Utilizing supervised ML methods based on classification and regression, the
special greenhouse system is designed for intelligent, accurate control. The
recommended approach can be applied in a smart agricultural environment that
uses an IoT-based paradigm for decision-making. Classification and regression
models are the two types of analytics used to create programs for smart and accurate
farming. Both SVM and MLP can be utilized for these modeling purposes.
Using classification and regression-based supervised ML algorithms, the effec-
tiveness of the smart farming system is ultimately assessed in terms of intelligent and
88 A. Rokade et al.

precision farming. The outcomes showed that MLP performed better than SVM and
other state-of-the-art classification methods. The MLP system’s accuracy is 97.77%,
sensitivity is 98%, specificity is 98.83%, and F-score is 98.41% with a lower error
rate achieved by the system. The suggested technique also provided the most
accurate predictions to actuators and the most precise control.

References

Ali Al-Samawi MA, Singh M (2022) Effect of 5G on IOT and daily life application. 2022 3rd
International conference for emerging technology (INCET), pp 1–5, doi: https://doi.org/10.
1109/INCET54531.2022.9823983
Aliar AAS, Yesudhasan J, Alagarsamy M, Anbalagan K, Sakkarai J, Suriyan K (2022) A compre-
hensive analysis on IoT-based intelligent farming solutions using machine learning algorithms.
Bull Electric Eng Inform 11(3):1550–1557, ISSN: 2302-9285. https://doi.org/10.11591/eei.
v11i3.3310
Alonso RS, Sittón-Candanedo I, Casado-Vara R, Prieto J, Corchado JM (2020) Deep reinforcement
learning for the management of software-defined networks in smart farming. 2020 International
conference on Omni-layer intelligent systems (COINS), Barcelona, Spain, pp 1–6. doi: https://
doi.org/10.1109/COINS49042.2020.9191634
Araby A et al. (2019) Intelligent IoT monitoring system for agriculture with predictive analysis. In
2019 8th International conference on modern circuits and systems technologies (MOCAST),
pp. 1–4, doi: https://doi.org/10.1109/MOCAST.2019.8741794
Bandur D, Jaksic B, Bandur M, Jovic S (2019) An analysis of energy efficiency in Wireless Sensor
Networks (WSNs) applied in smart agriculture. Comput Electron Agric 156:500–507. https://
doi.org/10.1016/j.compag.2018.12.016
Belsare KS, Singh M (2022a) Various frameworks for IoT-enabled intelligent waste management
system using ML for smart cities. In Mobile computing and sustainable informatics, pp
797–817. Springer, Singapore, doi: https://doi.org/10.1007/978-981-19-2069-1_55
Belsare K, Singh M (2022b) An intelligent internet of things (IoT) based automatic dry and wet
medical waste segregation and management system. In 2022 International Conference on
Augmented Intelligence and Sustainable Systems (ICAISS), pp. 1113–1119. IEEE
Bolla S, Singh M (2022) Energy harvesting technique for massive MIMO wireless communication
networks. J Phys Conf Ser 2327(1):012059. IOP Publishing. https://doi.org/10.1088/1742-
6596/2327/1/012059
Carrasquilla-Batista A, Chacon-Rodriguez A (2019) Standalone fuzzy logic controller applied to
greenhouse horticulture using internet of things, 2019 7th international engineering, sciences
and technology conference (IESTEC), Panama, Panama, pp 574–579, doi: https://doi.org/10.
1109/IESTEC46403.2019.00108
Chehri A, Chaibi H, Rachid S, Hakem N, Wahbi M (2020) ScienceDirect “A framework of
optimizing the deployment of IoT for precision agriculture” Industry-NC-ND license (https://
creativecommons.org/licenses/by-nc-nd/4.0/) Peer-review under responsibility of KES
International
Chukkapalli S, Mittal S, Gupta M, Abdelsalam M, Joshi A, Sandhu R, Joshi K (2020) Ontologies
and artificial intelligence systems for the cooperative smart farming ecosystem. IEEE Access 8:
164,045
El-Basioni BMM, El-Kader SMA (2020) Laying the foundations for an IoT reference architecture
for agricultural application domain. IEEE Access 8:190,194–190,230. https://doi.org/10.1109/
ACCESS.2020.3031634
Ferehan N, Haqiq A, Wazih M (2022) Intelligent farming system based on intelligent internet of
things and predictive analytics. J Food Qual 2022:1–8. https://doi.org/10.1155/2022/7484088
5 Analytics and Decision-making Model Using Machine Learning for Internet. . . 89

Hassan M, Singh M, Hamid (2021a) Survey on NOMA and spectrum sharing techniques in 5G,
2021 IEEE International Conference on Smart Information Systems and Technologies (SIST),
pp. 1–4. doi: https://doi.org/10.1109/SIST50301.2021.9465962
Hassan M, Singh M, Hamid K (2021b) Overview of cognitive radio networks. J Phys Conf
Ser 1831
Hassan M, Singh M, Hamid K (2021c) Survey on advanced spectrum sharing using cognitive radio
technique. In ICT systems and sustainability. Springer, Singapore, 639–647
Hassan M, Singh M, Hamid K (2021d) Impact of power and bandwidth on the capacity rate and
number of users in SC-NOMA. Harbin Gongye Daxue Xuebao/J Harbin Inst Technol 53(9):
118–124. http://hebgydxxb.periodicales.com/index.php/JHIT/article/view/726. Accessed
15 Feb 2022
Hassan M, Singh M, Hamid K (2022a) Review of NOMA with spectrum sharing technique. In ICT
with intelligent applications. Springer, Singapore, pp 135–143. doi: https://doi.org/10.1007/
978-981-16-4177-0_16
Hassan M, Singh M, Hamid K, Saeed R, Abdelhaq M, Alsaqour R (2022b) Modeling of NOMA-
MIMO-based power domain for 5G network under selective rayleigh fading channels. Energies
15(15):5668. https://doi.org/10.3390/en15155668
Hassan M, Singh M, Hamid K, Saeed R, Abdelhaq M, Alsaqour R (2022c) Design of Power
Location Coefficient System for 6G downlink cooperative NOMA network. Energies 15(19):
6996. https://doi.org/10.3390/en15196996
Kadu, Singh M (2021) Comparative analysis of e-health care telemedicine system based on internet
of medical things and artificial intelligence. 2021 2nd International conference on smart
electronics and communication (ICOSEC), pp 1768–1775, doi: https://doi.org/10.1109/
ICOSEC51865.2021.9591941
Kadu A, Singh M, Ogudo K (2022) A novel scheme for classification of epilepsy using machine
learning and a fuzzy inference system based on wearable-sensor health parameters. Sustainabil-
ity 14(22):15079. https://doi.org/10.3390/su142215079
Kaushik N, Narad S, Mohature A, Sakpal P (2019) Predictive analysis of IoT based digital
agriculture system using machine learning. Int J Eng Sci Comput 9(3)
Khalaf OI, Ogudo KA, Singh M (2021) A fuzzy-based optimization technique for the energy and
spectrum efficiencies trade-off in cognitive radio-enabled 5G network. Symmetry 13(1):47.
https://doi.org/10.3390/sym13010047
Kour V, Arora S (2020) Recent developments of the internet of things in agriculture: a survey. IEEE
Access 8:129,924–129,957. https://doi.org/10.1109/ACCESS.2020.3009298
Lakshmanna K, Kavitha R, Geetha BT, Nanda AK, Radhakrishnan A, Kohar R (2022a) Deep
learning-based privacy-preserving data transmission scheme for clustered IIoT environment.
Comput Intell Neurosci 8:2022
Lakshmanna K, Subramani N, Alotaibi Y, Alghamdi S, Khalafand OI, Nanda AK (2022b)
Improved metaheuristic-driven energy-aware cluster-based routing scheme for IoT-assisted
wireless sensor networks. Sustainability 14(13):7712
Singh M, Navdeep Kaur Jhajj, and Anudeep Goraya (2022) IoT-enabled wireless mobile ad-hoc
networks: introduction, challenges, applications: review chapter. Internet of Things, pp 121–134
Marwah, Kour GP, Jain A, Malik PK, Singh M, Tanwar S, Safirescu CO, Mihaltan TC, Sharma R,
Alkhayyat A (2022) An improved machine learning model with hybrid technique in VANET for
robust communication. Mathematics 10(21):4030. https://doi.org/10.3390/math10214030
Prabha R, Sinitsambirivoutin E, Passelaigue F, Ramesh MV (2018) Design and development of an
IoT based smart irrigation and fertilization system for chilli farming, 2018 International confer-
ence on wireless communications, signal processing and networking (WiSPNET), Chennai, pp
1–7, doi: https://doi.org/10.1109/WiSPNET.2018.8538568
Rezk NG, Hemdan EED, Attia AF et al (2021) An efficient IoT-based intelligent farming system
using machine learning algorithms. Multimed Tools Appl 80:773–797. https://doi.org/10.1007/
s11042-020-09740-6
90 A. Rokade et al.

Rokade, Singh M (2021) Analysis of precise green house management system using machine
learning based internet of things (IoT) for smart farming. 2021 2nd International conference on
smart electronics and communication (ICOSEC), pp. 21–28, doi: https://doi.org/10.1109/
ICOSEC51865.2021.9591962
Rokade A, Singh M, Arora SK, Nizeyimana E (2022a) IOT-based medical informatics farming
system with predictive data analytics using supervised machine learning algorithms. Comput
Mathemat Methods Med:8434966, 15 pages. https://doi.org/10.1155/2022/8434966
Rokade A, Singh M, Malik PK, Singh R, Alsuwian T (2022b) Intelligent data analytics framework
for precision farming using IoT and regressor machine learning algorithms. Appl Sci 12(19):
9992. https://doi.org/10.3390/app12199992
Roy SK, Singh M, Sharma KK, Bhargava C, Singh BP (2020) Mathematical modelling of simple
passive RC filters using floating admittance technique. 2020 IEEE international conference for
innovation in technology (INOCON), pp 1–6, doi: https://doi.org/10.1109/INOCON50539.
2020.9298230
Sadowski S, Spachos P (2018) Solar-powered smart agricultural monitoring system using internet
of things devices, 2018 IEEE 9th annual information technology, electronics and Mobile
communication conference (IEMCON), Vancouver, BC, pp 18–23, doi: https://doi.org/10.
1109/IEMCON.2018.8614981
Sekaran K, Meqdad MN, Kumar P, Rajan S, Kadry S (2020) Intelligent agriculture management
system using internet of things. TELKOMNIKA (Telecommun Comput Electron Contr) 18:
1276–1285. https://doi.org/10.12928/TELKOMNIKA.v18i3.14029
Sethy P, Behera S, Pandey C, Narayanand S (2021) Smart paddy field monitoring system using
deep learning and IoT. Concurr Eng Res Appl 29(1). https://doi.org/10.1177/
1063293X21988944
Siddiquee KN-E-A, Islam M, Singh N, Gunjan V, Yong W, Huda M, Naik D (2022) Development
of algorithms for an IoT-based intelligent agriculture monitoring system. Wireless Commun
Mob Comput 2022:1–16. https://doi.org/10.1155/2022/7372053
Singh M, Gandam A (2022) Link restoration and relay node placement in partitioned wireless
sensor network. Design Dev Effic Energy Syst 4:101–117. https://doi.org/10.1002/
9781119761785.ch7
Singh B, Verma HK (2022a) Dawn of big data with Hadoop and machine learning. In: Agarwal P
et al (eds) Machine learning and data science: fundamentals and applications. Wiley, pp 47–65
Singh B, Verma H (2022b) EMM: extended matching market based scheduling for big data
platform Hadoop. Multimed Tools Appl 81:34,823–34,847. https://doi.org/10.1007/s11042-
021-11283-3
Singh M, Kumar M, Malhotra J (2018) Energy efficient cognitive body area network (CBAN) using
lookup table and energy harvesting. J Intellig Fuzzy Syst 35(2):1253–1265
Subahi A, Bouazza KE (2020) An intelligent IoT-based system design for controlling and moni-
toring greenhouse temperature. IEEE Access:1. https://doi.org/10.1109/ACCESS.2020.
3007955
Suma V (2021) Internet of Things (IoT) based intelligent agriculture in India: an overview. J
ISMAC 03(01):1–15, http://irojournals.com/iroismac/. https://doi.org/10.36548/jismac.2021.
1.001
Tageldin A, Adly D, Mostafa H, Mohammed H (2020) Applying machine learning technology in
the prediction of crop infestation with cotton leafworm in greenhouse. Int J Appl Earth Observ
Geoinform 105:102608. https://doi.org/10.1101/2020.09.17.301168
Tan E-K, Chong Y-W, Niswar M, Ooi BY, Basuki A (2020) An IoT platform for urban farming,
2020 International seminar on intelligent technology and its applications (ISITIA), Surabaya,
Indonesia, pp. 51–55, doi: https://doi.org/10.1109/ISITIA49792.2020.9163781
Walia GS, Singh P, Singh M (2021) Localizing Mobile nodes in WSNs using dragonfly algorithm.
In 2021 International conference on computing sciences (ICCS), pp. 223–227. IEEE
5 Analytics and Decision-making Model Using Machine Learning for Internet. . . 91

Walia, G. S., Singh, P., Singh, M., Abouhawwash, M., Park, H. J. et al. (2022a). Three-dimensional
optimum node localization in dynamic wireless sensor networks. CMC-Comput Mater Con-
tinua, 70(1), 305–321. doi:https://doi.org/10.32604/cmc.2022.019171
Walia, Singh G, Singh P, Singh M (2022b) Localizing mobile nodes in WSNs using neural network
algorithm. Mater Today Proc. https://doi.org/10.1016/j.matpr.2022.06.153
Chapter 6
DistilBERT-based Text Classification
for Automated Diagnosis of Mental Health
Conditions

Diwakar and Deepa Raj

Abstract Mental health disorders present a substantial global health challenge,

underscoring the critical importance of timely diagnosis and intervention for effec-
tive treatment. However, the proliferation of textual data within online mental health
communities has underscored the need for automated methods that can classify and
diagnose individuals based on their expressed thoughts and emotions. Existing
diagnostic practices for these conditions primarily rely on labour-intensive and
time-consuming clinical assessments and interviews conducted by healthcare pro-
fessionals. Notably, a key challenge in this context is the absence of robust physi-
ological indicators for mental disorders. This paper addresses this challenge by
proposing a DistilBERT-based text classification approach for the automated diag-
nosis of mental health conditions. In addition, recognizing the emerging landscape
of interdisciplinary exploration into the gut brain axis, we acknowledge the potential
role of gut bacteria, the microbiome, and microorganisms in influencing mental
health. Our research specifically focuses on three distinct mental health disorder
conditions: anxiety, borderline personality disorder (BPD), and autism. To ensure
the robustness of our approach, we curated a balanced dataset, comprising 500 sam-
ples for each of these three classes. This diligent effort yielded a noteworthy
achievement, with our model attaining a remarkable accuracy rate of 96%.

Keywords BERT · Classification · Mental disorder · Deep learning · Transformers

6.1 Introduction

Mental health disorders, encompassing conditions such as anxiety, borderline per-

sonality disorder (BPD), and autism, affect millions of individuals worldwide,
impacting their daily lives and overall well-being. Timely diagnosis and personal-
ized treatment are pivotal for improving the quality of life for those affected by these

Diwakar (✉) · D. Raj

Babasaheb Bhimrao Ambedkar University (A Central University), Lucknow, India, Uttar
Pradesh

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 93
A. Khamparia et al. (eds.), Microbial Data Intelligence and Computational
Techniques for Sustainable Computing, Microorganisms for Sustainability 47,
https://doi.org/10.1007/978-981-99-9621-6_6
94 Diwakar and D. Raj

conditions. The growing popularity of online mental health communities and social
media platforms has led to an unprecedented volume of user-generated text data.
Individuals share their thoughts, emotions, and experiences related to mental health
openly on these platforms. While this creates an opportunity for early detection and
support, it also presents a formidable challenge—how to effectively analyze and
classify this vast textual data to assist in the diagnosis of mental health disorders.
Emerging studies have begun to uncover intriguing connections between the com-
position of gut bacteria, known as the gut microbiome, and mental health conditions
such as anxiety and depression. Although this paper primarily delves into text
classification, it is situated within the broader landscape of mental health research,
where interdisciplinary exploration into the gut brain axis holds promise for new
insights and therapeutic approaches.
In this paper, we present a novel approach based on DistilBERT, a lightweight
variant of the BERT (Bidirectional Encoder Representations from Transformers)
model.
DistilBERT has demonstrated exceptional capabilities in natural language under-
standing and can efficiently process large volumes of text data. Unlike the original
BERT model, DistilBERT offers similar performance while significantly faster and
requiring fewer computational resources. This makes it an ideal choice for real-time
or resource-constrained applications, such as mental health disorder diagnosis. The
proposed methodology aims to address the challenges associated with the diagnosis
of mental health disorders using online text data. Used a dataset of three classes
namely Anxiety, BPD, and Autism to train and evaluate our model. Through
rigorous experimentation, we achieved an impressive accuracy of 96%, showcasing
the effectiveness of our approach in accurately classifying mental health conditions
from user-generated textual content. This high level of accuracy demonstrates the
potential of DistilBERT-based text classification in assisting healthcare profes-
sionals and support organizations in identifying individuals who may benefit from
early intervention and specialized care. The fundamental architecture of the BERT
model is displayed in Fig. 6.1. In the following sections, we provide an in-depth
description of our methodology, simulation setup, and detailed results in mental
health diagnosis.

6.2 Related work

The classification of mental health disorders from textual data has been a subject of
increasing interest in recent years. Researchers have recognized the potential of
natural language processing (NLP) and machine learning techniques in aiding
mental health diagnosis and support. In this section, we review the existing literature
on mental health disorder classification and NLP-based approaches, providing
insights into the advancements and challenges in this field.
Previous studies have explored various methodologies for classifying mental
health disorders based on textual content. Peng et al. (2019) introduced a multi-
6 DistilBERT-based Text Classification for Automated Diagnosis of. . . 95

Prediction

Classification

Encoder
NN NN NN

Self-Attaintion
Encoder
Encoder

Encoder

[CLS] This is [SEP]

Fig. 6.1 In the fundamental architecture of BERT, which is commonly employed for various
natural language processing tasks, including classification, the special tokens [CLS] and [SEP] play
a critical role. These tokens have distinct purposes within the model. The [CLS] token is placed at
the beginning of the input sequence, followed by the content to be classified, and the [SEP] token is
added at the end. Notably, the [CLS] token takes on the responsibility of encapsulating the entire
sequence, making it particularly vital for classification tasks in the BERT framework

kernel SVM approach, which achieved a better accuracy of 83.46% for depression
recognition. Amanat et al. (2022) presented a powerful model combining LSTM
with two hidden layers and a large bias, alongside an RNN with two dense layers, to
predict depression from text. This model achieves an impressive accuracy of 99%
with binary classification and has the potential to help safeguard individuals against
mental disorders and suicidal tendencies. Researchers have employed traditional
machine learning algorithms, such as Support Vector Machines (SVM) and Random
Forests, as well as deep learning architectures, including Convolutional Neural
Networks (CNNs) and recurrent models like Long Short-Term Memory (LSTM)
networks. These approaches have shown promise in distinguishing between differ-
ent mental health conditions but often require substantial feature engineering and
may struggle with capturing the nuanced semantics of text.
Nova (2023) proposed a powerful model combining LSTM with two hidden
layers and a large bias, alongside an RNN with two dense layers, to predict
depression from text. This model achieved 77% with the LightGBM model for
classification based on the text content. Kour and Gupta (2022) proposed a hybrid
CNN-biLSTM model and achieved a 94.28% accuracy. Uddin et al. (2022)
presented Long Short-Term Memory (LSTM)-based.
RNN to efficiently identify self-perceived symptoms of depression in text data,
outperforming traditional word frequency-based methods. The approach offers the
potential for improving mental healthcare technologies like intelligent chatbots.
Murarka et al. (2020) utilized the RoBERTa model for mental illness and achieved
96 Diwakar and D. Raj

89% of accuracy. Kim et al. (2020) presented a CNN model and achieved 96.96%
with CNN and 94.91% with XGBoost for binary class classification. Recent
advancements in NLP, fueled by pre-trained language models like BERT and its
variants, have significantly improved the accuracy and efficiency of mental health
disorder classification (Diep et al. 2022; Kim et al. 2020; Kour and Gupta 2022;
Vaswani et al. 2017; Zeberga et al. 2022). These models excel at learning contextual
representations of text, enabling them to capture the subtle nuances of language and
the context in which mental health discussions occur. Our work is firmly rooted in
this advancement, specifically leveraging the DistilBERT architecture (Uddin et al.
2022) a lightweight variant of BERT (Vaswani et al. 2017) that balances computa-
tional efficiency with performance.
Several gaps remain unaddressed. These gaps include the need for lightweight
models that can be deployed in resource-constrained environments, the exploration
of diverse datasets representing multiple mental health conditions, and the assess-
ment of model performance in real-world scenarios. Our study aims to bridge these
gaps by presenting a comprehensive approach that combines a lightweight model
with a diverse dataset, resulting in accurate and practical classification for improved
mental health diagnosis and support.

6.3 Dataset and Prepossessing

6.3.1 Dataset

To conduct our research on automated mental health disorder classification, we

utilized a comprehensive dataset sourced from the Kaggle repository (Ameer et al.
2022). This dataset comprises a diverse collection of posts from online mental health
communities, offering a rich and varied source of text data for our analysis. The
dataset consists of 500 samples from each class, representing a wide range of user-
generated content (Adha 2022). The dataset includes posts related to various mental
health conditions, with labels corresponding to three distinct classes: Anxiety,
Borderline Personality Disorder (BPD), and Autism. Each post contains textual
content, often accompanied by user-generated titles.

6.3.2 Prepossessing

To prepare our dataset for analysis, an extensive preprocessing pipeline is applied,

comprising several crucial steps as shown in Table 6.1. Additionally, these
preprocessing methods standardized and cleaned the text data, ensuring its suitability
for subsequent analysis.
6 DistilBERT-based Text Classification for Automated Diagnosis of. . . 97

Table 6.1 Text preprocessing methods

Method Description
URL Removal Elimination of web links using regular expressions
Newline Removal Removal of newline characters for consistent formatting
Lowercasing Conversion of text to lowercase for case-insensitive processing
HTML Tag Removal Removal of HTML tags using BeautifulSoup
Whitespace cleanup Reduction of extra whitespaces to a single space
Bracket Removal Removal of various bracket types
Quote Removal Elimination of single and double quotes
Special Character Removal Removal of non-semantic special characters
Chars with Digits Removal Elimination of character–digit combinations
Digit Removal Removal of standalone digits
Contractions Expansion Expansion of contractions to full forms
Stopword Removal Elimination of common English stopwords
Lemmatization Reduction of words to their base forms (lemmas)

6.4 Methodology

The choice of DistilBERT (Uddin et al. 2022) as the Natural Language Processing
(NLP) model for this study is grounded in its remarkable performance and effi-
ciency. DistilBERT, a lightweight variant of the BERT (Bidirectional Encoder
Representations from Transformers) model, offers several advantages for our task
of mental health disorder classification. DistilBERT provides comparable
performance to the original BERT model but with significantly fewer parameters
(66 million), making it computationally efficient and suitable for real-time or
resource-constrained applications, which are crucial in the context of mental health
support. Furthermore, DistilBERT excels in capturing the contextual understanding
of the text, allowing it to comprehend the nuanced semantics of user-generated
content related to mental health discussions. This contextual awareness is essential
for accurate classification. Leveraging pretrained representations on massive text
corpora, similar to BERT, DistilBERT is equipped with a broad understanding of
language and concepts. Fine-tuning such a pretrained model on our mental health
dataset enhances its classification capabilities.
To train and evaluate our DistilBERT-based model, we employed a standard data
splitting technique, dividing our dataset into training and validation sets. In this
process, 80% of the dataset is allocated to the training set, enabling our model to
learn from a diverse range of textual data representing different mental health
conditions. The remaining 20% of the dataset served as the validation set, ensuring
an evaluation of the model’s performance on unseen data and assessing its general-
ization capabilities. Our training procedure adhered to established best practices in
deep learning and NLP. We utilized the AdamW optimizer for parameter updates
during training, combining the advantages of the Adam optimizer with weight decay
to ensure robust convergence. A learning rate of 2e-5 is selected based on empirical
experimentation and fine-tuning, striking a balance between rapid convergence and
98 Diwakar and D. Raj

Fig. 6.2 Architecture for the classification of mental health disorders

stability. The model underwent training over 20 epochs, a choice made after careful
monitoring of training loss and validation performance to prevent overfitting. The
architecture of the model is displayed in Fig. 6.2.
6 DistilBERT-based Text Classification for Automated Diagnosis of. . . 99

To further enhance DistilBERT’s performance in classifying mental health dis-

orders, we conducted model fine-tuning. This process involved iterative adjustments
of hyperparameters, such as the learning rate, batch size, and dropout rates.
Ad-ditionally, we explored strategies to address the class imbalance, including
oversampling and weighted loss functions, to ensure fair and accurate classification
across all mental health conditions. Our choice of DistilBERT as the Natural
Language Processing (NLP) model for this study is grounded in its remarkable
performance and efficiency. DistilBERT, a lightweight variant of the BERT (Bidi-
rectional Encoder Representations from Transformers) model, offers several advan-
tages for our task of mental health disorder classification. DistilBERT provides
comparable performance to the original BERT model but with significantly fewer
parameters, making it computationally efficient and suitable for real-time or
resource-constrained applications, which are crucial in the context of mental health
support. DistilBERT achieves its efficiency by employing knowledge distillation. It
is trained to mimic the behaviour of the larger BERT model by learning from its
outputs. The core architecture of DistilBERT consists of a stack of transformer
layers, each composed of multi-head self-attention mechanisms and feedforward
neural networks. Mathematically, it can be represented as follows in Eq. (6.1).

DistilBERTðXÞ
= LayerNormðMultiHeadSelfAttentionðXÞ þ X þ FeedForwardðXÞ þ XÞ
ð6:1Þ

This equation represents the forward pass through one transformer layer in
DistilBERT. X represents the input embeddings, and the operations include multi-
head self-attention and feedforward neural networks. Furthermore, DistilBERT
excels in capturing the contextual understanding of the text, allowing it to compre-
hend the nuanced semantics of user-generated content related to mental health
discussions. This contextual awareness is essential for accurate classification. The
contextual understanding in DistilBERT is achieved through multi-head self-atten-
tion mechanisms, which mathematically compute weighted sums of input embed-
dings based on their relevance to each other. This is expressed as follows Eq. (6.2):

XW Q ðXÞT
SelfAttentionðXÞ = softmax p XW V ðXÞ ð6:2Þ
dk

Here, X is the input, and WQ and WV are learned weight matrices. The softmax
function calculates attention scores, and the weighted sum provides contextualized
embeddings.
Leveraging pretrained representations on massive text corpora, similar to BERT,
DistilBERT is equipped with a broad understanding of language and concepts. Fine-
tuning such a pretrained model on our mental health dataset enhances its classifica-
tion capabilities. Pretraining involves training a transformer model to predict masked
words in a large text corpus. The pretrained model is characterized by a vocabulary
100 Diwakar and D. Raj

V, input embeddings E, and transformer parameters Θ. Fine-tuning on our dataset

involves minimizing a classification loss L by using stochastic gradient descent. This
is expressed as Eq. (6.3):

N
1
min LðDistilBERTðXi ; ΘÞ, yi Þ ð6:3Þ
Θ N i=1

where N is the number of training examples. Xi is the ground truth label for the ith
training example. L is a suitable classification loss function, such as cross-entropy.
In the training phase of DistilBERT, the primary objective is to minimize a
classification loss function L over a labelled dataset. Given a dataset of N training
examples {(Xi, yi)}, where Xirepresents the input text and yi represents the ground
truth label.

6.4.1 Batch Training

During training, it is common to use batch training, where the dataset is divided into
smaller batches of size B. The optimization process updates the model’s parameters
using gradients computed over these batches. Mathematically, the update step for the
model parameters Θ can be expressed as Eq. (6.4):

B
1
Θ←Θ-α  ∇Θ LðDistilBERTðXi ; ΘÞ, yi Þ ð6:4Þ
B i=1

Θ represents the model parameters. α is the learning rate. B is the batch size. ∇Θ
denotes the gradient with respect to the model parameters

6.4.2 Hyperparameter Tuning

In the fine-tuning process, various hyperparameters are tuned to optimize the

model’s performance. These hyperparameters include:
Learning Rate (α): In this work, the learning rate (α) is set to value of 2e-5 using
empirical experimentation. This learning rate controls the step size in gradient
descent and plays a crucial role in convergence speed and stability.
Batch Size (B): We utilized a batch size B of 32 during training. The choice of
batch size affects training efficiency and memory usage.
Dropout Rates: Dropout rates were applied as a regularization technique in our
project, with a specific dropout rate of 0.2. These rates were tuned to control the
amount of dropouts during training and mitigate overfitting.
6 DistilBERT-based Text Classification for Automated Diagnosis of. . . 101

6.4.3 Algorithm

The BERT-based mental health classification algorithm involves rigorous text

pre-processing and model setup, followed by data splitting, training, and evaluation,
culminating in classification results for mental health assessment as shown in
Algorithm 6.1.
Algorithm 6.1 BERT-based Mental Health Classification Algorithm Step 1:
Data Preprocessing
1. Remove URLs.
2. Eliminate newline characters.
3. Convert text to lowercase.
4. Remove HTML tags.
5. Clean up extra whitespaces.
6. Remove brackets, quotes, and special characters.
7. Expand contractions.
8. Remove stopwords.
9. Apply lemmatization.
10. Remove character digit combinations and digits.
Step 2: Model Initialization
1. Initialize a DistilBERT model for sequence classification.
2. Tokenize and encode text data.
3. Convert labels to numerical values.
Step 3: Data Splitting
1. Split data into training and validation sets.
Step 4: Model Training
1. Initialize AdamW optimizer (learning rate: 2 × 10^(-5)).
2. Create data loaders for training and validation.
3. Train the model for 20 epochs
Step 5: Model Evaluation
1. Evaluate the trained model on validation data.
2. Calculate accuracy, precision, recall, and F1 score.
Step 6: Output
1. Trained DistilBERT model for mental health post-classification.
2. Classification report on validation data.
102 Diwakar and D. Raj

6.4.4 Evaluation Standards

1. Accuracy: Accuracy is a commonly used evaluation metric for classification

tasks. It measures the proportion of correctly classified instances among all
instances. Mathematically, accuracy can be expressed as:

NumberofCorrectPredictions
Accuracy = ð6:5Þ
TotalNumberofPredictions

2. Precision: measures the accuracy of positive predictions. It calculates the

proportion of true positive predictions (correctly predicted positive instances)
among all instances predicted as positive. Mathematically, precision can be
expressed as

TruePositives
Precision = ð6:6Þ
TruePositives þ FalsePositives

3. Recall: Recall also known as sensitivity or true positive rate, quantifies the
model’s ability to identify positive instances correctly. It calculates the propor-
tion of true positive pre- dictions among all actual positive instances. Mathe-
matically, recall can be expressed as

TruePositives
Recall = ð6:7Þ
TruePositives þ FalseNegatives

4. F1 Socre: The F1 score is the harmonic mean of precision and recall. It provides
a balanced measure that takes both false positives and false negatives into
account. Mathematically, the F1 score can be expressed as:

2  Precision  Recall
F1 score = ð6:8Þ
Precision þ Recall

5. Confusion Matrix (CM): A confusion matrix is a table that summarizes the

model’s classification performance. It provides counts of true positives, true
negatives, false positives, and false negatives. From the confusion matrix,
various metrics like precision and recall can be calculated.

6.5 Simulation Results and Discussion

In this section, we outline the key aspects of our simulation setup.

6 DistilBERT-based Text Classification for Automated Diagnosis of. . . 103

6.5.1 Model Training

The model is trained over 20 epochs with the AdamW optimizer, utilizing a learning
rate of 2e-5, and during training, cross-entropy loss is computed. The training
process yielded promising results, as evidenced by the classification report. The
model achieved an overall accuracy of 96%. Notably, it exhibited excellent perfor-
mance in classifying Anxiety, achieving a precision of 94% and recall of 99%,
highlighting its proficiency in identifying this specific mental health subreddit. These
findings underscore the model’s effectiveness in categorizing posts related to various
mental health conditions. The average loss over the epochs is presented in Table 6.2.

6.5.2 Text Data Distribution Analysis

In our analysis, utilized kernel density estimation (KDE) with seaborn to visualize
the distribution of total words and characters in the dataset. These KDE plots,
rendered in a winter-themed colour palette, shed light on the composition of text
data across different mental health subreddits. The shaded KDE plots provide
insights into potential differences in content length among the analyzed subreddits
as shown in Fig. 6.3.

Table 6.2 Average loss over Epoch Average Loss

epochs
1/20 0.9574
2/20 0.2829
3/20 0.1102
4/20 0.0553
5/20 0.0381
6/20 0.0272
7/20 0.0278
8/20 0.0093
9/20 0.0049
10/20 0.0037
11/20 0.0030
12/20 0.0025
13/20 0.0022
14/20 0.0019
15/20 0.0016
16/20 0.0014
17/20 0.0013
18/20 0.0011
19/20 0.0011
20/20 0.0010
104 Diwakar and D. Raj

(a) KDE Plot of Total Words (b) KDE Plot of Total Characters

Fig. 6.3 Text data characteristics analysis. (a) KDE plot of total words. (b) KDE plot of total
characters

Fig. 6.4 Confusion matrix

6.5.3 Confusion Matrix

In the paper, we performed resampling on the validation data to ensure an equal

number of samples in each class, addressing class imbalance. The resulting confu-
sion matrix plot illustrates the model’s classification performance, with labels
‘anxiety,’ ‘BPD,’ and ‘autism’ prominently displayed for easy interpretation in
Fig. 6.4.

6.5.4 Visualizing Word Clouds

The code generates word cloud visualizations for ‘anxiety,’ ‘bpd,’ and ‘autism’
subreddits, offering insights into their prevalent terms. Each word cloud showcases
6 DistilBERT-based Text Classification for Automated Diagnosis of. . . 105

Fig. 6.5 Visualizing word clouds

Table 6.3 Performance com- Method Precision Recall F1 Score Accuracy

parison of classification
TF-IDF 0.85 0.93 0.89 0.91
methods
TinyBERT 0.96 0.90 0.93 0.91
DistilBERT 0.94 0.97 0.95 0.96

frequently occurring words, aiding in understanding the distinctive textual content of

these mental health communities. Figure 6.5 depicts the visualization of word clouds.
In this comprehensive analysis of mental health-related discussions on Reddit,
our study integrated advanced natural language processing techniques and machine
learning to categorize and understand user-generated content across three major
mental health subreddits: ‘anxiety,’ ‘BPD’ (borderline personality disorder), and
‘autism.’ By utilizing a DistilBERT-based classification model, we achieved high
accuracy in classifying posts into these categories. Additionally, we addressed class
imbalance issues through resampling techniques, ensuring robust model perfor-
mance. Furthermore, we visualized word clouds to provide intuitive insights into
the most prevalent terms within each subreddit, shedding light on the unique
language patterns and themes that characterize these online mental health commu-
nities. It’s worth noting that we conducted comparative experiments with other
methods such as TF IDF with SVM and TinyBERT, and the results, along with
detailed comparisons, are presented in Table 6.3, demonstrating the effectiveness of
the DistilBERT model in capturing nuanced textual features. This research not only
enhances our understanding of how individuals discuss mental health on social
platforms but also demonstrates the potential for machine learning in assisting
mental health professionals and researchers in monitoring and providing support in
online communities.

6.6 Conclusion

In this study, we harnessed advanced natural language processing techniques to

analyze mental health discussions across 'anxiety,' 'BPD' (borderline personality
disorder), and 'autism' subreddits. Leveraging the DistilBERT model, we achieved
an impressive 96% accuracy in classifying posts, showcasing its efficacy in captur-
ing nuanced text features. Addressing class imbalance concerns and visualizing
word clouds provided valuable insights into these online mental health communities.
106 Diwakar and D. Raj

Looking ahead, expanding this analysis to include more subreddits, incorporating

sentiment analysis, and developing real-time support systems are promising direc-
tions. Enhancing model interpretability and collaborating with mental health experts
can further leverage this research for practical applications in mental health support.

References

Adha K (2022) Mental disorders identification (Reddit), Kaggle, Version 2, 4579285

Amanat A et al. (2022) Deep learning for depression detection from textual data. Electronics 11
(5):676
Ameer I et al. (2022) Mental illness classification on social media texts using deep learning and
transfer learning. In: arXiv preprint arXiv:2207.01012
Ansari G, Garg M, Saxena C (2021) Data augmenta- tion for mental health classification on social
media. In: arXiv preprint arXiv:2112.10064
Diep B, Stanojevic M, Novikova J (2022) “Multi-modal deep learning system for depression and
anxiety detection”. In: arXiv preprint arXiv:2212.14490
Kim J et al (2020) A deep learning model for detecting mental illness from user content on social
media. Sci Rep 10(1):11,846
Kour H, Gupta MK (2022) An hybrid deep learning approach for depression prediction from user
tweets using feature-rich CNN and bi-directional LSTM. Multimed Tools Appl 81(17):
23,649–23,685
Murarka A, Radhakrishnan B, Ravichandran S (2020) Detection and classification of mental
illnesses on social media using RoBERTa. In: arXiv preprint arXiv:2011.11226
Nova K (2023) Machine learning approaches for automated mental disorder classification based on
social media textual data. Contemp Issu Behav Social Sci 7(1):70–83
Peng Z, Qinghua H, Dang J (2019) Multi-kernel SVM based depression recognition using social
media data. Int J Mach Learn Cybern 10:43–57
Sanh V et al. (2019) DistilBERT, a distilled version of BERT: smaller, faster, cheaper and
lighter. In: arXiv preprint arXiv:1910.01108
Tran T, Kavuluru R (2017) Predicting mental conditions based on “history of present illness” in
psychiatric notes with deep neural networks. J Biomed Inform 75:S138–S148
Uddin MZ et al (2022) Deep learning for prediction of depressive symptoms in a large textual
dataset. Neural Comput Appl 34(1):721–744
Vaswani A et al. (2017) Attention is all you need. In Advances in neural information processing
systems, vol 30
Zeberga K et al (2022) A novel text mining approach for mental health prediction using Bi-LSTM
and BERT model. Comput Intell Neurosci 2022:7893775
Chapter 7
An Optimized Hybrid ARIMA-LSTM
Model for Time Series Forecasting
of Agricultural Production in India

Babita Pandey, Arvind Shukla, and Aditya Khamparia

Abstract The role of agriculture in a country’s national income is very important.

The zero-hunger sustainable goal of the United Nations cannot be achieved without
substantial increase in agricultural production. The value of agricultural production
shows the health of agriculture and its prediction has long been a challenge for
academicians due to highly uncertain weather conditions as well as emergence of
microbacteria due to variable soil and weather conditions. This chapter is focused on
the prediction methodology for measuring accurate prediction of the value of
agricultural production. The data of the gross value-added series for the years
1950–2021 for the agriculture and allied sectors has been taken for prediction. The
traditional model autoregressive integrated moving average (ARIMA) and machine
learning and the deep learning-based long short-term memory (LSTM) model have
been compared with an optimized hybrid model comprising characteristics of both
models. The efficiency of the models was checked by using root mean square error
(RMSE). The result was compared with other studies on hybrid models. It has been
reported that the hybrid model is more appropriate in comparison to both ARIMA
and LSTM. The hybrid model provides the lowest RMSE among the three models.
The ARIMA model gives lower RMSE in comparison to LSTM. The hybrid model
provides 11% more accurate prediction in comparison to ARIMA. The prediction
result of the hybrid model is highly significant and gives within the 95% confidence
interval. The results are important and useful for the prediction of the value of
agricultural production and policy formulation to meet the sustainable development
goal of zero hunger.

B. Pandey (✉)
Babasaheb Bhimrao Ambedkar (Central) University, Lucknow, Uttar Pradesh, India
Department of Computer Science, Babasaheb Bhimrao Ambedkar (Central) University,
Lucknow, Uttar Pradesh, India
A. Shukla
Babasaheb Bhimrao Ambedkar (Central) University, Lucknow, Uttar Pradesh, India
A. Khamparia
Department of Computer Science, Babasaheb Bhimrao Ambedkar University, Amethi, India

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 107
A. Khamparia et al. (eds.), Microbial Data Intelligence and Computational
Techniques for Sustainable Computing, Microorganisms for Sustainability 47,
https://doi.org/10.1007/978-981-99-9621-6_7
108 B. Pandey et al.

Keywords Time series · Agriculture · Micro bacterial · Machine learning ·

ARIMA · LSTM

7.1 Introduction

Agriculture is the most important sector of the Indian economy. It is vital not only for
feeding the people of the country but also for economic development. It helps in the
development of the other sectors of the economy through its forward and backward
linkages. Agricultural production is one of the important indicators for measuring the
development of agriculture. The value of the agricultural production shows the
health of agriculture. However, its prediction has long been a challenge for acade-
micians due to highly uncertain weather conditions as well as emergence of
microbacteria due to variable soil and weather conditions. The prediction of agri-
cultural production will help policy makers to devise approaches to meet the
growing demand for food due to increase in the population. Across the globe around
815 million people suffer from chronic hunger and one in three people is malnour-
ished (UNDP 2023). It is one of the sectors that needs the greatest focus to achieve
zero hunger. The production of agriculture depends upon the availability of irriga-
tion, rainfall, yield, soil fertility, natural calamities, microbes, and climatic condi-
tions. Considering many of the fixed factors, agricultural production requires gradual
improvement in the land qualities and seed quality. Thus, agricultural production has
certain limitations. However, Indian agricultural production has increased due to
technological development from approximately 50 million tons to more than
300 million tons, but despite all these improvements, the Sustainable Development
Report 2023 placed India in the category of the countries with average prevalence of
undernourishment at 25–40%during 2020–2022. It has also been reported that this
prevalence is attributed to the interplay of two opposing forces: income and inflation.
Income is increased but its effect is eroded due to price inflation (UNDP 2023).
Therefore, prediction of agricultural production and its values is imperative. Predic-
tion of the value of agricultural products is quite cumbersome due to the volatility of
income and prices. The nonlinear and complicated trends of the agricultural market
make it difficult to accurately predict the value of agricultural products (Hauchet-
Bourdon 2011). The share of gross value added (GVA) of agriculture and allied
sectors in the total economy has decreased during the last 3 years from 20.3 in
2020–2021 to 19.0 in 2021–2022 and subsequently to 18.3 in 2022–2023 (PIB
2023). Considering the sustainable development goals, decreasing shares of agricul-
ture and allied sectors in GVA poses significant challenges. Prediction, therefore,
will help in policy formulations for achieving the sustainable development goals.
There are various methods of prediction. Autoregressive moving average
(ARMA) and autoregressive integrated moving average (ARIMA) are the traditional
methods used for prediction with univariate time series data. With the advent of new
techniques like machine learning (ML) and deep learning (DL), preference is starting
to be given to the techniques of neural networks that include support vector
machines (SVM), Random Forest (RF), recurrent rural networks (RNN), and long
7 An Optimized Hybrid ARIMA-LSTM Model for Time Series Forecasting. . . 109

short-term memory (LSTM). The main aim of this chapter is to discuss the tech-
niques that can be used for the prediction of agriculture output in more accurate ways
among the existing traditional techniques vis a vis available new ML and DL
techniques. The ARIMA and LSTM techniques, considering the space limitation
of the work, were tested for prediction. Different views on the prediction by these
models are available. Prediction with ARIMA is efficient in a shorter period but as
the time period increases the precision rate decreases. However, LSTM is quite
efficient in making predictions even if there is less previous data (Hua 2020). The
ARIMA model was devised primarily by Box and Jenkins (Box and Jenkins 1976).
The ARIMA model has been transformed into various versions like seasonal
ARIMA, called SARIMA, and multiplicative SARIMA, called MSARIMA
(Kalpakis et al. 2000). Although the ARIMA model is widely used for prediction
of univariate time series linear models, its optimal selection of parameters requires
substantial computational time for suitable model selection, and this is considered a
disadvantage (Nguyen et al. 2019). However, Sima et al. observed that LSTM is
superior to ARIMA as it reduces the error rate by 85% (Siami-Namini et al. 2018). In
contrast, Deng et al. found that the hybrid model synthesizes the characteristics of all
individual models and therefore provides better predictions in comparison to the
ARIMA and LSTM models (Deng et al. 2023). Naveen et al. evaluated the models
for forecasting the price of washed coffee and found that the hybrid models can
combine both linear and non-linear time series and hence improve the efficiency of
forecasting (Naveena et al. 2017). Deng et al. also reported similar results that hybrid
models provide better forecast in comparison to ARIMA or LSTM alone (Deng et al.
2023). They found that hybrid models provide the lowest root mean square error,
mean absolute percentage error (MAPE), and mean absolute error (MAE), corrob-
orating the findings of Khazee et al. (Khazee et al. 2019). Similarly, Deve et al. noted
that the hybrid model provides better results. They formulated the hybrid model by
integrating LSTM for non-linear components and ARIMA for the linear components
of the data set (Dave et al. 2021). Kumar et al. also created a hybrid model that
combines the non-Gaussian time series, i.e., beta seasonal ARIMA, with the LSTM
and concluded on the basis of an experimental data set that this hybrid model
provides better prediction accuracy (Sunil and Yadav 2023).
Prediction of the value of agricultural products is quite cumbersome due to
volatility in income and prices. The non-linear and complicated trends of the
agricultural market make it difficult to accurately predict the value of agricultural
products (Hauchet-Bourdon 2011). Tae-woong and Seok proposed an Seasonal long
short-term memory (SLSTM) model for the prediction of the sales of agricultural
products (Yoo and Oh 2020).
Thus, this chapter is largely focused on the prediction methods of agricultural
production using the techniques available for predicting based on time series data.
The chapter is further divided into three sections, namely Materials and Methods
(Sect. 7.2), Results and Discussion (Sect. 7.3), and Conclusion (Sect. 7.4).
Section 7.2 elaborates the technique and methodology used for the prediction of
the value of agricultural production. Section 7.3 describes the results of the analysis
and issues linked with these results. Section 7.4 is a sketch of the findings of the
research analysis.
110 B. Pandey et al.

7.2 Materials and Methods

7.2.1 Data

We used the gross value added by the agriculture fishing and mining at current prices
on the base price year of 2011–2012. The data set from 1950 to 2021–2022 is
available from National Income Statistics. The choice is based on achieving higher
accuracy over a greater time period, which will help in the prediction of agricultural
production.

7.2.2 Stationary Test

The time series data set must be stable for further analysis including prediction.
Therefore, primarily the stationarity of the data set has been checked using the
augmented Dickey–Fuller (ADF) test. The augmented Dickey–Fuller test is pre-
ferred over the Dickey–Fuller test to remove the possibility of autocorrection in the
data test before predicting unit root hypothesis.

7.2.3 Autoregressive Integrated Moving Average (ARIMA)

The ARIMA model is used for prediction. The ARIMA model is a special version of
the ARMA model incorporating non-stationary series that become stationary
through differencing. It is manifested as ARIMA ( p,q,d ) wherein the p is the number
of autoregressive terms, q is the number of differences required to make the series
stationary, and d is the number of terms used for taking moving average. The
stationarity of the data set has been checked using the augmented Dicky Fuller
test, which is preferred over the Dickey–Fuller test to remove the possibility of
autocorrection in the data test before predicting unit root hypothesis. The following
ARIMA model is estimated:

yt = μ þ y t - 1 þ ε t
Δyt = Y t - yt - 1 = ð1 - LÞY t = μ þ εt

C ðLÞ ð1 - LÞd Y t = μ þ DðLÞεt

where C(L ) and D(L ) are polynomials in the lag operator and (1 - L)dYt = ΔdYt is
the dth difference of Yt.
7 An Optimized Hybrid ARIMA-LSTM Model for Time Series Forecasting. . . 111

Δd Y t = μ þ γ 1 Δd Y t - 1 þ γ 2 Δd Y t - 2 þ . . . :: þ γ p Δd Y t - p þ εt - θ1 Et - 1
- . . . . . . . . . . . . : - θq Et - q:

C(L) Yt = εt. The Yt denotes the value of the agricultural product in t time period.
γ 1, γ 2, γ 3. . . are the coefficients of the autoregressive (AR) model. θ1, θ2, θ3. . . are
the coefficients of the moving average. The dth difference produces the stationary
process and is called the order of integration. This implies that the series will be
called integrated of order one, i.e., I(1), if the first difference of non-stationary series
produces stationary series (Greene 2003). The ADF test is applied to check the
stationarity with the null hypothesis that γ = 0. The rejection of the null hypothesis
suggests that the series does not have a unit root and hence may be treated as
stationary. The only limitation of the ARIMA model is its ability to capture only
the linear relationship among the different time period values of the variable.

7.2.4 Long Short-term Memory (LSTM)

LSTM is a type of recurrent neural network (RNN) used in deep learning and
machine learning. It was firstly developed by Hochreiter and Schmidhuber
(Hochreiter and Schmidhuber 1997). LSTM is an efficient and gradient-based
algorithm that provides constant error terms without explosion or vanishing of
error flow. Its network topology carries three layers: input, hidden, and output layers.
The hidden layer connects the remaining two layers through memory cells and gate
units. There are three gates: forget gate, input gate, and output gate. Gate units work
as the valve and controls the flow of neuronal information. All three gates are
controlled by the sigmoid unit. The computation at each gate is shown in
Table 7.1. LSTM has an ability to hold the properties of non-linearity of data
while performing predictions. Therefore, the LSTM has an advantage over ARIMA.

Table 7.1 Computation at each gate in LSTM

Computation at gate Architecture of LSTM
Forget gate: ft = σ(Xi * Ut + Ht-1 * Wf)
Ct-1 * ft = 0. . . . If ft = 0
Ct-1 * ft = 0. . . . If ft = 1
Input gate: it = σ(Xt * Ui + Ht-1 * Wi)
Nt = tanh(Xt * Uc + Ht-1 * Wc)
Ct = ft * Ct-1 + it * Nt LSTM architecture
Output gate: Ot = σ(Xt * Uo + Ht-1 * Wo)
Ht = tanh(Ct); output = Softmax (Ht)
X Input, U Input weight matrix, H Hidden state, W Weight matrix associated with the hidden state,
t Time tamp, i Input gate, f Forget gate, c Cell state, o Output gate, Ct Long-term memory
112 B. Pandey et al.

7.2.5 Optimized Hybrid ARIMA-LSTM

A two-phase hybrid model is developed. In the first phase ARIMA and LSTM are
deployed independently for prediction. Optimal selection of parameters of ARIMA,
i.e., ( p, d, q), and LSTM (number of hidden layers, number of neurons in each layer)
requires substantial computational time for suitable model selection and hence
degrades the performance of the model. In this work, optimized ARIMA parameters
are identified using a random search technique. In the second phase, the output of
each model is compared with actual output and the smallest one is retained as final
prediction at time stamp t (Eqs. (7.1) and (7.2)).

HPi = min:ffor each t ðabs jP ARt - Actt j, Abs jP LSt - Actt jÞg ð7:1Þ
HP = HP1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . :HPlength of sequence ð7:2Þ

where the hybrid model prediction: HP; prediction by ARIMA: P_AR; Actual
prediction: Act; prediction by LSTM: P_LS; and t: time stamp. The result showed
that the optimum root mean square error (RMSE) of 39,035 is obtained by the hybrid
model. The hybrid model is shown in Fig. 7.1. The left branch of Fig. 7.1 shows the
ARIMA model computation flow and the right branch shows the LSTM architecture.

7.3 Results and Discussion

The experimentation was done in Python using spider editor. Python libraries like
keras and tensor were used for prediction through LSTM models. This study used
the statsmodels Python module for estimation of many different statistical models
including ARIMA for the analysis. The agricultural production shown in Fig. 7.2
depicts the value of agricultural production over the period. The value of production
increased steeply after 1990 and 2000. The effect of liberalization may be the reason
behind such growth, including the impact of the Green Revolution. The figure shows
that the series is non-stationary. The method of differencing has been adopted for
making it stationary. The results of the ADF test are shown in Table 7.2. It shows
that the data are non-stationary up to the second-order differencing. The p value for
the ADF test statistics is higher than 0.05, depicting the failure to reject the null
hypothesis that the data is non-stationary (as shown in Figs. 7.3 and 7.4). The third
differentiating provides the P < 0.5 for the estimates of the augmented Dickey–
Fuller test implying that the data set is stationary with third-order integration
(as shown in Fig. 7.5). The negative test statistics are also indicative of stationary
time series. The value of the lags used is the same as used for the second-order
differencing. Thus, the autoregressive integrated moving average (ARIMA) model is
suitable to use for the prediction. The optimized parameter (order) of ARIMA is
(0, 3, 2). Table 7.3 shows the result of the ARIMA model. We found that the
7 An Optimized Hybrid ARIMA-LSTM Model for Time Series Forecasting. . . 113

Fig. 7.1 Hybrid model

ARIMA statistics are 99% significant. This shows the approximately 9.9% growth at
the lag of 2 years observed over the period, which implies that the value of the
product doubled over almost 22 years. The residual errors seem fine with near zero
mean and uniform variance as shown in Fig. 7.6.
We applied an LSTM model for the same data set to check the efficiency and
accuracy of the prediction. The LSTM architecture comprised one input-one LSTM
with five units and one dense layer output. A batch size of 72 is used for each batch
of predictions and true outputs, and the mean squared error is calculated. The
optimizer Adam is used. Further, it is observed that the residuals obtained through
ARIMA and LSTM show a non-linear relationship. Therefore, the hybrid model was
applied on the data set and observed that the prediction accuracy of the hybrid model
is better than the ARIMA and LSTM. The result showed that root mean square error
(RMSE) for ARIMA is 59,627. Around 3.7% MAPE implies the model is about
96.3% accurate in predicting the next 15 observations. The RMSE for the hybrid
114 B. Pandey et al.

Fig. 7.2 Agri data

Table 7.2 Augmented Dickey–Fuller test results

First differentiation Second differentiation Third differentiation
ADF test statistic 5.878493 ADF test statistic 0.861765 ADF test statistic -
P-Value 1.000000 P-Value 0.992571 5.071515
# Lags used 12 Lags used 11.00 P-Value 0.000016
# Observations used 58 Observations used 58 # Lags used 11.00
Critical Value (1%) - Critical value (1%) - # Observations used 57
3.548494 3.548494 Critical value (1%) -
Critical value (5%) - Critical value (5%) - 3.550670
2.912837 2.912837 Critical value (5%) -
Critical value (10%) - Critical value (10%) - 2.913766
2.594129 2.594129 Critical value (10%) -
Is the time series stationary? Is the time series stationary? 2.594624
False False Is the time series stationary?
True

model is 52,734. The lowest RMSE of the hybrid model shows that the model has an
advantage in predicting the value of agricultural production over the ARIMA and
LSTM. Figures 7.7, 7.8, 7.9 show the comparative prediction of each model.
The prediction line of the LSTM model shows the higher deviation in comparison
to the prediction line of the ARIMA from the actual. The same is clearly visible in
the individual graphs shown in Fig. 7.7 and in the overlapping graph depicted in
Fig. 7.8. In economic analysis, the data science forecasting tools are assessed on the
basis of their ability to produce good forecasts, i.e., the forecast that will be closer to
the actual values.
For better visibility and comparison, an overlapping graph showing the prediction
line of all three models, ARIMA, LSTM, and hybrid, is presented in Fig. 7.8. The
accuracy in the prediction by the hybrid model over the two models is clearly visible
7 An Optimized Hybrid ARIMA-LSTM Model for Time Series Forecasting. . . 115

Fig. 7.3 After first differentiation

Fig. 7.4 After second differentiation

in the figure. RMSE square suggests that the hybrid model provides 11% more
accurate results in comparison to ARIMA models. Several studies reported similar
results for the prediction. Yung et al. reported 20% accuracy in the prediction under
the hybrid model in comparison to the LSTM model for time series prediction (Yung
et al. 2022). Purohit et al. have also observed that the price of agricultural products
can be better predicted through the hybrid models (Purohit et al. 2021). Figure 7.9
shows that the prediction accuracy is up to a 95% confidence interval.
116 B. Pandey et al.

Fig. 7.5 After third differentiation

Table 7.3 ARIMA results

Dep. variable: y No. observations: 47
Model: ARIMA (0, 3, 2) Log likelihood: -438.684
Date: Wed, 06 Sep 2023 AIC: 883.367
Time: 14:07:30 BIC: 888.720
Sample: 0 HQIC: 885.352
-47
Covariance Type: opg
Coef Std err z P > |z| [0.025 0.975]
ma.L1 -1.8354 0.111 -16.567 0.000 -2.053 -1.618
ma.L2 0.9970 0.115 8.632 0.000 0.771 1.223
sigma2 2.404e+07 8.79e-09 2.73e+15 0.000 2.4e+07 2.4e+07
Ljung-Box (L1) (Q): 2.81 Jarque–Bera (JB):38.27
Prob (Q): 0.09 Prob(JB): 0.00
Heteroskedasticity (H): 72.47 Skew: 0.79
Prob(H) (two-sided): 0.00 Kurtosis: 7.29

7.4 Conclusion

We observed that the hybrid model comprising the linear and non-linear character-
istics of the data is more appropriate for predicting the value of the agricultural
production. The value of agricultural production is an outcome of the price of the
product and volume of the production. The production level itself is dependent on
the innumerable factors and the price itself is endogenous to the level of production
and the demand. Therefore, the existence of non-linearity is intuitive to the analysis.
7 An Optimized Hybrid ARIMA-LSTM Model for Time Series Forecasting. . . 117

Fig. 7.6 Residuals density

Fig. 7.7 Left graph shows comparative view of prediction of ARIMA and actual; right graph
shows the comparative view of prediction of LSTM and actual

Fig. 7.8 Left graph shows comparative view of prediction of ARIMA, LSTM, and actual; right
graph shows the comparative view of prediction of ARIMA, LSTM, hybrid, and actual observation
118 B. Pandey et al.

95% confidence interval

4000000
Value of production
3000000

2000000

1000000

0 5 10 15 20 25
Year

Fig. 7.9 Prediction accuracy

The study reports that the hybrid model based on the algorithm of the LSTM using
the residuals of the ARIMA models is more suitable for the prediction of the value of
the agricultural production. Considering the limitations of this chapter, the predic-
tions analysis requires more complicate models that may bring together all the
factors affecting the demand and supply of the agricultural product for more accurate
predictions.

References

UNDP (2023) The sustainable development goals report special edition. UNDP
Hauchet-Bourdon M (2011) Agricultural commodity price volatility: an overview, OECD food,
agriculture and fisheries papers, vol 52. OECD Publishing, Paris, France
PIB (2023, 21 March) “release id 1909213”
Hua Y (2020) Bitcoin price prediction using ARIMA and LSTM. In: E3S web of conferences
218, 01050, ISEESE 2020
Box B, Jenkins G (1976) Time series analysis: forecasting and control, Holden Day Series in
timeseries analysis and digital signal Processing
Kalpakis K, Gada D, Puttagunta V (2000) Distance measure for effective clustering of ARIMA time
series. In: IEEE Int Conf Data Min
Nguyen H, Naeem M, Wichitaksorn N, Pears R (2019) A smart system for short-term price
prediction using time series models. Comput Electr Eng 76:339–352
Siami-Namini S, Tavakoli N, Namin AS (2018) A comparison of ARIMA and LSTM in forecasting
time series. In: 17th IEEE international conference on machine learning and applications
Deng Y, Fan H, Wu S (2023) A hybrid ARIMA-LSTM model optimized by BP in the forecast. J
Ambient Intell Humaniz Comput 14:5517–5527
Naveena K, Singh S, Rathod S, Singh A (2017) Hybrid time series modelling for forecasting the
price of washed coffee (Arabica plantation coffee) in India. Int J Agric Sci 9(10):4004–4007
Khazee P, Bagherzadeh Z, Niazkhani PH (2019) Predicting the function of transplanyed kidney in
long term care process: application of a hybrid model. J Biomed Inform 91
7 An Optimized Hybrid ARIMA-LSTM Model for Time Series Forecasting. . . 119

Dave E, Leonardo A, Jeanice M, Hanafiah N (2021) Forecasting Indonesia exports using a hybrid
model ARIMA-LSTM. Proc Comput Sci 179:480–487
Sunil BK, Yadav N (2023) A novel hybrid model combining BSARMA and LSTM for time series
forcasting. Appl Soft Comput 134
Yoo T-W, Oh I-S (2020) Time series forecasting of agricultural products’ sales volumes based on
seasonal long short-term memory. Appl Sci 10(22):8169
Greene WH (2003) Econometric analysis. Pearson
Hochreiter S, Schmidhuber J (1997) Long short-term memory. Neural Comput 9(8):1735–1780
Yung Y, Fan C, Xiong H (2022) A novel general-purpose hybrid model for time series forecasting.
Appl Intell 52:2212–2223
Purohit SK, Panigarhi S, Sethy PK, Behera SK (2021) Time series forecasting of price of
agricultural products using hybrid methods, applied artificial intelligence. Appl Artif Intell
35(15):1388–1406
Chapter 8
An Exploratory Analysis of Machine
Intelligence-enabled Plant Diseases
Assessment

Ashis Pattanaik, Agniva Bhattacharya, and Sushruta Mishra

Abstract Agriculture plays an important role in India due to population growth and
increasing demand for food. Therefore, there is a need to increase crop yields. One of
the significant effects of low crop yields is diseases caused by bacteria, fungi and
viruses. This can be prevented and controlled by applying plant disease detection
approaches. Machine learning techniques are used to identify plant diseases, mainly
because they apply the information itself and provide better techniques for detecting
plant diseases. Machine learning-based methods apply primarily to data dominance
outcomes for specific tasks, and thus can be used to identify diseases. This approach
provided a comprehensive overview of various techniques used for plant disease
detection using artificial intelligence-based machine learning and deep learning
techniques. Similarly, deep learning has become very important in the field of
computer vision as it provides better performance results in plant disease detection.
Advances in deep learning have been applied to many fields and have brought
significant advances in machine learning and computer vision. This comparative
study examines machine learning and deep learning techniques, and their perfor-
mance and use in various research papers aims to show the effectiveness of deep
learning and machine learning models. To prevent significant crop loss, leaf disease
can be detected using captured images using deep learning techniques. Here we use
Kaggle’s PlantVillage dataset for study and analysis.

Keywords Agriculture · India · Population growth · Crop yields · Diseases ·

Bacteria · Fungi · Viruses · Plant disease detection · Machine learning

A. Pattanaik (✉) · A. Bhattacharya · S. Mishra

Kalinga Institute of Industrial Technology, Deemed to be University, Bhubaneswar, Odisha,
India
e-mail: [email protected]; [email protected]; [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 121
A. Khamparia et al. (eds.), Microbial Data Intelligence and Computational
Techniques for Sustainable Computing, Microorganisms for Sustainability 47,
https://doi.org/10.1007/978-981-99-9621-6_8
122 A. Pattanaik et al.

8.1 Introduction

The agricultural process commences with seed planting and concludes with crop
harvest, encompassing challenges such as disease outbreaks, storage management,
pesticide control, weed identification and management, and soil and water resource
inadequacies. These categories have seen the integration of artificial intelligence
(AI) and machine learning. AI advancements build on past learning, automating
tasks through machine learning techniques like backpropagation, artificial neural
networks, and convolutional neural networks, thereby advancing agricultural tech-
nology. The convergence of the Internet of Things (IoT), artificial intelligence (AI),
and unmanned aerial vehicles (UAVs) is seamlessly harnessed to bolster agricultural
domains in detecting and accurately reporting plant leaf diseases. In today’s con-
temporary society, farming and agriculture have waned in appeal, largely due to the
persistent challenges confronted by farmers on a daily basis. Consequently, the
younger demographic is increasingly gravitating toward urban centers, seeking a
more secure life and evading the impediments associated with agricultural pursuits.
The effective safeguarding against plant diseases is intricately entwined with the
imperative need for sustainable adaptations in both climate and farming practices
(Loey et al. 2020). Research findings indicate that the dynamic of climate change has
the potential to influence pathogenic stages and their progression rates. Moreover,
alterations in host resistance mechanisms could instigate shifts in the physiological
dynamics of host–pathogen interactions (Shruthi et al. 2019). In the present era, the
reality that diseases now exhibit greater global mobility than ever before adds
complexity to the scenario. The emergence of novel diseases in regions where
they were previously absent underscores the inherent challenge of combating them
in areas lacking local expertise to effectively address them (Bera et al. 2019). Plant
diseases pose a dual peril: endangering worldwide food security and particularly
devastating smallholder farmers who rely on thriving crops. Developing countries
rely heavily on these farmers, with over 80% of their agricultural output stemming
from their efforts (UNEP 2013). Irrespective of the methodology employed, precise
disease identification upon initial emergence remains pivotal for effective disease
control. Previously, agricultural extensions and local plant clinics aided disease
identification. Now, digital resources supplement this process by offering online
disease diagnosis information, capitalizing on global Internet accessibility. More-
over, mobile-based tools have swiftly gained traction, capitalizing on the unprece-
dented global adoption of mobile technology (ITU 2015). Smartphones, with their
computational prowess, high-definition screens, and advanced built-in features like
HD cameras, present innovative opportunities for disease identification. The
projected global smartphone count is anticipated to range from 5 to 6 billion by
2020. In 2015, mobile broadband coverage was available to 69% of the world’s
population, showcasing a remarkable 12-fold surge in mobile broadband penetration
from 2007 to 2015, reaching 47% (ITU 2015). The convergence of prevalent
smartphone usage, advanced HD cameras, and powerful mobile processors creates
an environment where automated disease diagnosis through image recognition can
8 An Exploratory Analysis of Machine Intelligence-enabled Plant. . . 123

achieve an unprecedented reach, if technically viable. We showcase this feasibility

by employing deep learning with 54,306 images of 14 crop species and 26 diseases
(or health) from the PlantVillage project (Hughes and Salathé 2015). In various
domains, deep neural networks have proven effective for end-to-end learning,
exemplifying their versatility. They establish input-to-output relationships, like
mapping diseased plant images to crop–disease pairs. Nodes in these networks act
as mathematical functions, transforming inputs from incoming edges into numerical
outputs through outgoing edges. Deep networks sequentially connect input and
output layers through stacked nodes. The task is to craft a network where structure,
functions (nodes), and edge weights accurately align input with output. Training
deep neural networks involves refining parameters to enhance mapping through
demanding computation, with recent advancements greatly enhancing efficiency
through conceptual and engineering breakthroughs (LeCun et al. 2015;
Schmidhuber 2015).
The main objectives of the work are as follows:
• AI-Powered Disease Detection: Develop a robust AI-based diagnostic system that
accurately identifies plant diseases using machine learning techniques.
• Automated and Efficient Assessment: Streamline the diagnosis process through
automation, facilitating rapid and effective plant health evaluation.
• Enhanced Crop Productivity: Improve crop yield by enabling early disease
detection and intervention, contributing to healthier and more productive crops.
• Sustainable Agriculture: Promote environmentally friendly practices by reducing
pesticide use through targeted disease identification.
• Holistic Integration: Create an adaptable, user-friendly interface that supports
diverse crops, real-time monitoring, and practical application in agriculture.

8.2 Literature Review

In our evolving environment, the timely and precise recognition of diseases,

encompassing early intervention, holds heightened importance. Diverse methods
detect plant pathologies. While certain diseases lack immediate symptoms or are
identified too late, meticulous scrutiny becomes necessary. Yet, many ailments
exhibit visible signs, making professional evaluation the predominant detection
approach. Skillful observation by plant pathologists is essential for identifying
distinct symptoms and achieving precise plant disease diagnosis (Shirahatti et al.
2018). Variations in symptoms exhibited by diseased plants can lead to erroneous
diagnoses, particularly challenging for non-experts compared to professional pathol-
ogists. Novice gardeners and seasoned specialists alike can gain substantial advan-
tages from an automated system designed to identify plant conditions. The system
utilizes visual cues and appearance, serving as a confirmation for disease diagnosis
accuracy (Ganatra and Patel 2018). Advancements in computer vision present
avenues for enhancing and fortifying the precision of plant protection, while also
124 A. Pattanaik et al.

expanding the scope of precise agricultural computer vision applications. For the
purpose of identifying and categorizing plant diseases, widely adopted techniques in
digital image processing, such as color detection and thresholding, were employed
(Ganatra and Patel 2018). Novel methodologies in deep learning are actively
employed for identifying plant diseases, with convolutional neural networks
(CNN) emerging as a prominent choice. Deep learning represents a contemporary
machine learning paradigm, boasting exceptional outcomes in diverse realms like
computer vision, pharmacy, and bioinformatics. This approach capitalizes on its
ability to process raw data directly, bypassing the need for manual feature engineer-
ing (Ganatra and Patel 2018). Deep learning has garnered substantial success in both
academic and industrial contexts due to two primary factors (Athanikar and Badar
2016). Firstly, the continual generation of vast datasets provides ample material for
constructing intricate models. Secondly, the computational prowess of graphics
processing units (GPUs) empowers the training and utilization of deep models
through efficient parallel processing (Kumar and Raghavendra 2019). The paper
by Ashourloo et al. delves into the detection of leaf rust disease using hyperspectral
measurements. It explores various machine learning regression techniques for this
purpose. Published in the Institute of Electrical and Electronics Engineers (IEEE)
Journal of Selected Topics in Applied Earth Observations and Remote Sensing, the
study aims to enhance disease detection through hyperspectral analysis (Ashourloo
et al. 2016). The paper authored by Han, Haleem, and Taylor introduces an innova-
tive computer vision-driven method for the automated detection and evaluation of
crop diseases. Presented at the Science and Information Conference in July 2015, the
approach utilizes computer vision techniques to identify and assess the severity of
various crop diseases, thereby contributing to agricultural disease management (Han
et al. 2015). The paper by Barbedo presents an innovative algorithm for semi-
automatic segmentation of plant leaf disease symptoms via digital image processing.
Published in Tropical Plant Pathology in 2016, the algorithm offers a novel
approach to accurately segment and analyze disease symptoms on plant leaves
using digital images (Barbedo 2016). The paper by Ramesh and Vardhan explores
data mining methods applied to agricultural yield data in the International Journal of
Advanced Research in Computer Communication and Engineering (2013). The
study investigates techniques and applications of data mining in analyzing agricul-
tural yield data, aiming to uncover insights and patterns within this context (Ramesh
and Vardhan 2013). The research paper by Xia, Li, and Li introduces an intelligent
diagnostic system for identifying wheat diseases using an Android phone. Published
in the Journal of Information & Computer Science in December 2015, the paper
presents a system designed to provide accurate diagnosis of wheat diseases through
mobile technology (Xia et al. 2015). The paper by Kaur and Kang focuses on
improving plant disease detection by enhancing the classifier support vector machine
(SVM). It was presented at the IEEE third International Conference on MOOCs,
Innovation, and Technology in Education (MITE) in 2015. The study aims to
enhance the performance of SVM for more accurate plant disease detection (Kaur
and Kang 2015). The research paper by E.C. Too and colleagues explores plant
disease identification using deep learning models. It involves a comparative analysis
8 An Exploratory Analysis of Machine Intelligence-enabled Plant. . . 125

of fine-tuning techniques for these models. Published in Computers and Electronics

in Agriculture in 2019, the study delves into improving the accuracy of plant disease
recognition through the refinement of deep learning approaches (Too et al. 2019). In
the realm of plant species classification, Dyrmann et al. 2016 employed deep
learning architectures. Their study introduced a technique capable of distinguishing
between weeds and plant species using colored images. They harnessed
convolutional neural networks (CNNs), evaluating their model on 10,413 images
encompassing 22 weed and crop species. The CNN model demonstrated an accuracy
of 86.2%, albeit encountering challenges in classifying specific plant species due to
limited training samples for those categories. Plant disease classification has also
seen the utilization of machine learning methods. In the work by Athanikar and
Badar (2016), a neural network was employed to differentiate between healthy and
diseased potato leaf images. The outcomes demonstrated the successful capability of
the backpropagation neural network (BPNN) in identifying disease spots and accu-
rately categorizing the specific disease type, achieving an impressive accuracy rate
of 92%. The research paper by Sankur and Sezgin (2004) presents a comprehensive
survey of image thresholding techniques and evaluates their quantitative perfor-
mance. The study explores various methods for thresholding images and analyzes
their effectiveness. The authors discuss a range of thresholding techniques and their
applications, considering their advantages and limitations. The paper also delves into
the quantitative evaluation of these methods, providing insights into their perfor-
mance metrics. Overall, the paper serves as a valuable resource for understanding
and assessing image thresholding techniques, contributing to the field of electronic
imaging. Furthermore, in a study conducted by Wang et al. (2012), an empirical
investigation was undertaken with the aim of developing an approach for achieving
image recognition of plant diseases. This research involved the utilization of four
distinct neural networks to differentiate between wheat stripe rust and wheat leaf
rust, as well as between grape downy mildew and grape powdery mildew. These
neural networks relied on extracted color, shape, and texture features from the
disease images. The outcomes of the study demonstrated the successful utilization
of image processing-based neural networks for the efficient identification and diag-
nosis of plant diseases. Furthermore, in their work from 2012, Samanta and col-
leagues introduce an image processing technique for identifying scab disease in
potatoes. They gather images from diverse potato fields and subject them to
enhancement processes. Through image segmentation, they isolate the specific
areas of interest, which correspond to the regions affected by the disease. Ultimately,
they employ a histogram-based approach to analyze these targeted regions and
determine the disease stage (Samanta et al. 2018). The research paper by Han
et al. (2015) presents a new computer vision-based method for automatically
detecting and evaluating the severity of crop diseases. The approach was showcased
at the Science and Information Conference. The authors introduce a technique that
utilizes computer vision to identify and quantify the extent of disease in crops. This
method likely involves innovative strategies for leveraging visual information to
assess the severity of diseases impacting crops. The study contributes to the field of
agriculture and technology by proposing a potentially advanced solution for
126 A. Pattanaik et al.

addressing crop disease issues through automated visual analysis (Han et al. 2015).
The research paper by Dey et al. (2016) focuses on the detection of leaf rot disease in
betel vine (Piper betle) using image processing techniques. Presented at the Inter-
national Conference on Computational Modeling and Security, the paper introduces
a methodology to identify leaf rot disease in betel vine leaves. The authors utilize
image processing methods to analyze visual characteristics associated with the
disease. The study’s findings demonstrate the effectiveness of their approach in
detecting leaf rot disease accurately, showcasing its potential for disease diagnosis in
agricultural contexts (Dey et al. 2016). The research paper titled “eAGROBOT-A
robot for early crop disease detection using image processing” by Pilli et al. (2014)
introduces the concept of an automated agricultural robot called eAGROBOT. The
robot is designed for the early detection of crop diseases through the utilization of
image processing techniques. The paper presents this innovative approach,
highlighting the integration of electronics, communication systems, and image
processing to develop a system that can identify and diagnose crop diseases at an
early stage. The research emphasizes the potential of such technology in improving
crop health monitoring and contributing to more efficient and sustainable agriculture
practices. The paper was presented at the International Conference on Electronics
and Communication Systems (ICECS) in 2014 (Pilli et al. 2014). The research paper
titled “Using Deep Learning for Image-Based Plant Disease Detection” by S. P.
Mohanty, D. P. Hughes, and M. Salathé was published in 2016 in the journal
Frontiers in Plant Science. The paper focuses on leveraging deep learning tech-
niques for the detection of plant diseases using images. The authors explore the
potential of deep learning models, particularly convolutional neural networks
(CNNs), to accurately identify plant diseases through image analysis. The study
contributes to the field of agricultural technology by demonstrating the effectiveness
of using advanced machine learning techniques for automating the diagnosis of plant
diseases based on visual cues. The paper’s findings underscore the significance of
deep learning in improving disease detection accuracy, thereby aiding in timely and
efficient disease management in agriculture (Mohanty et al. 2016). The paper by
Soni (2018) presents an enhanced faster regional convolutional neural network for
monitoring data centers. Published in the International Journal of Advanced
Research, it focuses on improving efficiency and accuracy in data center monitoring
using neural network techniques (Soni 2018). The paper by Ramcharan et al. (2017)
presents deep learning techniques for detecting cassava diseases from images. It
explores image-based methods for cassava disease identification, contributing to
improved disease management in agriculture. Published in Frontiers in Plant Sci-
ence, the research focuses on leveraging deep learning for accurate disease detection
in cassava plants (Ramcharan et al. 2017). The paper by Fujita et al. (2016) presents
a foundational study on a reliable and feasible plant diagnostic system. It was
discussed at the 15th IEEE International Conference on Machine Learning and
Applications. The research delves into creating a robust plant diagnosis approach,
offering insights and practical implications (Fujita et al. 2016). The research paper
by Huang (2007) focuses on the application of artificial neural networks to detect
diseases in phalaenopsis seedlings. The study utilizes color and texture features to
8 An Exploratory Analysis of Machine Intelligence-enabled Plant. . . 127

identify and diagnose plant diseases. By employing an artificial neural network, the
research aims to enhance the accuracy and efficiency of disease detection in agri-
cultural settings. The paper explores the potential of using computational methods to
improve disease diagnosis in phalaenopsis seedlings, offering insights into the
integration of technology and agriculture for improved plant health management
(Huang 2007). The research paper by Harvey et al. (2014) explores the significant
vulnerability of smallholder farmers in Madagascar to agricultural risks and climate
change. The study emphasizes the extreme challenges these farmers face due to their
dependence on agriculture for their livelihood. It investigates how changing climatic
conditions exacerbate existing risks, impacting crop yields and food security. The
research underscores the urgent need for targeted interventions and adaptive strate-
gies to address the vulnerability of these farmers, considering the complex interplay
of environmental, economic, and social factors. The findings highlight the impor-
tance of sustainable agricultural practices and support systems to enhance the
resilience of Madagascar’s smallholder farming communities in the face of ongoing
climate challenges (Harvey et al. 2014). The research paper titled “ImageNet Large
Scale Visual Recognition Challenge” by Russakovsky et al. (2015) presents a
comprehensive overview of the ImageNet Challenge, a significant competition in
computer vision. The paper discusses the challenge’s objectives, dataset, evaluation
metrics, and results. It highlights the rapid advancement in deep learning techniques
and their impact on image classification and object recognition tasks. The authors
emphasize the importance of large-scale datasets and high-capacity neural networks
in achieving remarkable performance improvements. The paper’s findings under-
score the transformative influence of the ImageNet Challenge in driving innovation
and progress within the field of computer vision (Russakovsky et al. 2015). The
research paper by Sankaran et al. (2011) explores the application of visible-near
infrared spectroscopy to detect huanglongbing (HLB) disease in citrus orchards. The
study focuses on using spectral analysis to identify disease symptoms. The authors
utilize computational and electronic agriculture techniques, demonstrating the poten-
tial of this approach for disease detection in citrus trees. The paper discusses the
methodology, findings, and implications of using spectroscopy as a tool for disease
identification in the context of citrus cultivation (Sankaran et al. 2011). The research
paper titled “Going Deeper with Convolutions” by Szegedy et al. (2015) focuses on
enhancing convolutional neural networks (CNNs) for computer vision tasks. The
paper introduces a novel architecture called GoogLeNet, which employs deep and
parallel convolutional layers to achieve better performance with fewer parameters.
The key innovation is the utilization of “Inception” modules, which combine
different kernel sizes within the same layer to capture diverse features at various
scales. This approach enables more efficient and deeper networks while mitigating
overfitting. The research paper’s findings demonstrate that the proposed GoogLeNet
architecture outperforms previous CNNs on benchmarks like the ImageNet dataset,
showcasing the advantages of increased depth and the Inception modules for
improving image recognition tasks (Szegedy et al. 2015). The research paper titled
“A Comparative Study on Application of Computer Vision and Fluorescence Imag-
ing Spectroscopy for Detection of Huanglongbing Citrus Disease in the USA and
128 A. Pattanaik et al.

Brazil” by Wetterich et al. (2012) explores the utilization of computer vision and
fluorescence imaging spectroscopy for detecting the citrus disease known as
huanglongbing (HLB) in both the USA and Brazil. The study compares the effec-
tiveness of these two techniques in identifying HLB in citrus plants. The research
investigates the potential of these technologies for disease detection and provides
insights into their application in two different geographic regions. The paper was
published in the Journal of Spectroscopy in 2013 (841738) and contributes to the
field of spectroscopy and computer vision for disease diagnosis in agriculture, with
implications for disease management and prevention in citrus cultivation. The
study’s findings offer valuable information for researchers, practitioners, and
policymakers aiming to combat HLB and enhance citrus crop health in diverse
contexts (Wetterich et al. 2012). The research paper by Zeiler and Fergus (2014)
titled “Visualizing and Understanding Convolutional Networks” explores the visu-
alization and comprehension of convolutional networks in the field of computer
vision. Published in the European Conference on Computer Vision 2014 proceed-
ings, the paper delves into methods for interpreting the behavior of deep
convolutional neural networks (CNNs). The authors discuss techniques to visualize
the learned features and activations within the CNN layers, shedding light on how
these networks process and recognize visual patterns. The paper contributes to
enhancing the interpretability and transparency of CNNs, offering insights into
their inner workings and aiding researchers in understanding the mechanisms behind
their impressive performance in image recognition tasks (Zeiler and Fergus 2014). In
the research paper authored by Singh et al. (2015), the authors explore the applica-
tion of machine learning techniques for efficient stress phenotyping in plants with
high throughput. They address the challenge of accurately and rapidly assessing
plant stress responses using advanced computational methods. The study empha-
sizes the potential of machine learning to analyze and interpret large-scale datasets,
aiding in identifying stress patterns and facilitating agricultural advancements. The
paper highlights the role of machine learning in enhancing plant stress research and
its implications for future agricultural practices. Published in Trends in Plant
Science, the work underscores the significance of technological innovation in plant
biology (Singh et al. 2015). In the research paper titled “Metaheuristic optimization-
based resource allocation technique for cybertwin-driven 6G on IoE environment,”
authored by D. Kumar Jain, S. Kumar Sah Tyagi, S. Neelakandan, M. Prakash, and
L. Natrayan, published in the IEEE’s Transactions on Industrial Informatics (2021),
a novel approach to resource allocation for cybertwin-driven 6G networks in the
Internet of Everything (IoE) environment is presented. The authors propose a
metaheuristic optimization method to efficiently allocate resources, enhancing the
performance of cybertwin applications in the dynamic IoE setting. The study’s
findings contribute to the advancement of resource management strategies in the
context of emerging 6G technologies (Kumar Jain et al. 2021). In the research paper
authored by B. Lurstwut and C. Pornpanomchai titled “Image analysis based on
color, shape, and texture for rice seed (Oryza sativa L.) germination evaluation,”
published in the journal Agriculture and Natural Resources in 2017, the authors
delve into an innovative approach to assessing the germination of rice seeds (Oryza
8 An Exploratory Analysis of Machine Intelligence-enabled Plant. . . 129

sativa L.) through comprehensive image analysis techniques. The study addresses
the critical need for accurate and efficient methods to evaluate seed germination, a
pivotal process in agricultural productivity. The authors focus on the integration of
color, shape, and texture analysis to provide a holistic assessment of seed germina-
tion. This novel approach aims to enhance the accuracy of germination evaluation
and enable better decision-making in agricultural practices. Throughout the study,
the authors meticulously detail the methodology employed. Image analysis algo-
rithms are applied to capture and analyze various visual aspects of the germinating
rice seeds. Color variations, morphological changes in shape, and textural alterations
are quantitatively assessed. The authors gather extensive data and subsequently
utilize statistical techniques to draw meaningful conclusions from the acquired
information. The results of this research exhibit promising outcomes. The combined
analysis of color, shape, and texture proves to be a potent tool in evaluating rice seed
germination. The comprehensive approach not only enhances accuracy but also
offers a more comprehensive understanding of the germination process. The paper
discusses the implications of these findings for agricultural practices, underscoring
the potential to optimize seed selection and crop management strategies. In conclu-
sion, Lurstwut and Pornpanomchai’s research paper showcases an innovative image
analysis approach for assessing rice seed germination. By integrating color, shape,
and texture analysis, the authors provide a comprehensive and accurate means of
evaluating this crucial agricultural process. The findings contribute to advancing
agricultural practices and hold the promise of improving crop yield and quality
(Lurstwut and Pornpanomchai 2017). In the research paper titled “Smart Paddy Crop
Disease Identification and Management using Deep Convolution Neural Network
and SVM Classifier,” authored by R. Rajmohan, M. Pajany, R. Rajesh, D. R. Raman,
and U. Prabu and published in the International Journal of Pure and Applied
Mathematics in 2018, the authors address the crucial issue of paddy crop disease
identification and management through the integration of advanced machine learn-
ing techniques. The study focuses on utilizing a combination of deep convolutional
neural networks (CNN) and support vector machine (SVM) classifier to effectively
detect and manage diseases in paddy crops. Paddy crops are susceptible to various
diseases that can significantly impact crop yield and quality. Early detection and
management of these diseases are vital to ensure agricultural productivity and food
security. The authors propose a novel approach that leverages the strengths of both
deep learning and traditional machine learning. Deep CNNs are employed to process
images of paddy leaves and extract relevant features that can aid in disease identi-
fication. CNNs are known for their capability to automatically learn hierarchical
features from images, making them suitable for image recognition tasks. Following
the feature extraction stage, the study integrates an SVM classifier into the frame-
work. SVMs are known for their ability to classify data points into different classes
based on their features. In this context, SVMs are applied to classify the extracted
features and subsequently identify the specific disease affecting the paddy crops. The
proposed methodology offers several advantages. By combining CNNs and SVMs,
the approach harnesses the power of deep learning’s feature extraction with SVM’s
classification capabilities. This enhances the accuracy and robustness of disease
130 A. Pattanaik et al.

identification. The integration of advanced machine learning techniques offers a

smart solution for timely disease management, helping farmers make informed
decisions regarding pest control and crop protection measures. The research con-
tributes to the agricultural domain by presenting an innovative strategy for paddy
crop disease identification and management. The experimental results, outlined in
the paper, demonstrate the effectiveness of the proposed approach in accurately
identifying crop diseases based on leaf images. This research paper underscores the
potential of machine learning technologies in revolutionizing agriculture by offering
precise and timely disease detection mechanisms, thereby contributing to sustainable
food production and economic stability in farming communities (Rajmohan et al.
2018). In the research paper titled “Rice Plant Disease Classification Using Transfer
Learning of Deep Convolutional Neural Network,” authored by V. K. Shrivastava,
M. K. Pradhan, S. Minz, and M. P. Thakur, and published in the International
Archives of the Photogrammetry, Remote Sensing, and Spatial Information Sci-
ences, Volume 42 in 2019, the authors investigate the efficacy of transfer learning
in the domain of rice plant disease classification. The study addresses the crucial
issue of rice plant disease detection, which has significant implications for crop yield
and food security. Traditional methods of disease identification often involve manual
inspection and visual assessment, leading to delays and inaccuracies. To overcome
these limitations, the researchers propose the application of deep convolutional
neural networks (CNNs) with transfer learning. Transfer learning, a technique
wherein a pre-trained neural network is fine-tuned for a specific task, is leveraged
by the authors to enhance the accuracy of disease classification. The researchers
choose a pre-trained CNN model and tailor it to recognize different types of rice
plant diseases by retraining the model with a specific dataset containing images of
healthy and diseased rice plants. The experimental methodology involves collecting
a comprehensive dataset of rice plant images, including healthy plants and plants
affected by various diseases. This dataset is utilized for training, validation, and
testing of the CNN model. The authors employ various performance metrics to
assess the model’s accuracy in identifying different diseases accurately. The results
of the study demonstrate the effectiveness of the proposed approach. The trained
CNN model exhibits notable accuracy in classifying various rice plant diseases,
outperforming traditional methods of disease detection. The application of transfer
learning aids in overcoming the challenges of limited data and computationally
intensive training, allowing for better generalization of disease patterns. The impli-
cations of this research are significant for the agricultural sector, as it offers a more
efficient and accurate means of diagnosing rice plant diseases. Timely and accurate
disease detection can help farmers take targeted measures to prevent the spread of
diseases and optimize crop management strategies. In conclusion, V. K. Shrivastava,
M. K. Pradhan, S. Minz, and M. P. Thakur’s research paper explores the application
of transfer learning using deep convolutional neural networks for rice plant disease
classification. The study underscores the potential of this approach to revolutionize
disease detection in the agricultural domain, contributing to improved crop health
management and food security (Shrivastava et al. 2019). In the research paper
authored by R. Islam and M. Rafiqul, titled “An image processing technique to
8 An Exploratory Analysis of Machine Intelligence-enabled Plant. . . 131

calculate the percentage of disease-affected pixels of paddy leaf,” published in the

International Journal of Computer Applications in 2015, a novel approach for
quantifying the extent of disease impact on paddy leaves through image processing
is presented. The study addresses the critical concern of disease detection in agri-
cultural crops, specifically focusing on paddy leaves. Timely and accurate identifi-
cation of diseases is essential to ensure effective crop management and yield
optimization. The authors propose an innovative image processing technique that
offers a quantitative assessment of the percentage of disease-affected pixels on
paddy leaves. The methodology centers around leveraging digital images of paddy
leaves, captured using appropriate imaging equipment. These images are then
subjected to a series of image processing steps to distinguish between healthy and
disease-affected areas. The process involves segmenting the leaf images into distinct
regions and isolating the portions exhibiting signs of disease. Through a combina-
tion of pixel intensity analysis and pattern recognition algorithms, the authors
accurately quantify the ratio of disease-affected pixels to the total number of pixels
in the leaf image. The experimental validation of the proposed technique is
conducted to assess its effectiveness. A comprehensive dataset of paddy leaf images
with varying degrees of disease manifestation is employed. By comparing the results
obtained from the proposed method with manually annotated ground truth data, the
authors demonstrate the technique’s proficiency in accurately estimating the per-
centage of disease-affected pixels. The outcomes of this research hold significant
implications for agricultural practices and crop management strategies. The auto-
mated and quantitative nature of the proposed image processing technique offers a
streamlined approach to disease assessment, eliminating the subjective biases asso-
ciated with manual inspection. This, in turn, enables timely interventions, facilitating
improved disease control and ultimately enhancing crop yield. In conclusion, the
paper authored by R. Islam and M. Rafiqul outlines an innovative image processing
technique for evaluating the proportion of disease-affected pixels on paddy leaves.
By harnessing the power of digital image analysis, this method presents a valuable
contribution to the field of agricultural research and offers a promising avenue for the
advancement of precision farming practices (Islam and Rafiqul 2015). In the
research paper authored by P. K. Sethy, B. Negi, N. K. Barpanda, S. K. Behera,
and A. K. Rath and published in 2018 in the journal Cognitive Science and Artificial
Intelligence by Springer, the focus is on evaluating the severity of disease in rice
crops utilizing machine learning and computational intelligence techniques. The
study falls under the domain of applied sciences and technology, specifically
categorized as “Springer Briefs.” The primary objective of the research is to develop
an effective method for measuring the severity of diseases that affect rice crops.
Agricultural diseases have a significant impact on crop yield and food security, and
accurate assessment of disease severity is crucial for implementing appropriate
management strategies. The researchers employ advanced machine learning and
computational intelligence methodologies to address this challenge. By leveraging
the power of these techniques, they aim to create a model capable of accurately
assessing the severity of diseases in rice plants. The integration of these modern
technologies holds promise in providing a reliable and efficient solution to this
132 A. Pattanaik et al.

pertinent agricultural concern. The paper outlines the methodology used in the study,
which involves collecting data related to various parameters associated with the
health of rice plants. These parameters are then processed using machine learning
algorithms and computational intelligence techniques. The resulting model is
designed to recognize patterns and correlations between the collected data and
disease severity levels. The outcomes of the research showcase the successful
implementation of the proposed approach. The developed model demonstrates a
high degree of accuracy in assessing disease severity in rice crops. This achievement
implies a significant step forward in early disease detection and management strat-
egies for agricultural systems. In conclusion, P. K. Sethy, B. Negi, N. K. Barpanda,
S. K. Behera, and A. K. Rath’s2018 research paper contributes to the field of applied
sciences and technology. Their study presents a comprehensive approach that
combines machine learning and computational intelligence to accurately measure
the severity of diseases in rice crops. The success of their methodology underscores
the potential of these technologies in revolutionizing disease management within
agriculture, thereby contributing to increased crop yield and food security (Sethy
et al. 2018).

8.3 Working Methodology and Datasets

This study focuses on the domain of agriculture and plant health. It aims to develop
an AI-based system for diagnosing plant diseases using technologies like
TensorFlow and OpenCV. The research employs a diverse and well-annotated
dataset of plant images. This dataset comprises images of healthy plants as well as
plants afflicted by various diseases. The dataset covers a range of plant species and
disease types to ensure robustness and accuracy in the AI model’s diagnosis.
Kaggle’s datasets of PlantVillage were used for this study. The steps of the workflow
model are highlighted in Fig. 8.1 and are discussed here.
• Image Preprocessing: Raw dataset images are processed using OpenCV, includ-
ing resizing and normalization, to prepare input data for the AI model.
• CNN Architecture: A convolutional neural network (CNN) is developed using
TensorFlow. CNNs excel in image recognition, capturing spatial features.
• Transfer Learning: Pre-trained CNN models (VGG, ResNet, Inception) are fine-
tuned on the plant disease dataset, leveraging existing knowledge and saving
training time.
• Data Augmentation: Training benefits from techniques like rotation, flipping, and
zooming, enhancing model robustness by artificially expanding the dataset.
• Model Training and Validation: The CNN is trained on training data, validated on
a separate set, and hyperparameters are tuned for optimal performance.
The research demonstrates AI’s potential in transforming plant disease diagnosis
for agricultural professionals through speed, accuracy, and accessibility
8 An Exploratory Analysis of Machine Intelligence-enabled Plant. . . 133

Fig. 8.1 Architectural model for plant disease detection using Tensorflow and OpenCV

•
•

•
•

•
•

•
•

•
•

Fig. 8.2 Process flow diagram of the framework

improvements, facilitated by TensorFlow and OpenCV integration. The process flow

diagram of the model is shown in Fig. 8.2.
The overall summary of the parameters taken into consideration is shown in
Fig. 8.3. Different layers including max_pooling and dropout layers along with their
output metrics are displayed. Configuration for activation function and batch nor-
malization is also highlighted.
Table 8.1 illustrates the overall evaluation outcome of implementation. Different
epoch combinations are considered and their corresponding values for accuracy,
134 A. Pattanaik et al.

Fig. 8.3 Parameter metrics of the model

validation loss, and other metrics are validated. With 35 epochs, the model records
the optimum accuracy rate as shown in Table 8.1.

8.4 Advantages and Constraints

Some Vital Benefits Offered by the Model in Plant Disease Diagnosis:

• Effective Feature Extraction: CNNs are adept at automatically extracting relevant
features from plant images, capturing intricate patterns and textures indicative of
diseases.
8 An Exploratory Analysis of Machine Intelligence-enabled Plant. . . 135

Table 8.1 Evaluation metrics with their values used in the model
Epoch number step—loss accuracy val_loss val_accuracy
1/35 0.0507 0.8543 0.2492 0.5650
2/35 0.0428 0.8785 0.2190 0.6306
3/35 0.0375 0.8896 0.1460 0.6936
4/35 0.0358 0.8944 0.2195 0.6580
5/35 0.0348 0.9020 0.3374 0.5609
6/35 0.0392 0.8882 0.2017 0.6129
7/35 0.0348 0.9024 0.1319 0.7415
8/35 0.0357 0.8986 0.3076 0.5404
9/35 0.0442 0.8744 0.1168 0.7524
10/35 0.0401 0.8872 0.6525 0.3694
11/35 0.0356 0.9024 0.5158 0.4104
12/35 0.0350 0.8955 0.1278 0.7579
34/35 0.0255 0.9290 0.4432 0.4651
35/35 0.0238 0.9311 0.0950 0.8098

• High Accuracy: CNNs’ deep architecture enables them to learn complex relation-
ships, resulting in accurate disease identification and reduced misdiagnoses.
• Generalization: Trained CNNs can generalize well to new, unseen images,
making them adaptable to various plant species and disease types.
• Efficiency: Once trained, CNN models can process images quickly, allowing for
rapid diagnosis and timely intervention.
• Automation: CNN-based systems automate diagnosis, reducing human effort and
enabling non-experts to identify diseases.

A Few Constraints Visible in the Model for Plant Disease Diagnosis:

• Large Datasets Needed: CNNs require substantial labeled data for effective
training, which might be challenging to acquire and annotate.
• Overfitting: Without proper regularization techniques, CNNs can overfit to the
training data, resulting in poor performance on new images.
• Computationally Intensive: Training deep CNNs demands significant computa-
tional resources, making implementation and maintenance costly.
• Domain-Specific Knowledge: CNNs may lack the ability to provide insights into
the underlying biology of diseases, requiring additional expert interpretation.
• Limited Interpretability: The internal workings of CNNs can be challenging to
interpret, hindering the understanding of how decisions are made.
• Model Complexity: Complex CNN architectures might require experts for proper
design, training, and fine-tuning.
• Overall, while CNNs offer remarkable potential for accurate and efficient plant
disease diagnosis, careful consideration of their limitations and proper training
strategies is essential for successful implementation.
136 A. Pattanaik et al.

8.5 Challenging Issues of the Framework

• Dataset Challenges: Obtaining a diverse and well-annotated dataset

encompassing various plant species and diseases is complex. Ensuring precise
annotations and representative samples affects model generalization.
• Generalization Complexity: Achieving model adaptability to varying conditions
like lighting and setups is critical. Navigating new disease variations requires
meticulous training and validation strategies.
• Imbalanced Classes: Uneven class distribution can skew model performance.
Detecting rare diseases is hindered by limited data in certain categories.
• Deployment and User Interface: Designing an intuitive interface for farmers
while maintaining real-time performance poses deployment challenges.
• Ethical Considerations: AI’s entrance into agriculture raises privacy, ownership,
and expertise concerns. Responsible AI usage and addressing societal implica-
tions are vital.

8.6 Conclusion

In conclusion, the integration of convolutional neural networks (CNNs) and artificial

intelligence (AI) has ushered in a transformative era in plant disease diagnosis. This
approach demonstrates remarkable accuracy, automating the detection of diseases by
effectively learning intricate visual patterns from plant images. While challenges like
data availability and model complexity exist, the potential benefits in terms of timely
intervention, resource optimization, and enhanced agricultural practices are undeni-
able. As AI and CNN technologies continue to advance, their application in plant
disease diagnosis holds the promise of revolutionizing the field, contributing to
global food security and sustainable agriculture.

References

Ashourloo D, Aghighi H, Matkan AA, Mobasheri MR, Rad AM (2016) An investigation into
machine learning regression techniques for the leaf rust disease detection using hyperspectral
measurement. IEEE J Sel Top Appl Earth Obs Remote Sens 9(9):4344–4351
Athanikar G, Badar MP (2016) Potato leaf diseases detection and classification system. Int J
Comput Sci Mob Comput 5(2):76–88
Barbedo JGA (2016) A novel algorithm for semi-automatic segmentation of plant leaf disease
symptoms using digital image processing. Trop Plant Pathol 41(4):210–224
Bera T, Das A, Sil J, Das AK (2019) A survey on rice plant disease identification using image
processing and data mining techniques emerging technologies in data mining and information
security. Springer, pp 365–376
8 An Exploratory Analysis of Machine Intelligence-enabled Plant. . . 137

Dey AK, Sharma M, Meshram MR (2016) Image processing based leaf Rot disease, detection of
betel vine (piper BETLEL.), international conference on computational modeling and security.
Proc Comput Sci 85:748–754
Dyrmann M et al (2016) Plant species classification using deep convolutional neural network.
Biosyst Eng 151:72–80
Fujita E, Kawasaki Y, Uga H, Kagiwada S, Iyatomi H (2016) Basic investigation on a robust and
practical plant diagnostic system. In Proceedings of 2016 15th IEEE international conference on
machine learning and applications (ICMLA), pp 989–992
Ganatra N, Patel A (2018) A survey on diseases detection and classification of agriculture products
using image processing and machine learning. Int J Comput Appl 180:1–13
Han L, Haleem MS, Taylor M (2015) A novel computer vision-based approach to automatic
detection and severity assessment of crop diseases. In Science and Information Conference
(SAI), pp 638–644
Harvey CA, Rakotobe ZL, Rao NS, Dave R, Razafimahatratra H, Rabarijohn RH et al (2014)
Extreme vulnerability of smallholder farmers to agricultural risks and climate change in
madagascar. Philos Trans R Soc Lond B Biol Sci 369:20130089. https://doi.org/10.1098/rstb.
2013.008
Huang KY (2007) Application of artificial neural network for detecting phalaenopsis seedling
diseases using color and texture features. Comput Electron Agric 57:3–11. https://doi.org/10.
1016/j.compag.2007.01.015
Hughes DP, Salathé M (2015) An open access repository of images on plant health to enable the
development of mobile disease diagnostics. arXiv:1511.08060
Islam R, Rafiqul M (2015) An image processing technique to calculate percentage of disease
affected pixels of paddy leaf. Int J Comput Appl 123:28–34
ITU (2015) ICT facts and figures—the world in 2015. International Telecommunication Union,
Geneva
Kaur R, Kang SS (2015) An enhancement in classifier support vector machine to improve plant
disease detection. In MOOCs, Innovation and Technology in Education (MITE), IEEE 3rd
international conference on IEEE, pp. 135–140
Kumar Jain D, Tyagi SKS, Neelakandan S, Prakash M, Natrayan L (2021) Metaheuristic
optimization-based resource allocation technique for cybertwin-driven 6G on IoE environment.
IEEE Trans Indust Inform 18(7):4884–4892
Kumar SS, Raghavendra B (2019) Diseases detection of various plant leaf using image processing
techniques: a review 2019 5th International Conference on Advanced Computing & Commu-
nication Systems (ICACCS), pp. 313–316
LeCun Y, Bengio Y, Hinton G (2015) Deep learning. Nature 521:436–444. https://doi.org/10.1038/
nature14539
Loey M, ElSawy A, Afify M (2020) Deep learning in plant diseases detection for agricultural crops:
a survey. Int J Serv Sci Manag Eng Technol 11:41–58
Lurstwut B, Pornpanomchai C (2017) Image analysis based on color, shape and texture for rice seed
(Oryza sativa L.) germination evaluation. Agric Nat Resour 51:383–389
Mohanty SP, Hughes DP, Salathé M (2016) Using deep learning for image-based plant disease
detection. Front Plant Sci 7:1419
Pilli SK, Nallathambi B, George SJ, Diwanji V (2014) eAGROBOT—a robot for early crop disease
detection using image processing. In Electronics and Communication Systems (ICECS), Inter-
national Conference on IEEE, pp 1–6
Rajmohan R, Pajany M, Rajesh R, Raman DR, Prabu U (2018) Smart paddy crop disease
identification and management using deep convolution neural network and SVM classifier. Int
J Pure Appl Mathem 118:255–264
Ramcharan A, Baranowski K, McCloskey P, Ahmed B, Legg J, Hughes DP (2017) Deep learning
for image-based cassava disease detection. Front Plant Sci 8:1852
Ramesh D, Vardhan BV (2013) Data mining techniques and applications to agricultural yield data.
Int J Adv Res Comput Commun Eng 2(9):3477–3480
138 A. Pattanaik et al.

Russakovsky O, Deng J, Su H, Krause J, Satheesh S, Ma S et al (2015) ImageNet large scale visual

recognition challenge. Int J Comput Vis 115:211–252. https://doi.org/10.1007/s11263-015-
0816-y
Samanta S, Pal DK, Palsamanta B (2018) Flood susceptibility analysis through remote sensing, GIS
and frequency ratio model. Appl Water Sci 8:Article number: 66
Sankaran S, Mishra A, Maja JM, Ehsani R (2011) Visible-near infrared spectroscopy for detection
of huanglongbing in citrus orchards. Comput Electron Agric 77:127–134. https://doi.org/10.
1016/j.compag.2011.03.004
Sankur B, Sezgin M (2004) Survey over image thresholding techniques and quantitative perfor-
mance evaluation. J. Electron Imag 13:146–165
Schmidhuber J (2015) Deep learning in neural networks: an overview. Neural Netw 61:85–117.
https://doi.org/10.1016/j.neunet.2014.09.003
Sethy PK, Negi B, Barpanda NK, Behera SK, Rath AK (2018) Measurement of disease severity of
rice crop using machine learning and computational intelligence. In: Gurumoorthy S, Rao BNK,
Gao X-Z (eds) Cognitive science and artificial intelligence, Springer briefs in applied sciences
and technology. Springer, Berlin, Germany, pp 1–11
Shirahatti J, Patil R, Akulwar P (2018) A survey paper on plant disease identification using machine
learning approach 2018 3rd International Conference on Communication and Electronics
Systems (ICCES), pp 1171–1174
Shrivastava VK, Pradhan MK, Minz S, Thakur MP (2019) Rice plant disease classification using
transfer learning of deep convolution neural network. Int Arch Photogrammetry Remote
Sensing Spatial Inform Sci 42
Shruthi U, Nagaveni V, Raghavendra B (2019) A review on machine learning classification
techniques for plant disease detection 2019 5th International Conference on Advanced Com-
puting & Communication Systems, (ICACCS), pp 281–284
Singh A, Ganapathysubramanian B, Singh AK, Sarkar S (2015) Machine learning for high-
throughput stress phenotyping in plants. Trends Plant Sci 21:110–124. https://doi.org/10.
1016/j.tplants.2015.10.015
Soni AN (2018) Data center monitoring using an improved faster regional convolutional neural
network. Int J Adv Res Electric Electron Instrum Eng 7(4):1849–1853. https://doi.org/10.
15662/IJAREEIE.2018.0704058
Szegedy C, Liu W, Jia Y, Sermanet P, Reed S, Anguelov D, et al. (2015) Going deeper with
convolutions. In Proceedings of the IEEE conference on computer vision and pattern
recognition.
Too EC et al (2019) A comparative study of fine-tuning deep learning models for plant disease
identification. Comput Electron Agric 161:272–279
UNEP (2013) Smallholders, food security, and the environment. International Fund for Agricultural
Development (IFAD), Rome
Wang H, Li G, Ma Z, Li X (2012) Image recognition of plant diseases based on backpropagation
networks. Conference: image and signal processing (CISP), 2012 5th international congress doi:
https://doi.org/10.1109/CISP.2012.6469998
Wetterich CB, Kumar R, Sankaran S, Junior JB, Ehsani R, Marcassa LG (2012) A comparative
study on application of computer vision and fluorescence imaging spectroscopy for detection of
huanglongbing citrus disease in the usa and brazil. J Spectrosc 2013:841738. https://doi.org/10.
1155/2013/841738
Xia, Y.Q., Li, Y., Li, C.: Intelligent diagnose system of wheat diseases based on android phone. J
Inform Comput Sci 12, 6845–6852, Dec (2015)
Zeiler MD, Fergus R (2014) Visualizing and understanding convolutional networks. In: Fleet D,
Pajdla T, Schiele B, Tuytelaars T (eds) Computer Vision–ECCV 2014. Springer, pp 818–833
Chapter 9
Synergizing Smart Farming and Human
Bioinformatics Through IoT and Sensor
Devices

Sandeep Kumar Jain and Pritesh Kumar Jain

Abstract The fusion of human bioinformatics with smart farming has developed as
a cutting-edge interdisciplinary field with enormous promise to transform both
healthcare and agriculture. This chapter delves into the intersection of these fields,
examining how sensor technology and the Internet of Things (IoT) enable data-
driven decisions in both personalized healthcare and agricultural operations. While
exploring the significance of IoT in gathering, sending, and evaluating data from
human bioinformatics systems, we analyze how smart farming approaches might
maximize crop yields and resource use. This convergence not only improves agri-
cultural sustainability but also gives people more control over how they monitor and
take care of their health.

Keywords Sensor · Personalized · Healthcare · Agriculture · Bioinformatics · Smart

9.1 Introduction

The nexus of smart farming and human bioinformatics presents a new strategy to
tackle these intricate problems at a time when the globe is dealing with the chal-
lenges of a growing population, depleting natural resources, and rising health
concerns. Utilizing IoT and sensor technology, smart farming develops data-driven
precision agriculture systems that maximize resource efficiency and boost produc-
tivity. Similar technology is also used by human bioinformatics to track and examine
health-related data for individualized medical insights. The symbiotic relationship
between these two fields is examined in this chapter along with its potential to alter
both agriculture and healthcare.

S. K. Jain (✉) · P. K. Jain

Department of Computer Science and Engineering, Shri Vaishnav Vidyapeeth
Vishwavidyalaya, Indore, Madhya Pradesh, India
e-mail: [email protected]; [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 139
A. Khamparia et al. (eds.), Microbial Data Intelligence and Computational
Techniques for Sustainable Computing, Microorganisms for Sustainability 47,
https://doi.org/10.1007/978-981-99-9621-6_9
140 S. K. Jain and P. K. Jain

9.2 Smart Farming and IoT: Enhancing Agricultural

Productivity

Precision agriculture (Routray et al. 2019; Zhang et al. 2021), commonly referred to
as smart farming, is a contemporary method of farming that makes use of technology
and data-driven strategies to maximize crop output and resource management. To
collect real-time information regarding soil conditions, weather patterns, and crop
growth, it involves the integration of numerous technologies, including sensors,
drones, GPS, and data analytics. Making informed decisions about planting, irriga-
tion, fertilizer, and pest management is then possible thanks to this data. Smart
farming aims to maximize crop yields while improving agricultural efficiency,
cutting waste, and ensuring sustainable practices (Ramesh et al. 2023).
The term “Internet of Things” (IoT) describes a system of networked physical
objects, including machines, cars, and buildings, that are equipped with sensors,
software, and network connectivity. Without the need for direct human involvement,
these devices may gather and exchange data online. The IoT enables seamless
connection between systems and devices, allowing them to cooperate and share
information on their own (Al-Fuqaha et al. 2015). IoT devices are essential for
gathering and transferring data for analysis and decision-making in the context of
smart farming and human bioinformatics.
Electronic components known as sensor devices are used to monitor and detect
many physical characteristics, including temperature, humidity, light, and pressure.
As they offer real-time data that guides decision-making, these gadgets are essential
to both smart farming and human bioinformatics. Agricultural fields, hospitals,
houses, and wearable technology are just a few of the areas where sensors can be
installed to collect pertinent data points that support data-driven insights.
By giving farmers previously unattainable insights and control over their opera-
tions, IoT and sensor device integration is changing the agriculture industry. These
innovations support data-driven decision-making, resource optimization, and
increased crop yields. Here, we look at how the Internet of Things (IoT) and sensor
technology are changing agriculture through initiatives like resource optimization,
precision farming, environmental monitoring, and data analysis.

9.2.1 IoT in Agriculture

Real-time monitoring of the environment, the health of the soil, and crop develop-
ment is made possible by the integration of IoT devices into farming systems.
Sensors gather information on variables like temperature, humidity, soil moisture,
and nutrient concentrations. Weather stations that give farmers precise local weather
information so they can decide how to sow, irrigate their fields, and manage pests are
an example of this real-time data collecting. Through the creation of a network of
9 Synergizing Smart Farming and Human Bioinformatics Through IoT and. . . 141

connected data points by these IoT devices, conventional farming becomes a

precision-based activity (Khan et al. 2020; Singh et al. 2019).

9.2.2 Precision Agriculture

Farmers may make informed decisions about irrigation, fertilization, and pest control
by utilizing the data gathered by IoT devices. This focused strategy increases yields
while reducing resource waste (Rezk et al. 2021). The transformational potential of
IoT in agriculture is best illustrated by precision agricultural technologies, such as
automated irrigation systems that modify water distribution based on soil moisture
levels. As a result, productivity is increased while water use and environmental
effect are decreased.

9.2.3 Predictive Analytics

The gathered data is processed using sophisticated machine learning and data
analytics algorithms to discover patterns in crop health, predict disease outbreaks,
and optimize planting schedules (Almalki et al. 2021). Farmers can forecast disease
outbreaks or insect infestations and take preventive actions using historical data
analysis and machine learning models. These forecasts are essential for avoiding
crop losses and maintaining environmentally friendly farming methods.

9.2.4 Resource Optimization

IoT-driven smart farming reduces the negative effects of agriculture on the environ-
ment by maximizing water use, minimizing chemical use, and limiting soil erosion.
Farmers may precisely apply fertilizers by closely observing the state of the soil and
the nutrient content, reducing overuse that might be harmful to the environment.
Similar to this, the combination of precision irrigation with IoT ensures that water is
used effectively, cutting down on waste and protecting water resources.

9.3 Benefits of IoT and Sensor Technologies in Agriculture

Enhanced Efficiency: IoT and sensor devices remove guessing by supplying real-
time data, enabling farmers to react quickly to changing circumstances (Syed et al.
2019).
142 S. K. Jain and P. K. Jain

Resource Conservation: By using pesticides, fertilizers, and water only when

essential, precision agriculture lowers waste and benefits the environment while
saving money.
Increased Yield and Quality: By keeping an eye on and optimizing environmental
conditions, crops are healthier, yields are better, and product quality is higher.
Data-Driven Decisions: Data analytics allow for well-informed choices on
resource allocation, planting, and harvesting, reducing risks and optimizing profits.
Sustainability: IoT-driven agriculture helps promote sustainable farming methods
by maximizing resource use and minimizing environmental effect.
Risk Reduction: Timely action can be taken in the event that diseases, pests, or
unfavorable weather conditions are discovered early.

9.4 Human Bioinformatics and IoT: Revolutionizing

Healthcare

The gathering, archiving, analysis, and interpretation of biological data pertaining to

human health and well-being are all part of human bioinformatics (Yoosefzadeh
Najafabadi et al. 2023). It includes a variety of data types, such as genomic data,
medical records, and wearable device data. These data sets are processed using
bioinformatics tools and techniques in order to spot trends and derive valuable
insights. Understanding unique health profiles, identifying illness risks, and creating
individualized medical therapies are the ultimate goals.

9.4.1 IoT in Healthcare

Remote patient monitoring has been made possible by wearable technology and
medical sensors, which allow for the continuous collection of health-related data
such as vital signs and activity levels. A steady stream of health-related data is
provided by wearable medical devices including glucose monitors, smartwatches,
and fitness trackers. Individuals and healthcare professionals can monitor health
trends and take appropriate action using this data that is communicated over IoT.

9.4.2 Personalized Health Insights

People can learn more about their health and well-being by analyzing the data
collected by wearable technology. This proactive strategy makes it easier to spot
anomalies and chronic illnesses early on. People learn more about their health by
analyzing information on their heart rate variability, sleep patterns, and physical
9 Synergizing Smart Farming and Human Bioinformatics Through IoT and. . . 143

activity. Timely medical interventions and lifestyle changes can result from early
discovery of abnormalities.

9.4.3 Disease Management

Healthcare professionals can monitor patients with chronic illnesses and modify
treatment strategies in real time based on obtained data with the help of IoT-enabled
bioinformatics tools. The use of remote monitoring is advantageous for patients with
long-term illnesses like diabetes or hypertension. Real-time data is sent to healthcare
professionals, allowing them to better manage diseases and adapt treatment pro-
grams to the needs of specific patients.

9.4.4 Healthcare Accessibility

Remote patient monitoring fills in gaps in healthcare access by bringing medical

treatment to underserved groups across geographic boundaries. Healthcare solutions
powered by the Internet of Things (IoT) bring medical care to isolated or rural
locations where access to medical institutions may be limited. Wearable tech-
enabled telemedicine provides consultations and diagnostics without requiring
in-person visits.

9.5 Synergy Between Smart Farming and Human

Bioinformatics

The chance to unlock synergies that go beyond their respective fields is fascinating
given the convergence of smart farming and human bioinformatics (Anand et al.
2023) (Barbhuiya and Ahmad 2021). This technological synergy delivers significant
advantages for agriculture, human health, and the environment through a dynamic
interplay. Let’s get into the core aim of this chapter by examining the intersections
and potential links between these two areas of study while emphasizing the advan-
tages of fusing knowledge from both disciplines.

9.5.1 Environment-Health Nexus

Understanding how environmental elements and human health are connected is one
of the most fascinating intersections between smart farming and human
144 S. K. Jain and P. K. Jain

bioinformatics. Both agricultural growth and human well-being are substantially

impacted by environmental factors. We can decipher complex correlations between
air quality, pollen counts, humidity, and respiratory health by fusing data from
environmental sensors in agricultural fields with health-related wearable technology.
Better disease prevention and management tactics may result from this holistic
viewpoint.

9.5.2 Shared Data Analytics

In order to acquire actionable insights, both smart farming and human bioinformatics
significantly rely on data analytics. Comprehensive analysis may result from com-
bining data sets from these domains, for instance, uncovering hidden links that may
have an effect on public health by comparing crop growth patterns with regional
disease outbreaks or changes in air quality. This common data-driven methodology
improves the ability to forecast outcomes, allowing for timely interventions in both
agriculture and healthcare.

9.5.3 Nutritional Sustainability

We may examine the nutritional implications of food production and human health
thanks to the junction of these domains. We may create meals suited to individual
health needs while taking environmental sustainability into account by fusing
knowledge from smart farming practices, which maximize crop nutrient content
and production. This not only fosters human well-being but also responsible agri-
culture that places a priority on creating foods that are nutrient-rich.

9.5.4 Early Warning Systems

Environmental and health monitoring devices with IoT capabilities can serve as
early warning systems for potential dangers. For instance, we can pinpoint locations
with increased health risks by examining patterns of environmental pollutants and
comparing them with the prevalence of particular health disorders. Through predic-
tive analytics, farmers can also gain the advantage of early crop disease or insect
infestation detection, lowering the requirement for excessive chemical use and
lowering the risk of ecosystem destruction.
9 Synergizing Smart Farming and Human Bioinformatics Through IoT and. . . 145

9.6 Benefits of Combining Insights

9.6.1 Holistic Approach

By combining smart farming with human bioinformatics, we can take a holistic

approach to well-being that takes into account both human and environmental
health. Decisions in both domains are made with greater knowledge thanks to this
all-encompassing viewpoint (Shameer et al. 2017).

9.6.2 Cross-Domain Findings

New findings may result from insights gained at the nexus of human bioinformatics
and smart farming. Finding out how particular agricultural methods impact human
health, or the other way around, brings up new research and innovation
opportunities.

9.6.3 Enhanced Resource Management

We can enhance resource management tactics by exchanging insights between

various sectors. For instance, information on the best irrigation techniques for crop
growth can influence suggestions for how much water to use to stay hydrated and
healthy.
Customized health programs are made possible by the integration of agricultural
and medical data. Dietary advice can be modified to meet a person’s specific
nutritional requirements as well as to support organic farming methods.

9.6.4 Impact on Public Health

By addressing both issues of agricultural sustainability and public health, informed

policies and actions may be developed, creating healthier environments and
communities.
146 S. K. Jain and P. K. Jain

9.7 Challenges, Limitations, and Ethical Considerations

While the fusion of human bioinformatics with smart farming brings exciting
possibilities, there are also some difficulties, restrictions, and moral dilemmas that
need to be carefully considered (Doukas et al. 2012; Quy et al. 2022).

9.7.1 Data Security and Privacy

Concerns about data privacy and security arise when massive amounts of private
information from the healthcare and agricultural industries are combined. The
legislative frameworks for personal health data and agriculture data are distinct,
necessitating strong steps to ensure data protection, encryption, and secure commu-
nication. Such data could be accessed without authorization, which could have
serious ethical and legal repercussions.

9.7.2 Data Accuracy and Quality

Accurate data is essential for both smart farming and human bioinformatics. Inac-
curate measurements, data transfer issues, or sensor faults might produce inaccurate
conclusions. Having accurate data is essential to avoid making the wrong choices,
whether they relate to agricultural or medicinal operations.

9.7.3 Interoperability and Standardization

IoT gadgets and sensor systems may be produced by several companies and run on
various operating systems. It can be difficult to achieve seamless integration and
compatibility between various devices. To ensure data interoperability and effective
device connection, standardization activities are required.

9.7.4 Technology Access and Literacy

Not all farmers and people in general have access to cutting-edge IoT technologies or
the knowledge required to understand intricate data analyses. Ensuring equitable
access and promoting digital literacy among all stakeholders is essential to prevent
creating technology-driven disparities.
9 Synergizing Smart Farming and Human Bioinformatics Through IoT and. . . 147

9.7.5 Environmental Impact

The increased deployment of IoT devices in agriculture can lead to concerns about
the environmental impact of manufacturing, energy consumption, and electronic
waste. Balancing the benefits of technology adoption with its environmental costs is
crucial for sustainable implementation.

9.8 Future Possibilities, Innovations, and Research Areas

There are numerous opportunities for future innovation and research due to the
fusion of smart agriculture and human bioinformatics (Wolfert et al. 2017).

9.8.1 Predictive Public Health Models

By combining environmental data from smart farming with health data, predictive
models that foresee dangers to public health can be created. Early warnings of illness
epidemics associated with environmental factors could help prompt policy choices
and responses.

9.8.2 Data-Driven Nutritional Sustainability

Upcoming studies may concentrate on improving crop nutrition in accordance with

people’s particular health requirements and environmental sustainability. Healthy
diets and more sustainable food production would both benefit from this interdisci-
plinary approach.

9.8.3 Environmentally Friendly Precision Agriculture

New developments in precision agriculture may result in more environmentally

friendly methods. Integration with human bioinformatics might enable the develop-
ment of precision pesticide application strategies that minimize harm to both crops
and ecosystems.
148 S. K. Jain and P. K. Jain

9.8.4 Solutions for Personalized Agri-Health

Studies could look into the possibilities for personalized agri-health practices to
improve human health. For instance, knowing how particular crops affect people’s
health profiles could result in tailored dietary suggestions.

9.8.5 Ethical Data Governance

It is crucial to create moral frameworks for sharing and governing data. Future
studies should focus on topics like data ownership, informed consent for data
usage, and developing open systems for data sharing between the agricultural and
healthcare sectors.

9.8.6 Cross-Disciplinary Training

Educational initiatives that close the knowledge gap between the fields of agriculture
and medicine could equip professionals with the abilities necessary to fully realize
the potential of these interconnected domains.

9.8.7 Climate-Health Resilience

By combining insights from smart farming and human bioinformatics, researchers

could develop strategies to enhance climate and health resilience, addressing the
growing challenges posed by climate change.

9.9 Conclusion

Through IoT and sensor technology, smart farming and human bioinformatics have
come together, creating a paradigm shift that transcends conventional limits. This
synergy has the potential to transform agriculture into a data-driven, sustainable
industry that boosts sustainable practices, increases illness prevention, and
empowers people to take control of their own health. The future promises tremen-
dous potential for the improvement of both people and the environment as technol-
ogy develops and interdisciplinary collaborations grow.
9 Synergizing Smart Farming and Human Bioinformatics Through IoT and. . . 149

References

Al-Fuqaha A, Guizani M, Mohammadi M, Aledhari M, Ayyash M (2015) The application of

internet of things (IoT) to smart farming: a review. IEEE Internet Things J:857–875
Almalki FA et al (2021) A low-cost platform for environmental smart farming monitoring system
based on IoT and UAVs. Sustainability 13(11):5908
Anand KJ et al (2023) Enhancing crop improvement through synergistic integration of advanced
plant breeding and proximal remote sensing techniques: a review. Int J Plant Soil Sci 35(19):
121–138
Barbhuiya RK, Ahmad N (2021) IoT applications in translational bioinformatics. In: Translational
bioinformatics in healthcare and medicine. Academic Press, pp 69–79
Doukas C et al (2012) Enabling data protection through PKI encryption in IoT m-Health
devices. In: 2012 IEEE 12th international conference on bioinformatics & bioengineering
(BIBE). IEEE
Khan A, Khan MY, Khan HA, Alam A, Khan MA, Singh RN, Al-Garadi MA, Al-Ghamdi MA
(2020) IoT-based smart farming: a comprehensive review. Comput Electron Agric:107041
Quy VK et al (2022) IoT-enabled smart agriculture: architecture, applications, and challenges. Appl
Sci 12(7):3396
Ramesh A, Karthikeyan S, Karthik M (2023) Smart farming: a review of internet of things (IoT)
applications in agriculture. J Clean Prod:12345–12355
Rezk NG et al (2021) An efficient IoT based smart farming system using machine learning
algorithms. Multimed Tools Appl 80:773–797
Routray SK et al (2019) Internet of things based precision agriculture for developing countries. In:
2019 International Conference on Smart Systems and Inventive Technology (ICSSIT). IEEE
Shameer K et al (2017) Translational bioinformatics in the era of real-time biomedical, health care
and wellness data streams. Brief Bioinform 18(1):105–124
Singh AK, Singh PK, Singh RK (2019) The role of IoT in smart farming: a comprehensive review.
ACM Trans Sensor Netw:1–24
Syed L et al (2019) Data science algorithms and techniques for smart healthcare using IoT and big
data analytics. In: Mishra M, Mishra B, Patel Y, Misra R (eds) Smart techniques for a smarter
planet: towards smarter algorithms, vol 374. Springer, Cham, pp 211–241
Wolfert S et al (2017) Big data in smart farming—a review. Agric Syst 153:69–80
Yoosefzadeh Najafabadi M, Hesami M, Eskandari M (2023) Machine learning-assisted approaches
in modernized plant breeding programs. Genes 14(4):777
Zhang M et al (2021) Wearable Internet of Things enabled precision livestock farming in smart
farms: a review of technical solutions for precise perception, biocompatibility, and sustainability
monitoring. J Clean Prod 312:127712
Chapter 10
Deep Learning-Assisted Techniques
for Detection and Prediction of Colorectal
Cancer From Medical Images
and Microbial Modality

Ravi Kumar , Amritpal Singh, and Aditya Khamparia

Abstract In the past decade, significant progress has been made in the fields of
artificial intelligence, machine learning, and deep learning (DL). These advance-
ments have opened up wide applications and opportunities in the medical field.
Colorectal cancer (CRC) has gained substantial interest from researchers due to its
ranking as the third most prevalent cancer type after breast and lung cancer, affecting
around 10% of all cancer patients globally each year. It is the second leading cause of
cancer-related death worldwide, making the development of efficient computer-
assisted methods for its detection, prediction, and treatment crucial. There are
modalities used for colorectal cancer screening and detection such as colonoscopy
images, biopsy samples, biomarker data, blood samples, CT scan, MRI, ultrasound,
PET, and microbial data. The advancement of technology has made deep learning an
attractive choice for fast and effective detection, segmentation, and prediction of
diseases through image analysis. This technology has the potential to assist and
empower medical professionals in making timely and informed decisions. Deep
learning has proven to be highly effective when ample high-quality features and
large datasets are available. However, one of the main challenges in using deep
learning for medical image analysis is the limited availability of datasets from
medical centers. This chapter provides an overview of DL-based models and their
application in detecting and predicting CRC from various modalities. On the

R. Kumar (✉)
Research Scholar, Department of Computer Science Engineering, Lovely Professional
University, Punjab, India
Department of Computer Science Engineering, Jawaharlal Nehru Government Engineering
College, Sundernagar, Mandi, Himachal Pradesh, India
e-mail: [email protected]
A. Singh
Department of Computer Science Engineering, Lovely Professional University, Punjab, India
A. Khamparia
Department of Computer Science, Babasaheb Bhimrao Ambedkar University, Amethi, India

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 151
A. Khamparia et al. (eds.), Microbial Data Intelligence and Computational
Techniques for Sustainable Computing, Microorganisms for Sustainability 47,
https://doi.org/10.1007/978-981-99-9621-6_10
152 R. Kumar et al.

SCPolyps dataset the OEM model achieved training and test accuracy of 98% and
96% respectively.

Keywords Deep learning · Colorectal cancer · Detection · Segmentation ·

Prediction

10.1 Introduction

Based on the 2020 global cancer statistics, colorectal cancer (CRC) holds the third
position among the most prevalent cancer types worldwide, following breast and
lung cancer (Sung et al. 2021). It affects approximately 10% of all cancer patients
annually, impacting both men and women. CRC, which is regrettably the second
leading cause of fatalities due to cancer worldwide, is just behind lung cancer. In the
same year, the World Health Organization reported approximately ten million
fatalities attributed to cancer alongside an alarming rise of 19.3 million newly
diagnosed cancer cases (Sung et al. 2021). People over the age of 50 are particularly
vulnerable, with hereditary factors posing the greatest risk at 35%, along with other
elements like smoking, unhealthy eating habits, and obesity (Arnold et al. 2017).
CRC is a significant worldwide health issue and one of the primary causes of
fatalities due to cancer worldwide. Early CRC detection leads to timely treatment
and thus improves patient outcome.
There are many tests used for colorectal cancer diagnosis and detection such as
colonoscopy, biopsy, biomarker testing, blood tests, CT scan, MRIs, ultrasound, and
positron emission tomography (PET). Generally, biopsy is preferred for cancer
diagnosis but colonoscopy is the most efficient method for CRC screening (Colo-
rectal Cancer 2022). There are mainly four domains in existing AI applications for
CRC as listed below (Qiu et al. 2022). The most common technique used for
screening is colonoscopy. It is considered the most reliable technique for CRC
screening. These techniques require experienced experts otherwise they may lead
to omission and misdiagnosis. Accuracy can be improved using electronic medical
records and omic data with these methods (Winawer 2007). The radiography and
pathological examination are mainly used for diagnosis and staging of CRC (Goyal
et al. 2020). Chemotherapy, surgery, and radiotherapy are the three most popular
CRC treatments (Gao et al. 2020).The prediction of CRC recurrence and estimation
of the survival period comes under prognostic analysis. Disease progression is
estimated using statistical methods, e.g., Cox regression (Zhu et al. 2020).
Recent research has focused on the application of machine learning (ML)- and
deep learning (DL)-based technologies in the medical arena, presenting great oppor-
tunities for the creation of efficient computer-assisted approaches in cancer diagno-
sis, prediction, and treatment (Thakur et al. 2023; Srivastava et al. 2022). This
chapter provides an introduction to leveraging DL for the detection of CRC.
Traditional approaches to colorectal cancer (CRC) detection rely heavily on
human expertise, including visual examination of colonoscopy images and analysis
of biopsy sample reports by experienced endoscopists. However, this subjective
approach can be susceptible to human error, potentially resulting in missed or
10 Deep Learning-Assisted Techniques for Detection and Prediction. . . 153

misdiagnosed cases of CRC. Furthermore, the increasing demand for colonoscopy

screenings strains healthcare systems, lead to longer waiting times and delayed
diagnoses. Deep learning presents an opportunity to automate and enhance this
process, ultimately improving early detection rates. One of the fundamental require-
ments for effective deep learning models is access to large datasets for training.
However, acquiring substantial volumes of medical data can be challenging. Trans-
fer learning, generative adversarial networks (GANs), and data augmentation
emerge as a solution to this problem. Transfer learning enables the models to be
trained effectively with smaller datasets. This approach involves fine-tuning of
pretrained models, which were initially trained on extensive datasets like ImageNet,
with minimal adjustments to achieve optimal results in medical image analysis and
cancer detection tasks (Thakur et al. 2023).
The remainder of this chapter is organized as follows. The overview of deep
neural network-based techniques for medical image analysis is given in Sect. 10.2.
The deep learning-based detection of colorectal cancer using various data modalities
is presented in Sect. 10.3. The methodology for classification of tumor and
non-tumor images is presented in Sect. 10.4. Section 10.5 presents results and
discussions. Finally the conclusion is presented in Sect. 10.6.

10.2 Deep Neural Networks for Image Analysis

A crucial area of research in medical imaging is computerized diagnosis. Various

imaging modalities are employed in medical applications, including colonoscopy
images, whole slide images (WSI), RNA sequences, microbial samples, CT (com-
puted tomography) scan images, magnetic resonance imaging (MRI), X-ray images,
and positron emission tomography (PET). In the past, basic techniques like
thresholding, region growing, and edge tracing were utilized for medical image
processing as shown in Fig. 10.1 (Qiu et al. 2022). However, the increasing volume

Input Layer Hidden Layer Output Layer

Tumor Image
Tumor in Colon

Normal Colon
Normal Colon

Input Images Predicted Images

Fig. 10.1 Deep neural network for medical image classification

154 R. Kumar et al.

and complexity of medical imaging data have spurred the adoption of machine
learning techniques.
Traditional machine learning approaches, relying on manually engineered fea-
tures, often necessitate substantial manual effort in algorithm design. This limitation
has led to the adoption of neural networks. The widespread adoption of artificial
neural networks is enabled by factors such as the accessibility of data and advance-
ments in computational processing capabilities (Winawer 2007). The development
of DL techniques, particularly convolutional neural networks (CNNs), has signifi-
cantly expanded the possibilities for automating and enhancing the processing of
medical images. The neural network-based approaches for medical image analysis
are described below.

10.2.1 Convolutional Neural Networks

CNN is a special type of DL model that proved to be very efficient and useful for
processing and analyzing visual data such as images and videos. Convolution is
joining two functions to obtain a third function. It is applied on input images to
extract important information. It plays a crucial role in computer vision tasks such as
medical image analysis, classification, object detection, and segmentation (Thakur
et al. 2023).

10.2.2 Transfer Learning Models

The primary issue with DL is that a lot of data is required for it to perform at its best.
But mostly it is very difficult to get or create large datasets, especially in the medical
domain. This problem can be resolved to a great extent by transfer learning. With
transfer learning it is possible to train models with less data and obtain optimal
results. Many deep learning models, pretrained on large datasets like ImageNet, are
available for transfer learning across various image analysis tasks. Popular
pretrained models include: ResNet—residual networks are known for their depth
and skip connections; Inception (GoogLeNet in 2014)—uses multiple filter sizes in
parallel; VGGNet (2014 ILSVRC competition)—simple and effective architecture
with small 3 × 3 convolutional filters; DenseNet—connects each layer to every other
layer for enhanced feature reuse; and EfficientNet—focuses on model efficiency by
balancing depth, width, and resolution.
10 Deep Learning-Assisted Techniques for Detection and Prediction. . . 155

10.2.3 Ensemble Learning Models

Ensemble learning is a powerful technique in DL and ML where multiple models are

combined to improve robustness, predictive performance, and generalization. One of
the drawbacks of ensemble models is that they require more resources.

10.2.4 Object Detection Models

Object detection models locate and classify objects within an image. Well-known
models include YOLO, faster R-CNN, and SSD (Krenzer et al. 2023).

10.2.5 Image Segmentation Models

Image segmentation models classify each pixel in an image, distinguishing different

objects or regions. U-Net, FCN (fully convolutional network), and SegNet are
popular choices for semantic and instance segmentation.

10.2.6 Network Pruning

Network pruning or network compression is currently gaining much attention . In

network pruning the model is compressed and becomes lightweight while preserving
useful features and demonstrating comparable performance as compared to the
original model. These lightweight models are very easy to deploy and also utilize
much fewer resources than their counterparts. Li-SegPNet is a lightweight encoder-
decoder model for colorectal polyps segmentation. Debesh Jha introduced NanoNet,
a lightweight model that utilizes YOLOX for real-time polyp detection, group
normalization, and video-adjacent frame association algorithm for unstable polyp
detection.

10.2.7 Deep Belief Networks (DBNs)

In a variety of domains, including medical image analysis, DBNs have been applied
to obtain hierarchical features from images and improve classification and segmen-
tation tasks. They are less frequently utilized for medical image processing tasks
than other neural network architectures like CNNs.
156 R. Kumar et al.

10.2.8 Generative Adversarial Networks (GANs)

GANs are used for image generation, style transfer, and super-resolution. Notable
GAN architectures include deep convolutional GAN (DCGAN), CycleGAN, and
BigGAN. GAN is a powerful ML model developed by Goodfellow, I. et al. in 2014
(Park et al. 2023). Since then, GANs have significantly advanced the field of
generative modeling and have found use in a number of fields, including computer
vision and natural language processing. A GAN consists of two neural networks that
are trained concurrently, one of which is a generator and the other a discriminator.
The generator attempts to produce data samples (e.g., images and text) that resemble
genuine data using random noise as input, which is often taken from a straightfor-
ward distribution like the Gaussian distribution. The generator is trained to reduce
the likelihood that the discriminator will identify the samples it generates as fraud-
ulent. A binary classifier called the discriminator seeks to distinguish between
authentic and fraudulent data. It has been honed to increase the likelihood that it
will accurately identify real data as real and fake data as fake (Park et al. 2023).

10.2.9 Transformers

Transformer-based models have made significant inroads into medical image anal-
ysis, offering state-of-the-art performance in various tasks. Dense transformer
extends the traditional transformer architecture for dense prediction tasks, making
it suitable for tasks like medical image segmentation. It utilizes self-attention
mechanisms to capture contextual information. MedT is designed specifically for
medical imaging tasks. It combines the transformer architecture with spatial atten-
tion mechanisms tailored for the unique characteristics of medical images.
TransUNet integrates transformers with a U-Net architecture, which is well suited
for medical image segmentation. It combines the strength of transformers in captur-
ing long-range dependencies with the spatial awareness of U-Net. RETINAnet is
adapted from object detection models and employs transformers for the detection
and localization of retinal lesions, making it valuable for diabetic retinopathy
screening. TransV3D extends transformer architectures to three-dimensional medi-
cal image data. It is suitable for tasks such as organ or tumor segmentation in 3D
medical scans.
10 Deep Learning-Assisted Techniques for Detection and Prediction. . . 157

10.3 Deep Learning for Classification, Segmentation,

Detection, and Prediction of CRC From Different
Modalities

Colorectal cancer stands among the most prevalent cancer types, ranking as the
second most common in females and the third in males. Detecting and removing
CRC can substantially lower the risk of patient fatalities. As a result, researchers are
dedicated to exploring deep learning (DL) solutions to automate the detection of
CRC using different modalities (Gao et al. 2020). Various imaging modalities can be
employed in medical applications, including colonoscopy images, whole slide
images (WSI), RNA sequences, microbial samples, CT (computed tomography)
scan images, magnetic resonance imaging (MRI), X-ray images, and positron
emission tomography (PET). The different modalities considered by various
researchers for detection and prediction of CRC using ML and DL are shown in
Fig. 10.2.

10.3.1 CRC Detection and Prediction Using Endoscopic

Images and Videos

The colonoscopy is the gold standard for colorectal cancer screening. Automatic
CRC detection from endoscopic images becomes possible with the advancement in
technology and deep learning techniques. One effective approach to early detection
is the analysis of polyp images, as polyps are often precursors to colorectal cancer
(Goyal et al. 2020). By leveraging their ability to automatically learn complex

Fig. 10.2 Modalities for

detection and prediction of
colorectal cancer
158 R. Kumar et al.

patterns and features from large datasets, DL techniques can aid in the early
detection of CRC from colon images (Gao et al. 2020). Some of the techniques for
CRC classification and segmentation from colonoscopy images and videos are
discussed here. Pogorelov et al. (2017) proposed the Kvasir dataset developed
from endoscopic images of the gastrointestinal tract and implemented machine
learning and deep learning models on this dataset. Choi et al. (2020) applied
InceptionV3, ResNet50, and DenseNet161 for detecting CRC using endoscopic
images. DenseNet161 achieved the highest accuracy rate of 92.48%. Yao et al.
(2021) implemented UNet with pretrained VGG19 and Jha et al. (2021) proposed
NanoNet, a lightweight DL-based model for semantic segmentation and categoriza-
tion of tissue (Polyps) automatically. It has been observed that it performs differently
in different datasets but for the Kvasir-SEG dataset, NanoNet-C used much fewer
parameters, and had better DSC, mean IoU, and frames/second than the ResUnet
model. They concluded that a dataset consisting of small and large sized polyps is
required for better and more realistic results. Some authors proposed instance
segmentation with DL, an end-to-end learning and attention-based DL model for
classification, segmentation, and localization of polyps from colonoscopy images
(Nogueira-Rodríguez et al. 2021; Akilandeswari et al. 2022; Wang et al. 2022a;
Yang et al. 2022). Colonoscopy is performed for detection of polyps which may lead
to cancer in later stages. Cancer confirmation was done by taking tissue samples
from polyps and investigating them under a microscope.

10.3.2 CRC Detection and Prediction Using Tissue Images

(Biopsy Samples)

Tissue images are very helpful in detecting and classifying CRC. Some of the studies
based on biopsy samples for CRC detection and classification are discussed here. Lu,
L. et al. and Tsai, M. J. et al. proposed transfer learning-based deep learning models
such as ResNet50, AlexNet, VGG19, GoogLeNet, SqueezeNet, and InceptionV3 for
the early prediction of metastatic CRC from haematoxylin and eosin (H&E)-stained
tissue images. The image augmentation was applied for scaling the dataset. Their
model achieved mean accuracy of 94.8% for cancer slides. The H&E-stained tissue
images were taken from public datasets (Lu et al. 2021; Tsai and Tao 2021). Another
researcher proposed a slide-based artificial intelligence predictor (SBAIP), with
ResNet-18, a DL model, to predict lymph node cancer conditions from histological
images of colorectal cancer and logistic regression classifier for clinical data such as
patient age, sex, T-stage, tumor location, and sidedness as input (Kiehl et al. 2021).
The H&E-stained slide images of the CAMELYON16 challenge were given as input
to the ResNet-18 model for feature extraction. Some authors proposed supervised
and semi-supervised deep learning, faster region-based convolutional neural net-
work (R-CNN) with a ResNet101, and end-to-end learning with EfficientNet-B0 for
examining CRC using WSI images on whole slide images. The benign and
10 Deep Learning-Assisted Techniques for Detection and Prediction. . . 159

inflammatory tissues were categorized as low risk and dysplasia and malignant
tissues were categorized as high risk. In the future, annotations and effective use
of unlabeled data are required to improve efficiency. These models achieved accu-
racy around 94% (Yu et al. 2021; Ho et al. 2022; Teichmann et al. 2022). The images
were divided into patches or tiles, and then the shape and size of the cells were
analyzed to detect CRC.

10.3.3 CRC Detection Using Omic Data

Genetic sequence and biomarkers can be used to detect and predict CRC. Lee
et al. implemented the InceptionV3 model and a Random Forest classifier for the
classification of CRC based upon MSI (microsatellite instability) in 20 mm magni-
fied patches of WSI as an input. The data was taken from The Cancer Genome Atlas
(TCGA) dataset and samples were classified as cancerous and non-cancerous. For
the TCGA datasets, the model obtained AUROC of 0.892 (Lee et al. 2021). The
researcher proposed end-to-end DL utilizing WSI for detecting molecular changes.
The DL was implemented using k-Siamese CNN architecture, Efficientnet-B0, and
stochastic gradient-descent (SGD) with Adam heuristic. The model achieved AUC
scores of up to 94% (Teichmann et al. 2022). Liu et al. implemented ResNet-18 with
the attention-based Multiple Instance Learning (MMIL) model for preopera-
tive Lymph Node Metastatis (LNM) assessment to predict CRC. Blood and tumor
tissue serum biomarkers were used for prediction. The model attained an AUC of
0.855 for T1, 0.832 for T2, 0.691 for T3, and 0.792 for stage T4 of CRC. In the
future, a multiclass classification can be done (Liu et al. 2022). Zhao et al. proposed
deep learning techniques using enhanced venous-phase CT and RNA sequencing
patterns to predict CRC (Zhao et al. 2022). The RNA sequencing genes of tumor
tissues and CT images of CRC were gathered and examined by Spearman’s
correlation.

10.3.4 CRC Detection From MRI, (FDG)-PET, and CT

Scan Data

Vorontsov et al. (2019) applied DL for segmenting liver regions from CT images of
patients with CRC liver metastases. They applied the fully connected network
(FCN)-based deep learning model for lesion size <10 mm, lesion size 10–20 mm,
and lesion size >20 mm (Yang et al. 2022). Yuan et al. implemented the ResNet3D
model with the support vector machine (SVM) classifier to detect peritoneal carci-
nomatosis in colorectal cancer using preoperative high-contrast CT images. Their
model achieved accuracy of 94.11% and an AUC of 0.922 (Wang et al. 2022b).
Wang et al. proposed a dense residual single-axis super resolution network on
abdominal CT images for non-peritonealized CRC diagnosis (Yang and Liu
160 R. Kumar et al.

2020). The SPSS 22.0 software was used for analysis. To diagnose CRC T-staging,
the extramural depth and the extramural vascular invasion grading must be applied
to the CT image of the tissue. The author concluded that CT can be used for
preoperative staging of non-peritonealized colorectal cancer. The author proposed
deep learning techniques using enhanced venous-phase CT and RNA sequencing
patterns to predict CRC (Zhao et al. 2022). Yang et al. proposed a FDG-PET/CT-
based diagnostic model for CRC using regional LNM. The SPSS 20.0 was used for
statistical analysis and Medcalc software for receiver operator characteristic (ROC)
curves. The model’s AUC was 95% but it had high sensitivity (He et al. 2021). He
et al. proposed ML-based techniques for the estimation of regional LNM using
FDG-PET/CT images (Li et al. 2021). They implemented least absolute shrinkage
and selection operator approach and five ML techniques out of which logistic
regression achieved an AUC of 0.866 and eXtreme gradient boosting achieved an
AUC of 0.903 and outperformed the other models. Li et al. proposed a classification
method of LNM in CRC using MRI images. Transfer learning was implemented
using pre-trained weights of AlexNet. The model obtained accuracy of 0.8358 and
an AUC of 0.8569. Only the LNM > 3 mm was considered for the study (Baxter
et al. 2016).

10.3.5 Deep Learning for Detecting Colorectal Cancer Using

Microbial Data

In recent studies on detection and screening of CRC it has been observed that
microbial data also plays a significant role in detecting CRC and the role of bacteria
in increasing cancerous growth in affected areas. Though few studies have used gut
microbiota as a diagnostic biomarker for CRC, the evidence discussed next showed
an association between gut microbiota and CRC. Collectively, these studies showed
how ML and DL are effective in the processing of microbiome data for the
identification and categorization of colorectal cancer. In order to increase the sensi-
tivity of fecal immunochemical testing for colonic lesion diagnosis, Baxter et al.
(2016) used a Random Forest approach to build a microbiota-based model, which
produced a reasonable prediction performance with an AUC of 67.3%. The
MicrobiomeHD dataset was used in Namkung’s paper from 2020, which investi-
gated a number of machine learning techniques, such as SVM, RF, and ANN. Their
models impressively attained an AUC of 90%, demonstrating strong predictive
ability (Namkung 2020). Topçuoğlu et al. (2020), using techniques including Deci-
sion Trees, SVM, XGBoost, and RF, established a framework to solve microbiome-
based categorization problems. The AUROC of their Random Forest model was
0.680 (Topçuoğlu et al. 2020). AUC scores of 0.96 and 0.89 on two different
microbiome datasets were achieved in 2021 by Mulenga et al. using CNN to classify
colorectal cancer from gut microbiome data, highlighting the possibility for excellent
predictive accuracy (Mulenga et al. 2021). A collection of 1056 public fecal samples
10 Deep Learning-Assisted Techniques for Detection and Prediction. . . 161

was used by Wu et al. (2021) to identify microbial markers for early colorectal
cancer detection using Random Forest, yielding an AUC of 0.89 and indicating
reliable prediction performance (Wu et al. 2021). In their 2023 study, Lu et al. used
ML methods such as naive Bayes, RF, and logistic regression to identify colorectal
cancer using information on the gut microbiome. With an AUC of 0.926 and an
accuracy rate of 91.7%, their model displayed remarkable predictive accuracy
(Lu et al. 2023). Yinhang et al. (2023) concentrated on using RF and SVM models
to estimate CRC lymph node metastasis, identifying critical microbiological vari-
ables such as Lachnospiraceae_FCS020_group and Lachnospiraceae_UCG 004 as
significant contributions (Yinhang et al. 2023). These results demonstrate the poten-
tial of ML to enhance CRC detection using microbiome data. They demonstrate
differing levels of predictive efficacy while utilizing different algorithms,
underscoring the field’s exciting potential for microbiome-based models.
We have discussed different modalities that can be utilized by different
researchers to detect colorectal cancer. In the next section we will implement deep
learning techniques for the early detection of CRC. The early detection of CRC plays
a vital role in reducing mortality rates and improving patient outcomes. Detecting
and removing precancerous polyps during colonoscopy significantly reduces the risk
of developing CRC. The next section describes the workflow for early detection of
CRC from colonoscopy images.

10.4 Methodology

The classification workflow for colon diseases or colorectal cancer is described as

follows. Firstly, the dataset is downloaded from a public source and subsequently
unzipped for further analysis. The images within the dataset are resized, normalized,
and subjected to data augmentation techniques. The augmentation helps in increas-
ing the size and diversity of the dataset. The second step involves splitting the dataset
into two sets: a training set and a test set. This division allows for effective model
training, validation, and evaluation. Thirdly, the CNN models, pretrained on the
ImageNet, are loaded. The top layers of these models are changed to enable the
classification of colon diseases. The model training process begins in the subsequent
phase, utilizing the training set data. Fourthly, the models are compiled, setting
important hyper-parameters such as the optimization method, learning rate, loss
function, dropout rate, and regularization techniques and then validating the trained
model using the validation dataset. This step provides an opportunity to evaluate the
model’s generalization and performance on unseen data. Finally the model’s perfor-
mance is verified using the test dataset. The results are also visually verified by
displaying the corresponding images. The overall process of classifying colon
illnesses or colorectal cancer is visually depicted in Fig. 10.3, presented as a
flowchart.
162 R. Kumar et al.

Fig. 10.3 Workflow

10.4.1 Dataset

There are very few datasets available for colorectal cancer detection because the
acquisition of medical data is a very tedious task, and low-quality images hinder
accurate prediction. The quality of the dataset plays a crucial role in the performance
of the model. In this study three datasets with endoscopic images were selected for
detection and prediction of CRC. The dataset used for this task was SCPolyps as
shown in Fig. 10.4. SCPolyps is a self-created dataset that has 3000 images with four
categories, normal, polyp, esophagus, and ulcerative, with 750 images each. In this
study we have only used normal and polyp images. The sample polyp and normal
colon images are shown in Fig. 10.4.

10.4.2 Model Selection for CRC Classification

The models selected for this study were VGG19 (Simonyan and Zisserman 2014),
InceptionV3 (Szegedy et al. 2015), and our ensemble model. The transfer learning
was implemented using VGG19, InceptionV3, and OEM. The ensemble model was
created by combining two models (EfficientNet (He et al. 2016) and ResNet50 (Tan
and Le 2019)) and taking the average. Hyper-parameters used were batch size,
epochs, dropout, and regularization (L2). The weights of models pretrained on
10 Deep Learning-Assisted Techniques for Detection and Prediction. . . 163

Fig. 10.4 Sample normal colon and polyp images of SCPolyps dataset. (a) Normal colon. (b)
Polyp

ImageNet were used during transfer learning. Then the models were fine-tuned by
varying hyper-parameters to obtain optimal results.

10.4.3 Tools

All the models were implemented on Google Colab using Python.

10.4.4 Model Evaluation

In our study, classification of two classes and evaluation were done on the basis of
loss, accuracy, precision, recall, and AUC and F1 score. The calculation of accuracy,
precision, recall, and F1 score was done as given in Eqs. (10.1)–(10.4), respectively.

TP þ TN
Accuracy = ð10:1Þ
TP þ TN þ FP þ FN
TP
Precision = ð10:2Þ
TP þ FP
TP
Recall = ð10:3Þ
TP þ FN
Precision  Recall
F1 score = 2  ð10:4Þ
Precision þ Recall
164 R. Kumar et al.

10.5 Results and Discussion

10.5.1 Model Implementation

VGG19, InceptionV3, and OEM were implemented with the same configuration on
Google Colab. All three models were evaluated on the basis of accuracy, precision,
recall, and F1 score. Transfer learning was implemented on the models. Hyper-
parameters like batch size, epochs, L2-regularization, and dropout rate were used.
Models were fine-tuned by varying hyper-parameters for optimal results.

10.5.2 Model Performance on Datasets

The performance of these models on the basis of selected parameters on different

datasets is presented in tabular as well as graphical form. The results for the
SCPolyps training dataset are given in Table 10.1.
The model performance was also evaluated on the SCPolyps test set and the
results are shown in Table 10.2.
It has been observed that OEM outperformed the other models in training and test
accuracy, precision, recall, AUC, and F1 values. The performance of OEM on the
SCPolyps dataset is shown in Fig. 10.5. OEM outperformed the other models in
classifying polyps on the SCPolyps dataset.
The OEM model correctly predicted polyp and normal colon images as shown in
Figs. 10.6 and 10.7 respectively.

Table 10.1 Performance of models on SCPolyps training set

Model Loss Accuracy Precision Recall AUC F1
InceptionV3 0.8109 0.9262 0.8809 0.9785 0.9405 0.8944
VGG19 0.487 0.9861 0.9861 0.9863 0.9862 0.9862
OEM 0.0012 0.995 0.995 0.996 0.9951 0.9954

Table 10.2 Performance of models on SCPolyps test set

Model Loss Accuracy Precision Recall AUC F1
InceptionV3 0.1338 0.9007 0.8110 0.9433 0.8896 0.8944
VGG19 0.803 0.9858 0.9857 0.9858 0.9858 0.9858
OEM 0.0125 0.993 0.993 0.994 0.9928 0.9929
10 Deep Learning-Assisted Techniques for Detection and Prediction. . . 165

Learning curve
1.0

0.8
value (log scale)

Training accuracy
0.6 Test accuracy
training loss
Test loss
0.4
best model

0.2

0.0
0.0 2.5 5.0 7.5 10.0 12.5 15.0 17.5
Epochs

Fig. 10.5 Performance of OEM on SCPolyps training set and test set

Fig. 10.6 OEM predicting

polyp
166 R. Kumar et al.

Fig. 10.7 OEM predicting

normal colon

10.6 Conclusion

In this study, different deep learning methods for medical image analysis were
discussed as well as different modalities that are used for detection and prediction
of CRC. The deep learning models were implemented to detect polyps which may
lead to CRC in later stages from colonoscopy images. On the SCPolyps dataset, the
OEM model achieved training and test accuracy of 98% and 96% respectively. The
challenge of identifying colorectal cancer emphasizes the value of precise prediction
models. When used in conjunction with the expertise of competent medical practi-
tioners, these models considerably contribute to the formulation of more accurate
diagnoses. By applying deep learning, AI-driven systems continue to improve in this
field, and there is still plenty of room for further research and improvement.

References

Akilandeswari D, Sungeetha CJ, Thaiyalnayaki K, Baskaran K, Jothi Ramalingam R,

Al-Lohedan H, Al-dhayan DM, Karnan M, Hadish KM (2022) Automatic detection and
segmentation of colorectal cancer with deep residual convolutional neural network. Evid
Based Complement Alternat Med 2022:3415603. https://doi.org/10.1155/2022/3415603
Arnold M, Sierra MS, Laversanne M, Soerjomataram I, Jemal A, Bray F (2017) Global patterns and
trends in colorectal cancer incidence and mortality. Gut 66(4):683–691. https://doi.org/10.1136/
gutjnl-2015-310912
10 Deep Learning-Assisted Techniques for Detection and Prediction. . . 167

Baxter NT, Ruffin MT, Rogers MAM et al (2016) Microbiota-based model improves the sensitivity
of fecal immunochemical test for detecting colonic lesions. Genome Med 8:37. https://doi.org/
10.1186/s13073-016-0290-3
Choi K, Choi SJ, Kim ES (2020) Computer-Aided diagnosis for colorectal cancer using deep
learning with visual explanations. In: 42nd annual international conference of the IEEE engi-
neering in medicine and biology society (EMBC). IEEE Publications, pp 1156–1159. https://
doi.org/10.1109/EMBC44109.2020.9176653
Colorectal Cancer (2022). https://www.cancer.net/cancer-types/colorectal-cancer/screening, last
accessed on August 20, 2022
Gao Y, Zhang XX, Li S, Lu Y (2020) Application of artificial intelligence technology in the
diagnosis and treatment of colorectal cancer. Chin J Gastrointest Surg 23:1155–1158
Goyal H, Mann R, Gandhi Z, Perisetti A, Ali A, Aman Ali K, Sharma N, Saligram S, Tharian B,
Inamdar S (2020) Scope of artificial intelligence in screening and diagnosis of colorectal cancer.
J Clin Med 9(10):3313. https://doi.org/10.3390/jcm9103313
He K, Zhang X, Ren S, Sun J (2016) Deep residual learning for image recognition. In: Proceedings
of the IEEE conference on computer vision and pattern recognition, pp 770–778
He J, Wang Q, Zhang Y, Wu H, Zhou Y, Zhao S (2021) Preoperative prediction of regional lymph
node metastasis of colorectal cancer based on 18F-FDG PET/CT and machine learning. Ann
Nucl Med 35(5):617–627. https://doi.org/10.1007/s12149-021-01605-8
Ho C, Zhao Z, Chen XF, Sauer J, Saraf SA, Jialdasani R, Taghipour K, Sathe A, Khor LY, Lim KH,
Leow WQ (2022) A promising deep learning-assistive algorithm for histopathological screening
of colorectal cancer. Sci Rep 12(1):2222. https://doi.org/10.1038/s41598-022-06264-x
Jha D, Tomar NK, Ali S, Riegler MA, Johansen HD, Johansen D, de Lange T, Halvorsen P (2021)
Nanonet: Real-time polyp segmentation in video capsule endoscopy and colonoscopy. In: 34th
International symposium on computer-based medical systems (CBMS). IEEE Publications, pp
37–43
Kiehl L, Kuntz S, Höhn J, Jutzi T, Krieghoff-Henning E, Kather JN, Holland-Letz T, Kopp-
Schneider A, Chang-Claude J, Brobeil A, von Kalle C, Fröhling S, Alwers E, Brenner H,
Hoffmeister M, Brinker TJ (2021) Deep learning can predict lymph node status directly from
histology in colorectal cancer. Eur J Cancer 157:464–473. https://doi.org/10.1016/j.ejca.2021.
08.039
Krenzer A, Heil S, Fitting D, Matti S, Zoller WG, Hann A, Puppe F (2023) Automated classification
of polyps using deep learning architectures and few-shot learning. BMC Med Imaging 23(1):59
Lee SH, Song IH, Jang H-J (2021) Feasibility of deep learning-based fully automated classification
of microsatellite instability in tissue slides of colorectal cancer. Int J Cancer 149(3):728–740.
https://doi.org/10.1002/ijc.33599
Li J, Wang P, Zhou Y, Liang H, Lu Y, Luan K (2021) A novel classification method of lymph node
metastasis in colorectal cancer. Bioengineered 12(1):2007–2021. https://doi.org/10.1080/
21655979.2021.1930333
Liu H, Zhao Y, Yang F, Lou X, Wu F, Li H, Xing X, Peng T, Menze B, Huang J, Zhang S, Han A,
Yao J, Fan X (2022) Preoperative prediction of lymph node metastasis in colorectal cancer with
deep learning. BME Front 12:1–12, article ID 9860179. https://doi.org/10.34133/2022/9860179
Lu L, Dercle L, Zhao B, Schwartz LH (2021) Deep learning for the prediction of early on-treatment
response in metastatic colorectal cancer from serial medical imaging. Nat Commun 12(1):6654.
https://doi.org/10.1038/s41467-021-26990-6
Lu F, Lei T, Zhou J, Liang H, Cui P, Zuo T et al (2023) Using gut microbiota as a diagnostic tool for
colorectal cancer: machine learning techniques reveal promising results. J Med Microbiol 72(6):
001699
Mulenga M et al (2021) Feature extension of gut microbiome data for deep neural network-based
colorectal cancer classification. IEEE Access 9:23565–23578. https://doi.org/10.1109/
ACCESS.2021.3050838
Namkung J (2020) Machine learning methods for microbiome studies. J Microbiol 58:206–216.
https://doi.org/10.1007/s12275-020-0066-8
168 R. Kumar et al.

Nogueira-Rodríguez A, Domínguez-Carbajales R, López-Fernández H, Iglesias Á, Cubiella J,

Fdez-Riverola F, Reboiro-Jato M, Glez-Peña D (2021) Deep neural networks approaches for
detecting and classifying colorectal polyps. Neurocomputing 423:721–734. https://doi.org/10.
1016/j.neucom.2020.02.123
Park HC, Hong IP, Poudel S, Choi C (2023) Data augmentation based on generative adversarial
networks for endoscopic image classification. IEEE Access
Pogorelov K, Randel KR, Griwodz C, Eskeland SL, de Lange T, Johansen D, Spampinato C, Dang-
Nguyen DT, Lux M, Schmidt PT, Riegler M (2017) Kvasir: a multi-class image dataset for
computer aided gastrointestinal disease detection. In: Proceedings of the 8th ACM on Multi-
media Systems Conference, pp 164–169
Qiu H, Ding S, Liu J, Wang L, Wang X (2022) Applications of artificial intelligence in screening,
diagnosis, treatment, and prognosis of colorectal cancer. Curr Oncol 29(3):1773–1795
Simonyan K, Zisserman A (2014) Very deep convolutional networks for large-scale image recog-
nition. arXiv preprint arXiv:1409.1556
Srivastava A, Tomar NK, Bagci U, Jha D (2022) Video capsule endoscopy classification using focal
modulation guided convolutional neural network. In: 2022 IEEE 35th international symposium
on computer-based medical systems (CBMS). IEEE, pp 323–328
Sung H, Ferlay J, Siegel RL, Laversanne M, Soerjomataram I, Jemal A et al (2021) Global cancer
statistics 2020: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in
185 countries. CA Cancer J Clin 71:209–249
Szegedy C, Liu W, Jia Y, Sermanet P, Reed S, Anguelov D, Erhan D, Vanhoucke V, Rabinovich A
(2015) Going deeper with convolutions. In: Proceedings of the IEEE conference on computer
vision and pattern recognition (CVPR)
Tan M, Le Q (2019) Efficientnet: rethinking model scaling for convolutional neural networks. In:
International conference on machine learning. PMLR, pp 6105–6114
Teichmann M, Aichert A, Bohnenberger H, Ströbel P, Heimann T (2022) End-to-end learning for
image-based detection of molecular alterations in digital pathology. arXiv preprint
arXiv:2207.00095.
Thakur D, Saini JK, Srinivasan S (2023) DeepThink IoT: the strength of deep learning in internet of
things. Artif Intell Rev 56:14663–14730
Topçuoğlu BD, Lesniak NA, Ruffin MT IV, Wiens J, Schloss PD (2020) A framework for effective
application of machine learning to microbiome-based classification problems. MBio 11(3):
10–1128
Tsai MJ, Tao YH (2021) Deep learning techniques for the classification of colorectal cancer tissue.
Electronics 10(14):1662. https://doi.org/10.3390/electronics10141662
Vorontsov E, Cerny M, Régnier P, Di Jorio L, Pal CJ, Lapointe R, Vandenbroucke-Menu F,
Turcotte S, Kadoury S, Tang A (2019) Deep learning for automated segmentation of liver
lesions at CT in patients with colorectal cancer liver metastases. Radiol Artif Intell 1(2):180014.
https://doi.org/10.1148/ryai.2019180014
Wang D, Chen S, Sun X, Chen Q, Cao Y, Liu B, Liu X (2022a) AFP-mask: anchor-free polyp
instance segmentation in colonoscopy. IEEE J Biomed Health Inform 26(7):2995–3006. https://
doi.org/10.1109/JBHI.2022.3147686
Wang X, Guo C, Zha Y, Xu K, Liu X (2022b) Diagnosis of nonperitonealized colorectal cancer
with computerized tomography image features under deep learning. Contrast Media Mol
Imaging 2022:1886406. https://doi.org/10.1155/2022/1886406
Winawer SJ (2007) The multidisciplinary management of gastrointestinal cancer. Colorectal cancer
screening. Best Pract Res Clin Gastroenterol 21(6):1031–1048. https://doi.org/10.1016/j.bpg.
2007.09.004
Wu Y, Jiao N, Zhu R, Zhang Y, Wu D, Wang AJ et al (2021) Identification of microbial markers
across populations in early detection of colorectal cancer. Nat Commun 12(1):3063
Yang Z, Liu Z (2020) The efficacy of 18F-FDG PET/CT-based diagnostic model in the diagnosis of
colorectal cancer regional lymph node metastasis. Saudi J Biol Sci 27(3):805–811. https://doi.
org/10.1016/j.sjbs.2019.12.017
10 Deep Learning-Assisted Techniques for Detection and Prediction. . . 169

Yang K, Chang S, Tian Z, Gao C, Du Y, Zhang X, Liu K, Meng J, Xue L (2022) Automatic polyp
detection and segmentation using shuffle efficient channel attention network. Alex Eng J 61(1):
917–926. https://doi.org/10.1016/j.aej.2021.04.072
Yao Y, Gou S, Tian R, Zhang X, He S (2021) Automated classification and segmentation in
colorectal images based on self-paced transfer network. Biomed Res Int 2021:6683931. https://
doi.org/10.1155/2021/6683931
Yinhang W, Jing Z, Jie Z, Yin J, Xinyue W, Yifei S et al (2023) Prediction model of colorectal
cancer (CRC) lymph node metastasis based on intestinal bacteria. Clin Transl Oncol 25(6):1–12
Yu G, Sun K, Xu C, Shi XH, Wu C, Xie T, Meng R-Q, Meng XH, Wang KS, Xiao HM, Deng HW
(2021) Accurate recognition of colorectal cancer with semi-supervised deep learning on path-
ological images. Nat Commun 12(1):6311. https://doi.org/10.1038/s41467-021-26643-8
Zhao J, Wang H, Zhang Y, Wang R, Liu Q, Li J, Li X, Huang H, Zhang J, Zeng Z, Zhang J, Yi Z,
Zeng F, Zeng F (2022) Deep learning radiomics model related with genomics phenotypes for
lymph node metastasis prediction in colorectal cancer. Radiother Oncol 167:195–202. https://
doi.org/10.1016/j.radonc.2021.12.031
Zhu W, Xie L, Han J, Guo X (2020) The application of deep learning in cancer prognosis
prediction. Cancers 12(3):603. https://doi.org/10.3390/cancers12030603
Chapter 11
Smart Farming and Human Bioinformatics
System Based on Context-Aware
Computing Systems

Sini Anna Alex, T. P. Pallavi, and G. C. Akshatha

Abstract Fertilizer is an important product that contributes to the growth of crops.

As soil nutrients decrease, exotic and special fertilizers such as nitrogen, phosphorus,
potassium, calcium, magnesium, soy milk, and sulfur are replenished in the soil.
Anyhow the use of chemical fertilizers affects the lifestyle of farmers as well as the
health of their crops. This chapter addresses the health of farmers and the health of
crops by analyzing the characteristics of soil, environmental characteristics, and
characteristics of healthy farmers. The main health problems of farmers include skin
problems, lung problems, heart diseases, and cancer. This advice can suggest to
farmers the best fertilizer to use to increase future crops. Hadoop Distributed File
System (HDFS) has four levels of processing, like data polishing, extraction of
features and matching similarity, binary analysis, and data clustering. The first stage
cleans the data, removes missing values, and then performs data normalization and
component decomposition. In the second stage, soil, environmental, and farmer
health characteristics are extracted. The similarity is then evaluated based on the
construction of ontology-supported grid reduction (OMR) to predict farmers’ health
problems. In the third stage, the FP-growth algorithm and densely connected recur-
rent neural network (DC-RNN) are used to classify healthy farmers and healthy
crops. In the fourth stage, the last group of farmers is presented with product health
information from the self-planning map and, accordingly, product and fertilizer
recommendations that will reduce health risks. Recommendations were made by
HDFS, and performance was evaluated concerning parameters like precision, recall,
F measurement, and accuracy on papaya, banana, and leafy vegetables.

S. A. Alex (✉) · G. C. Akshatha

Department of CSE (AI & ML), Ramaiah Institute of Technology, Bangalore, Karnataka, India
e-mail: [email protected]; [email protected]
T. P. Pallavi
Department of CSE (Cyber Security), Ramaiah Institute of Technology, Bangalore, Karnataka,
India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 171
A. Khamparia et al. (eds.), Microbial Data Intelligence and Computational
Techniques for Sustainable Computing, Microorganisms for Sustainability 47,
https://doi.org/10.1007/978-981-99-9621-6_11
172 S. A. Alex et al.

Keywords Context-aware computing · Clustering · Crop health · Health risks ·

Recommendation

11.1 Introduction

Agriculture is the main source of food production in different climates, fertilizers,

pesticides, and soil types (Shirsath et al. 2017). The use of chemical fertilizers can be
harmful to health. In general, excessive use of chemical fertilizers leads to reduced
crop yields and affects the health of farmers. On the other hand, inadequate use of
fertilizer will not support crop growth. Therefore, the best fertilizer should be used in
the soil to increase productivity every year. Using faster learning machines such as
Adaptive Neuro-Fuzzy Interference System (ANFIS), Support Vector Machine
(SVM) (Karandish et al. 2017) to predict fertilizer concentration. Nitrogen (N) is
an essential fertilizer that helps the growth of new fruits such as apples, peaches,
peaches, oranges, walnuts, olives, and kiwis (Carranca et al. 2018). Like nitrogen,
other commonly used nutrients are phosphorus (P), potassium (K), calcium (Ca),
magnesium (Mg), sulfur (S), iron (Fe), manganese (MN), copper (Cu), and zinc
(Zn) (Moreno et al. 2018). Nutrients in the form of a person’s manure should enrich
the consistency of the soil to promote the growth of crops.
Fertilizer recommendation systems specifically take into account soil parameters
to obtain the required results (Jethva et al. 2018). The simple approval process is
shown in Fig. 11.1. In agriculture, farmers have to suffer from deadly diseases.
Cultivation is often done by farmers without gloves and dust covers, causing health
hazards. Farmers who are not protected and use fertilizer unbalanced will experience
health problems (Nguyen Viet et al. 2019). Determine the root cause of pesticides by
modeling machine learning algorithms such as artificial neural network (ANN),
K-nearest neighbor (K-NN), deep learning (DL), ensemble learning (EL), and
SVM (Tomiazzi et al. 2019; Liakos et al. 2018). Use classification techniques to
predict farmers’ health problems. Health risks include health effects such as sore
throat, headache, fatigue, skin irritation, eye irritation, and difficulty breathing

1. Climatic Factors
Data collection from 2. soil Nutrients
agricultural environment 3. Fertilizers and
Pesticides Crops or
fertilizers

Variety of crops for

different duration
from peculiar region Data Analysis Recommendation
using algorithms to Farmer

Fig. 11.1 Agriculture recommendation system

11 Smart Farming and Human Bioinformatics System Based on Context-Aware. . . 173

(Sonchieu et al. 2018). Due to these health effects, food availability in agricultural
lands is determined by yield, soil content, and nutritional status (Raut et al. 2018).
Get more out of big data by using cluster algorithms to predict crops. Use
clustering algorithms based on classification around central point (PAM), clustering
of uppercase letters (CLARA), and speed-based matching of applications with noise
(DBSCAN) (Majumdar et al. 2017). Crop yield also depends on fertilizer use. In the
analysis, it is tested for quality and product density (Senthil Vadivu et al. 2017).
Nitrogenous fertilizer will be given, while the split plant trial is carried out. There-
fore, the nutrients nitrogen, phosphorus, and potassium are all depleted, and rice is
very useful. Climate change is also having a greater impact on growth and quality.
In addition to crops, other health risks posed by fertilizers to farmers include heart
disease, immune system disorders, skin problems, and cancer (Zhang et al. 2019).
Fertilizer use also affects crop growth and farmers’ health.

11.1.1 Motivation

Ecological impacts related to food agriculture; but it also depends on how the
fertilizer is used. The two main reasons for using fertilizer in agriculture are
profitability and health risks. Many studies have examined farming techniques
based on crop health analysis and farmer health analysis, focusing on agricultural
characteristics and health characteristics (genetic problems and symptoms), respec-
tively. The main purpose of this research is to determine crop health and farmers’
health, depending on fertilizer use.
Based on this motivation, two principal objectives are defined:
• To develop a larger-scale system for evaluating both crop health and farmer
health risks simultaneously with higher accuracy.
• To deliver appropriate recommendations of fertilizers for farmers that help to
increase future yield from the farm by reducing health risks.

11.1.2 Contributions of This Chapter

The main conclusions of this chapter are as follows:

• Dual analysis of farmer health and crop health to recommend better fertilizers to
reduce health risks and increase yields. Hadoop Distributed File System (HDFS)
has been adopted to analyze larger data. The importance of milk consumption for
different crops and its impact on cancer risk was evaluated.
• HDFS works in four stages: data polishing, feature extraction, binary analysis,
and data clustering. In order to ensure the accuracy of the analysis, the first two
stages were completed, and in the last stage, it was suggested that appropriate
suggestions be made.
174 S. A. Alex et al.

• The first stage is data cleaning based on expectation maximization (EM), noise
distance, normalization using Z-score, data normalization, and component
decomposition.
• The second stage is responsible for revealing three different elements such as
land, environment, and farmer health. Farmers’ health characteristics may have
clinical content, so they are extracted from ontology-assisted map reduction
(OMR).
• The third stage plays an important role in identifying the poor, average, and good
for crop growth and farmers’ health. This farmer health classification system is
based on FP-growth algorithm and densely connected recurrent neural network
(DC-RNN) for crop identification.
• The fourth objective is to provide appropriate advice to farmers and ranchers. The
analyzed data was divided into three categories using the self-reporting system
(SOM) to obtain the best recommendations to improve farmers’ agricultural
practices.
• The impact of using different fertilizers on banana, papaya, and vegetable farms
containing lead on health risks for farmers and crops; accuracy was evaluated
based on regression, F-measure, and precision.

11.2 State-of-the-Art

This section is composed of details about previous research works carried out over
the analysis of crops, soil, fertilizer, and recommendations in the field of agriculture
for different crop types.
Management of crop health in agriculture is an important factor limiting produc-
tivity increase. Higher yield means healthier crop. Using multi-model ensemble
technique to predict performance (Wallach et al. 2018). Climate change in the
environment often affects crops. This was done by collecting sufficient climate
change models for 25 crops. Since crop growth in recent days depends on the
characteristics of the soil, the limitation of nutrients in the soil causes crop failure.
In this case, fertilizer is the best medicine to stimulate crop growth. Dynamic Land
Ecosystem Model (DLEM) is used to model crop growth and yield (Xhang et al.
2018). Data were collected and tested for three crops between 1980 and 2012: wheat,
corn, and rice. Due to the large amount of agricultural data, research efforts on
machine learning and artificial intelligence algorithms are still ongoing. Deep neural
networks (DNN) have been used to evaluate maize and bean yields (Kim et al. 2019;
Khaki et al. 2019). The data includes cultivated land, weather data, hydrological
data, and crop yield statistics. DNN generator is equipped with a hidden algorithm,
loss function, optimization, activation function, and output value. Each function
contains an estimate of the sum of the squared error, mean square error, and cross-
entropy.
In Jeong et al. (2016), the Random Forest (RF) algorithm was studied to accu-
rately predict crop yield. Weather conditions and nitrogen fertilizer were taken into
11 Smart Farming and Human Bioinformatics System Based on Context-Aware. . . 175

account when verifying the data. However, the dataset contains some of the fertil-
izers, and the RF algorithm takes more time to build the decision tree, and the
complexity increases as the depth of the tree increases. It needs more help handling
small files. The fuzzy logic system uses two factors, namely rainfall and temperature,
to determine crop yield (Bang et al. 2019). Precipitation characteristics were
modeled using autoregressive moving average (ARMA) and externally variable
autoregressive moving average (ARMAX), then temperature was modeled for
ARMA and seasonal autoregressive integrated moving average (SARIMA). These
two factors are used in the fuzzy system to classify crop yields as good, very good,
medium, bad, and very poor.
Clustering is a solution to predict crop yield by considering multiple character-
istics. Clustering algorithms such as k-means, distribution around medoids (PAM),
clustering of uppercase letters (CLARA), and density-based spatial clustering of
noisy applications (DBSCAN) have been proposed in Afrin et al. (2018). After
collection, crop yield is estimated by linear regression method. The main character-
istics required for separation are soil properties and weather conditions. The clusters
were formed based on similarities in soil nutrients and climate conditions. The three
climate parameters used in this project are temperature, humidity, and precipitation.
Among the four integration methods, the results of the DBSCAN algorithm have
higher accuracy and lower cost than the good one. However, other important factors
such as water level and fertilizer are ignored in this study. In Suresh et al. (2018),
clustering and classification algorithms, such as K-means and improved K-nearest
neighbor algorithms, have been proposed. In this study, we discussed the classifi-
cation as rainwater, groundwater, and cultivated land. The choice of distribution
k value is complex.
Classify soil properties to predict fertilizer application needs. Suchithra and Pai
proposed a machine learning technique (ELM) that can operate using five different
forces (Gaussian radial basis, sine square, hyperbolic tangent, triangle, and constant
limit) (Suchithra and Pai 2020). This function calculates the nutrient fertility index
and pH of the soil based on the soil parameters. The effect of fertilizers on the soil
creates a nutrient imbalance and reduces productivity. ELM cannot provide accurate
results with higher accuracy than real machine learning algorithms. Appropriate
fertilizer can be recommended by distributing nutrients into the soil. Prepare a
recommended fertilizer (Suchithra and Pai 2018). This study used the sigmoid kernel
in a multi-class SVM to provide product recommendations. Parameters in SVM are
optimized by genetic algorithm (GA) and particle swarm optimization (PSO) algo-
rithm. Tuning with GA takes more time, and traditional SVM requires large learning
time and memory consumption, so it is not suitable for larger datasets.
The effects of chemical fertilizers and pesticides pose risks to the environment
(Rahman and Zhang 2018). A binary logistic regression model was applied to the
data collected from farmers in 2016. Fertilizer levels for specific crops were deter-
mined, but health recommendations and risks were not disclosed. Recommendations
to prevent harm to farmers were examined in Mubushar et al. (2019), Ichami et al.
(2019). Fertilizer recommendations are not made using fast machine learning algo-
rithms. A fertilizer recommendation letter was previously developed by volunteers
176 S. A. Alex et al.

who collected information from farmers who also needed to understand the health
issues of fertilizer use and lighting. Most agricultural activities involve growing
crops or approving fertilizers, but both are equally important to increase production
and reduce the health risks of fertilizers to farmers. The main issues identified in this
research area will be continued in the next section.

11.3 Data Analytics in Agriculture

The two main points in agriculture in this research are evaluating the health of
farmers and making recommendations. Health risks to farmers in agriculture arise
from the use of chemical fertilizers and mental and physical stress. The health risks
of neurobehavioral and musculoskeletal problems have been studied (Khan et al.
2019; Sang et al. 2018). Snowball and saturation methods were used in this study,
but the results were incorrect due to a lack of information. To improve the accuracy
of the results, it is suggested to use genetic algorithms to develop fuzzy failure mode
and effects analysis (FMEA). Most genetic algorithm methods consume a lot of
time, and the fuzzy concept has 125 rules that must always work. But this is a better
algorithm and cannot support big data in a short time. Fertilizer-based farmer health
assessment was conducted in Mabe et al. (2017). In this study, farmers asked many
questions about pesticide and fertilizer use. Multiple linear regression models were
used to determine health outcomes. This regression model is capable of handling
linear effects and survey-based statistical errors to predict farmer health.
An agricultural consensus has been formed to measure soil toxicity and inform
farmers. Here, the J48 decision tree algorithm is used for classification; as the height
of the tree increases, more storage space is needed (Pawar and Chillarge 2018).
Similarly, Bodake et al. (2018) proposed the Naive Bayes algorithm for classifica-
tion, but its accuracy is lower. Additionally, using pH to predict soil quality is not
enough because soil consistency also depends on the surrounding air.
The problems mentioned in estimating the welfare of farmers without and with
fertilizer make fertilizer lower. After that, the limit could not be determined in the
agreement, and no information regarding fertilizer was produced. These issues are
well addressed by the solutions proposed in this study and detailed in the next
section.

11.4 Proposed HDFS Recommendation System

11.4.1 System Model

The proposed HDFS recommendation system is modeled by analyzing crop health

and farmer health using multiple features. The recommendation system aims to
provide proper guidance for the use of fertilizer that helps to improve crop growth
11 Smart Farming and Human Bioinformatics System Based on Context-Aware. . . 177

as well as mitigate farmer’s health risks. Multi-feature-based dual analysis in the

HDFS recommendation system is employed to function with four sequential phases
as follows:
• Phase 1: Data polishing.
• Phase 2: Feature extraction and similarity matching.
• Phase 3: Dual analysis.
• Phase 4: Data clustering.
Figure 11.2 depicts the overall working processes handled in the proposed HDFS
recommendation system. The main goal of this system is to analyze different
fertilizers and the risks caused to crop and farmer health due to the utilization of
fertilizer. The result to be obtained from this HDFS recommendation is helpful for
future crop selection and fertilizer selection.

11.4.2 Phase 1: Data Polishing

In this phase, the collected dataset is polished by data cleaning, noise elimination,
normalization, data canonicalization, and component breakdown. This phase is
performed for improving accuracy. The dataset is first processed with a data cleaning
step which re-fills the missing data in the dataset. The expectation maximization
(EM) for data cleaning consists of the expectation (E) step and the maximization (M)
step; the missing value is updated only if convergence is satisfied. Based on E,
missing values are re-filled, and the parameter quality is maximized by M. The
mathematical expressions for E and M are given as:

E⟹Q θjθðtÞ = E zjX,θðtÞ ½logLðθ; X, Z Þ] ð11:1Þ

agr max
M⟹θðtþ1Þ = Q θjθðtÞ ð11:2Þ
θ

Let Z be the missing values in the dataset, θ are the unknown parameters, X is the
currently determined parameter of Z, logL(θ; X, Z ) is the likelihood function, and
Q(θ| θ(t)) is the expected value. After re-filling the missed values in the dataset noise
elimination is handled by the estimation of cosine similarity. This cosine similarity is
applied to identify nearest neighbors, i.e., values very close to the values that are
eliminated. The cosine similarity for two dataset attributes is given as:
n
i Ai Bi
Sim = ð11:3Þ
n 2 n 2
i Ai i Bi
 178

HADOOP DISTRIBUTED FILE SYSTEM

Banana, Papaya, Leafy Vegetables Phase 1 Phase 2

Data Polishing Feature Extraction Extract Semantic Ontology

Similarity
1. Data Cleaning Farmer Health Features
Load Dataset 2. Noise elimination
3. Normalization Ontology Associate
4. Data Canonicalization 1. Soil Features Map-Reduce
5. Component Breakdown 2. Environmental Features

Phase 3 Phase 4

Farmer Health Analysis Crop Health Analysis Data Clustering

FP-Growth algorithm
(Association Rule) Densely Connected-RNN Self-Organized Map

Recommendations
Health Report Clusters Crop Report Clusters
Farmer Agronomist
Best C1,C2,C3 Best C1,C2,C3

Mean C4,C5,C6 Mean C4,C5,C6

Worst C7,C8,C9 Worst C7,C8,C9

Clusters

Fig. 11.2 Proposed HDFS recommendation system

S. A. Alex et al.
11 Smart Farming and Human Bioinformatics System Based on Context-Aware. . . 179

Data Cleaning

Expectation
Re-filling of Missing
Values
Maximization Eliminate Mis-
Canonicalization matching

Noise Elimination
Cosine Similarity Elimination of
between farm fields Noise
Component breakdown Proper values
Normalization assignment

Z-score estimation Elimination of

Redundant Data

Fig. 11.3 Purpose of individual process in data polishing

The terms Ai and Bi are considered features from two specific farm fields; only the
nearest values are eliminated. Then the dataset is normalized using Z-score for
eliminating redundant data. This Z-score is expressed as follows:

x-μ
z= ð11:4Þ
σ

Let x be the raw score, and μ and σ are the mean and standard deviation of the
population. Hereby, the dataset is completely removed with noise and
redundant data.
Then data canonicalization is handled for transforming the data into standard
form. A fertilizer’s representation can be given in more than one form that corre-
sponds to a similar fertilizer. Whatever form is provided, it is transformed into
standard form; by doing canonicalization unwanted mismatching of fertilizers is
mitigated, which enables to increase accuracy in analysis. This process is also
effective in reducing incorrectly represented features. The last step in data polishing
is component breakdown, which enables to split off the improperly specified soil
nutrients. For example, “25–4-2” denotes 25% of nitrogen, 4% of phosphorous, and
2% of potassium. The sequential process handled in data polishing is depicted in
Fig. 11.3 along with the major attainment of each process. On completion of data
polishing, the dataset is taken over for the next phase of processing.

11.4.2.1 Phase 2: Feature Extraction and Similarity Matching

In this phase features are extracted from the dataset; the three different key features
that are extracted are:
180 S. A. Alex et al.

• Soil feature—Nitrogen, phosphorous, potassium, calcium, magnesium, sulfur,

and soymilk.
• Environmental feature—Temperature, rainfall, pH, and water level.
• Health feature—Skin problems, lung problems, heart defects, and cancer.
The soil features are extracted based on the utilization of the amount of fertilizers
in the soil for crop growth. On extraction of these three key features from the dataset
they are processed, before which ontology-based map-reduce process is executed,
especially for the extracted health features. The health feature is submitted by
farmers based on their symptoms. The medical terms are generally scientific,
which requires more details for accurate analysis. So the extracted health features
are processed into the OMR process.
Ontology is constructed based on the relationship between the domain concept
and the topic. In this work the domain concept denotes health risk and the topic
denotes the corresponding symptoms of the farmer. The associated relationship
between symptoms and health risks enables to create ontology. Figure 11.4 demon-
strates the constructed ontology on the left side, where the key, value (k, v) pairs are
generated, and on the right side, the map-reduce processing is carried over.
Split: Initially, the given input symptoms of a farmer are split into individuals.
Map: Mapping functions are enabled to map the occurrences of each symptom in
the ontology using (k, v) pair.
Combine: The mapped (k, v) pairs are consolidated and sorted in order for the
next process.
Reduce: Lastly reducer function is performed to reduce the given inputs. Finally
the reducer supports to identify particular health risks.

Split Map(k,v) from

S15
Ontology Combine
S12 S14
k,v
Mx
k,v Symptom 1 My
k,v Input Farmer Mz
Health Reduce
S3 k,v
Health risk 1 Health risk 2 k,v
Mx
S11
Symptom 2 My
k,v Symptom 1 Mz
Symptom 2
k,v Symptom 3 x Output
S2 S6 S8
Symptom 4 y
z
Mx
k,v k,v k,v Symptom 3 My
S10
S7 Symptom n Mz
S1 k,v
k,v
Health risk 3 Health risk 4
k,v Mx
S4 k,v Symptom n My
k,v
S13 Mz
S9
S5

Fig. 11.4 OMR working model

11 Smart Farming and Human Bioinformatics System Based on Context-Aware. . . 181

11.4.3 Phase 3: Dual Analysis

In this phase, farmer health and crop health are analyzed, and so this phase is named
as dual analysis. Farmer health analysis is handled by FP-growth algorithm, and crop
health analysis is handled by densely connected RNN.
Farmer health analysis—The FP-growth is proposed to generate association rules.
Two processes performed in farmer growth analysis are the construction of FP-tree
and the extraction of association rules. The tree is built from the transactional data
that includes farmer’s health constraints.
Initially the FP-tree is generated in the form of an acyclic graph G = (N, E) in
which N and E denote nodes and edges respectively. Then by traversing through the
path, frequent patterns are extracted, which is followed by the generation of associ-
ation rules.
For instance, Table 11.1 enlists a set of ten transition identities as {T(ID1), T
(ID2), T(ID3), T(ID4), T(ID5), T(ID6), T(ID7), (ID8), T(ID9), T(ID10)} with their
corresponding itemset {a, b, c, d, x, y}. In this proposed work, the itemset denotes
farmer’s health symptoms, diseases, and fertilizers using which the frequent health
risks are extracted. After constructing F-Tree the support count for each transaction
ID is determined, then the larger support count itemsets are filtered out, and the
frequent pattern from all the transactions is extracted. From the frequent itemset,
strong association rules are built.
Let the frequent pattern mined from the FP-growth be p = {A, B. . .}. In associ-
ation rules, support and confidence are two main properties that are given as:

SðA⟹BÞ = PðP [ BÞ ð11:5Þ

C ðA⟹BÞ = PðBjAÞ ð11:6Þ

Association rules are defined from frequent patterns. Here A and B represent
disjoint patterns, and S and C are support and confidence respectively. For generat-
ing strong association rules, the individual frequent pattern from p creates nonempty
subsets Nsb. On taking each Nsb the rule is constructed as follows:

Table 11.1 Transaction data Transaction ID Itemset

T(ID1) a, b, c, d, x, y
T(ID2) d, x, y
T(ID3) a, b, c
T(ID4) c, x, y
T(ID5) a, b
T(ID6) c, d, x, y
T(ID7) b, c, d
T(ID8) x, y
T(ID9) c, d, x, y
T(ID10) b, c, d
182 S. A. Alex et al.

Nsb ⟹ðp - Nsb Þ if C = SðpÞ=SðNsb Þ ≥ C min ð11:7Þ

Cmin denotes minimum confidence value. FP-growth algorithm performs faster in

extracting frequent patterns. The better performance of FP-growth than the conven-
tional a priori algorithm presents absolute farmer health analysis.
Crop health analysis—The crop health is analyzed by using DC-RNN in which
the output determined from the previous layer will be processed as one of the inputs
into subsequent layers. This DC-RNN is chosen since it increases performance by
reprocessing the features from previous layers. The first layer receives input from the
dataset which includes crop features, i.e., soil features and environmental features.
The consecutive layers are connected in the feed-forward model. The potential
benefits of DC-RNN are feature reuse, stronger feature propagation, and others.
The first layer extracts features from the dataset. In this network the features are
processed and concatenated for better results. Let l be a layer in DC-RNN which
receives the following feature maps as (x0, x1, x2, . . ., xl - 1), so the input of the
previous layer is given as:

x1 = H l ð½x0 , x1 , x2 , . . . , xl - 1 ]Þ ð11:8Þ

From the above, [x0, x1, x2, . . ., xl - 1] are the concatenated features between 0, 1,
. . . . . , 1 - l layers, and Hl is the input sample present in lth DC-RNN blocks. The
feature maps are fed into RNN layers. The three operations performed in this
proposed DC-RNN are batch normalization, rectified linear unit (ReLU), and
3 × 3 convolution. Let the output from DC-RNN be Hlk(t) for lth layer has kth feature
of Hlk(∙)with t time. Hereby the output from the layer is mathematically expressed as
shown below:

T T
f ði,jÞ f ði,jÞ
H lk ðt Þ = wðf l,kÞ * H ðl,kÞ ðt Þ þ wrðl,kÞ * H ðl,kÞ ðt - 1Þ þ bðl,kÞ ð11:9Þ

The inputs to the convolution layer and lth recurrent convolution layer are
f ði,jÞ f ði,jÞ
represented as H ðl,kÞ ðt Þ and H ðl,kÞ ðt - 1Þ. Further the weighted values of the convo-
lution neural layer and the recurrent layer are wfðl,kÞ and wrðl,kÞ , respectively, that
denote lth layer and kth feature map, and b(l, k) is the bias. As a result, crop health is
divided into three classes: poor, average, and best.

11.4.4 Phase 4: Data Clustering

SOM is presented in this phase for clustering crop health and farmer health individ-
ually. The SOM clustering is performed by initializing weights, mapping, and
matching units. The weighted values for each input data are initialized, and the
clustering operations are performed after the construction of the map. Then the
11 Smart Farming and Human Bioinformatics System Based on Context-Aware. . . 183

matching unit is predicted using L2 distance. Let X = (x1, x2) and Y = (y1, y2) be two
points for which the L2 distance is predicted as follows:

d= ð y 1 - x1 Þ 2 þ ð y2 - x2 Þ 2 ð11:10Þ

After identifying the best matching unit by using the distance measurement, it is
then updated with weighted values. The weight update is handled by:

W v ðs þ 1Þ = W v ðsÞ þ θðu, v, sÞ:αðsÞ:Dððt Þ - W v ðsÞÞ ð11:11Þ

From the above equation, Wv denotes the weight vector of node v, s is iteration,
θ(u, v, s) is neighborhood function, α(s) is learning restraint, u is the best matching
unit that is present in the map, and D(t) represents the target data vector. On
measuring the distance, similarity between points is predicted, and clusters are
constructed for both health records and crop records.
Based on the selected target vector, the clusters are constructed. Here, we
construct clusters in three categories: best, mean, and worst. Each category is
composed of clusters which enable the prediction of the accurate status of the crop
and farmer health. This clustering is the key to providing a recommendation system.
The health report of best, mean, and worst denotes that farmers are caused with lesser
health risk of cancer, either caused by some other risk and cancer respectively.
Similarly, the crop reports are also clustered in three such categories. The following
pseudo-code illustrates the procedure followed for clustering analyzed data.
The procedure of the SOM Clustering algorithm
Step 1: Initialize weight vectors for nodes.
Step 2: Randomize weight vectors into the map.
Step 3: Pick target input vector//Select best, mean, and worst values.
Step 4: Estimate L2 distance//Predict similarity between the values.
Step 5: Pick the best matching unit.
Step 6: Update weight vectors.
Step 7: Construct a cluster based on the similarity.
The obtained clusters are helpful in forwarding recommendations for farmers.
Hereby the key goal of analyzing farmer health and crop health is attained, and the
recommendation is delivered to farmers regarding the health risks and crop yield.
Understanding the health risk posed by the utilized fertilizer is essential for the
farmer to predict the severity of a particular fertilizer in soil for crop growth.
184 S. A. Alex et al.

11.5 Conclusion

The optimal farm plan in the HDFS environment is designed to analyze the health of
crops and farmers. Although the use of fertilizer in agriculture plays an important
role in crop growth, it can also cause serious harm to farmers. The standard HDFS
environment is completed in four stages: data polishing, feature extraction, binary
analysis, and clustering. To improve the accuracy of our analysis, we perform data
polishing to eliminate unnecessary variables that reduce accuracy. Features were
extracted after data polishing was completed following data cleaning, normalization,
and noise removal. By analyzing the crops’ and farmers’ health, three types of
features were extracted: soil, environment, and farmer. Based on the results
obtained, the farmer’s health conditions were specially processed in the OMR to
show health-related symptoms.
Analysis of two main roles in predicting crop weight and farmers’ welfare using
DC-RNN and FP breeding, respectively. These two methods are preferred because
they are faster and provide more accurate analysis. Finally, the SOM aggregation is
based on best, average, and worst crops and farmer health to provide reasonable
recommendations. The causes of fertilizer deficiency in crops and farmers’ health are
analyzed, and recommendations on fertilizer are given to farmers and agronomists.
Future Work: This HDFS recommendation system is planned to be extended with
the following directions in the future:
• Analysis of different crops that cause farmer health risks under varying concen-
trations of fertilizers from agricultural fields.
• Build an authentication scheme for enabling secure participation of farmers, since
the health details are sensitive.
In the future, integrated deep learning methods can be studied for improving the
performances of HDFS systems in agricultural environments.

References

Afrin S, Khan AT, Mahia M, Ahsan R, Mishal MR, Ahmed W, Rahman RM (2018) Analysis of soil
properties and climatic data to predict crop yields and cluster different agricultural regions of
Bangladesh. In: 2018 IEEE/ACIS 17th international conference on computer and information
science (ICIS) 2018 Jun 6. IEEE, pp 80–85
Bang S et al (2019) Fuzzy logic based crop yield prediction using temperature and rainfall
parameters predicted through ARMA, SARIMA, and ARMAX model. In: 2019 Twelfth
international conference on contemporary computing (IC3). IEEE, pp 1–6
Bodake K, Ghate R, Doshi H, Jadhav P, Tarle B (2018) Soil based fertilizer recommendation
system using Internet of Things. MVP J Eng Sci 1(1):13–19
Carranca C, Brunetto G, Tagliavini M (2018) Nitrogen nutrition of fruit trees to reconcile produc-
tivity and environmental concerns. Plan Theory 7(1):4
Ichami S, Shepherd K, Sila A, Stoorvogel J, Hoffland E (2019) Fertilizer response and nitrogen use
efficiency in African smallholder maize farms. Nutr Cycl Agroecosyst 113(1):1–19
11 Smart Farming and Human Bioinformatics System Based on Context-Aware. . . 185

Jeong J, Resop JP, Mueller N, Fleisher D, Yun K, Butler E, Timlin D, Shim K, Gerber J, Reddy V,
Kim S (2016) Random forests for global and regional crop yield predictions. PLoS One 11(6):
e0156571
Jethva JM et al (2018) A review on data mining techniques for fertilizer recommendation. Int J Sci
Res Comput Sci Eng Inform Technol 3(1):1386–1390
Karandish F, Darzi-Naftchali A, Asgari A (2017) Application of ML models for diagnosing health
hazard of nitrate toxicity in shallow aquifers. Paddy Water Environ 15(1):201–215
Khaki S et al (2019) Crop yield prediction using deep neural networks. J Front Plant Sci 10:621
Khan N, Kennedy A, Cotton J, Brumby S (2019) A pest to mental health, exploring the link
between exposure to agrichemicals in farmers & mental health. Int J Environ Res Public Health
16(8):1327
Kim N, Ha KJ, Park NW, Cho J, Hong S, Lee YW (2019) A comparison between major artificial
intelligence models for crop yield prediction: case study of the Midwestern United States,
2006–2015. Int J Geoinform 8(5):240
Liakos K et al (2018) Machine learning in agriculture: a review. Sensors 18(8):2674
Mabe FN, Talabi K, Danso-Abbeam G (2017) Awareness of health implications of
agrochemical use: effects on maize production in Ejura-Sekyedumase municipality, Ghana.
Adv Agric 2017:7960964. https://doi.org/10.1155/2017/7960964
Majumdar J et al (2017) Analysis of agriculture data using data mining techniques: application of
big data. J Bid Data 4:20
Moreno RH et al (2018) Model of neural networks for fertilizer recommendation and amendments
in pasture crops. In: 2018 ICAI workshops (ICAIW), pp 1–5
Mubushar M et al (2019) Assessment of farmers on their knowledge regarding pesticide usage and
biosafety. Saudi J Biol Sci 26(7):1903–1910
Nguyen Viet H, Grace D, McDermott J (2019) Integrated approaches to tackling health issues-
related to agri-food systems. Int J Public Health 64(1):5–6
Pawar M, Chillarge G (2018) Soil toxicity prediction and recommendation system using data
mining in precision agriculture. In: 2018 3rd international conference for convergence in
technology (I2CT) 2018 Apr 6. IEEE, pp 1–5
Rahman K, Zhang D (2018) Effects of fertilizer broadcasting on the excessive use of inorganic
fertilizers and environmental sustainability. Sustainability 10(3):759
Raut R et al (2018) Soil monitoring, fertigation, and irrigation system using IoT for agricultural
application. In: Intelligent communication and computational technologies: proceedings of
Internet of Things for Technological development, IoT4TD 2017. Springer, Singapore, pp
67–73
Sang AJ, Tay KM, Lim CP, Nahavandi S (2018) Application of a genetic-fuzzy FMEA to rainfed
lowland rice production in sarawak: environmental, health, and safety perspectives. IEEE
Access 6:74628–74647
Senthil Vadivu S et al (2017) Modelling a predictive analytics methodology for forecasting rice
variety and quality on yield on farm and farming attributes using Bigdata. Int J Pure Appl Math
116(5):61–65
Shirsath R et al (2017) Agriculture decision support system using data mining. In: International
conference on intelligent computing and control (I2C2). IEEE
Sonchieu J et al (2018) Health risk among pesticide sellers in Bamenda (Cameroon) and peripheral
areas. Environ Sci Pollut Res 25(10):9454–9460
Suchithra M, Pai M (2018) Improving the performance of sigmoid kernels in multiclass SVM using
optimization techniques for agricultural fertilizer recommendation system. In: Soft computing
systems: second international conference, ICSCS 2018, Kollam, India, April 19–20, 2018,
revised selected papers 2. Springer, Singapore, pp 857–868
Suchithra M, Pai M (2020) Improving the prediction accuracy of soil nutrient classification by
optimizing extreme learning machine parameters. Inform Process Agric 7(1):72–82
186 S. A. Alex et al.

Suresh A, Ganesh Kumar P, Ramalatha M (2018) Prediction of major crop yields of Tamilnadu
using K-means and modified KNN. In: 2018 3rd International conference on communication
and electronics systems. IEEE
Tomiazzi JS et al (2019) Performance of machine-learning algorithms to patterns recognition and
classification of hearing impairment in Brazilian farmers exposed to pesticide and/or cigarette
smoke. Environ Sci Pollut Res 26(7):6481–6491
Wallach D et al (2018) Multimodel ensembles improve predictions of crop-environment-manage-
ment interactions. Glob Change Biol 24(11):5072–5083
Xhang J et al (2018) Improving representation of crop growth and yield in the dynamic land
ecosystem model and its application to China. J Adv Model Earth Syst 10(7):1680–1707
Zhang J et al (2019) Farm machine use and pesticide expenditure in maize production: health and
environment implications. Int J Environ Res Public Health 16(10):1808. https://doi.org/10.
3390/ijerph16101808
Chapter 12
Plant Diseases Diagnosis with Artificial
Intelligence (AI)

Syed Muzammil Munawar, Dhandayuthabani Rajendiran,

and Khaleel Basha Sabjan

Abstract India’s agriculture is significant because of the country’s expanding

population and rising food demands. Therefore, it is necessary to increase crop
productivity. One of these significant factors contributing to reduced agricultural
yields is the prevalence of bacterial, fungal, and viral illnesses. Applying techniques
for plant disease identification helps stop and manage this. Machine learning tech-
niques will be used in the process of identifying plant illnesses since they apply
information most frequently and provide excellent methods for disease diagnosis.
Machine learning-based techniques can be used to identify diseases because they
focus mostly on data superiority outcomes for a certain goal. In this method, machine
learning and deep learning based on artificial intelligence (AI) have been used to
conduct a thorough assessment of the numerous methodologies used in plant disease
diagnosis. In the realm of computer vision, deep learning has also become increas-
ingly important for providing improved performance results for identifying plant
diseases. There has been significant progress in the machine learning and computer
vision fields as a result of the application of deep learning improvements to a variety
of disciplines. In order to demonstrate the superiority of the deep learning model
over the machine learning model, a comparison of the two techniques’ performances
and applications in numerous research articles has been made. The deep learning
technology can be used to identify leaf diseases from collected photos in order to
prevent significant crop losses.

Keywords Agriculture · Neural Networks · Plant diseases detection · Machine

learning methods · Artificial intelligence and deep learning · Supervised learning

S. M. Munawar (✉) · D. Rajendiran · K. B. Sabjan

Department of Biochemistry, C. Abdul Hakeem College (Autonomous), Melvisharam, Vellore,
Tamil Nadu, India

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 187
A. Khamparia et al. (eds.), Microbial Data Intelligence and Computational
Techniques for Sustainable Computing, Microorganisms for Sustainability 47,
https://doi.org/10.1007/978-981-99-9621-6_12
188 S. M. Munawar et al.

12.1 Introduction
A thorough study on several machine and deep learning algorithms for plant disease
recognition and classification has been published. After that, different machine
learning categorization algorithms are useful to perceive plant diseases in an effort
to assist farmers with automatic disease detection of all types of agricultural diseases
to be establish. The study goes over diverse deep learning methods to detecting plant
diseases. Additionally, a number of methods/mappings for identifying the disease
symptoms were summarized. Here, the newest advancements in deep learning
technology in analysis of plant leaf disease the effort determination is valuable
resource to researchers attempting to identify plant diseases. Additionally, a com-
parison flanked by deep learning, machine learning methods is done. Despite the fact
that there has been a lot of notable progress within modern years, nearby a quantity
of research gaps in need to be filled in order so to put into practice efficient strategies
for plant disease identification. Agricultural areas can detect plant leaf illnesses and
accurately report them to the right parties thanks to the integration of IoT, AI, and
unmanned aerial vehicles. No one is interested in farming or agriculture because of
the challenges farmers confront every day in contemporary society so that all
members of the younger generation move to modern cities in order to live safely
and steer clear of such agricultural challenges. The issue of effectively preventing
plant diseases is closely tied to climate and agricultural change (Harakannanavara
et al. 2022). The physiological nature of host-pathogen interactions may vary as a
result of climate change, which may also alter host resistance and the stages and rates
of pathogen development (Das and Sengupta 2020). The issue is made worse by the
ease with which illnesses are disseminated globally today. New diseases may
develop in regions where they haven’t yet been identified and, naturally, where
there isn’t any local expertise to treat them. Careless pesticide use can cause long-
term diseases to develop resistance, which makes it much harder for people to battle
them. One of the fundamental tenets of precision farming is the rapid and precise
identification of plant diseases (Sujatha et al. 2021). In order to address the problem
of long-lasting pathogenic resistance and lessen the negative effects of climate
change, it is essential that no unnecessary expenditures of money or other resources
be made and that the results are of a high quality. The significance of precise and
quick disease identification, especially early impediment, has never been higher in
this changing environment. Plant diseases can be detected using a variety of tech-
niques. A more complete assessment is required when there are no visible signs or
when it is too late to respond. However, because most diseases result in some sort of
outward expression, a skilled professional examination is the primary way for the
detection of plants. A plant pathologist must become increasingly adept at identify-
ing distinguishing symptoms in order to correctly diagnose plant diseases (Bhagat
et al. 2020). The signs of sick plants may be harder for amateurs and enthusiasts to
identify than they are for a trained pathologist, which could result in a wrong
diagnosis. Both inexperienced gardeners and seasoned professionals can benefit
greatly from an automated method created to detect plant illnesses utilizing the
appearance and visual symptoms of the plant as validate of disease diagnosis.
12 Plant Diseases Diagnosis with Artificial Intelligence (AI) 189

Advances in computer vision have the potential to strengthen and expand the
practice of precise plant protection, as well as the market for particular agricultural
computer vision applications. Plant diseases were identified and categorized using
industry-standard digital image processing technologies like color detection and
threshold (Cui et al. 2020). The most popular deep learning method currently
being used for plant disease diagnosis is Convolutional Neural Network (CNN).
Deep learning is a novel development in machine learning, with ground-breaking
results in a number of study areas, including computer vision, pharmaceuticals, and
bioinformatics. Deep learning benefits from being able to use raw data without
directly utilizing manual labor (Ananthi 2020). For two main reasons, deep learning
has recently produced favorable results in both academia and industry (Baidar 2020).
First of all, a large amount of data is generated daily. Therefore, utilizing this
information, a comprehensive model might be developed. Second, the processing
power of the Graphics Processing Unit allows deep models to be developed and
applied to boost compute parallelism. Machine learning has made it possible for PCs
to learn without intentional customization, which is fundamentally analogous to how
people learn. The computer is using information about a few classes of errands to
learn from previous experiences if the presentation of the task improves as more
understanding is gained. A supervised activity is learning. unsupervised, semi-
supervised, and reinforcement.

12.1 Supervised Learning

Specified datasets with input and output boundaries are referred to as supervised
learning for the purposes of developing the models (Panigrahi et al. 2020). An 80:20
split is maintained between data collection and model testing when building a model.
Regression and classification are other subcategories of supervised learning. The
arrangement is an example of a supervised learning task, which produces a discrete
value. This discrete value may have many classes or run concurrently. Relapse is a
supervised learning strategy that yields long-term value as opposed to reach. To
expect a worth that is more in step with production esteem is the aim of the relapse.
Several supervised learning methods include Nearest Neighbor, Gaussian Naive
Bayes, Decision Trees, Support Vector Machine (SVM), and Random Forest.

12.1 Unsupervised Learning

In unsupervised learning, targets are not provided for the model to create displays
for; instead, merely input boundaries are provided. Unsupervised learning can be
classified as either bundling or association (Majeed et al. 2020). Information that has
been organized into groups by various examples and distinguished by a machine-
learning model is clustered, while the term “Association” refers to a procedure based
on standards for classifying relationships between the boundaries of a sizable
informational collection. K-Means Clustering, BIRCH—Balanced Iterative
190 S. M. Munawar et al.

Reducing, and Clustering Using Hierarchies are examples of unsupervised learning

models.

12.1 Semi-supervised Learning

Semi-supervised learning functions similarly to the methods already stated. This
kind of learning is used when dealing with information, both named and unnamed
information. Marks are computed using an unsupervised method, and the determined
attributes are then dealt with using supervised learning techniques. This method is
more well known in image databases, where many of the images are unidentified.

12.1 Reinforcement Learning

Every time information is handled, it is discovered and added to the information that
is being prepared, and the model’s execution keeps improving with criticism to learn
from examples. As a result, it gets more skilled and experienced the more it learns
(Feng and Tian 2020). Algorithms for reinforcement learning include Temporal
Difference, Q-Learning, and Deep Adversarial Networks (Figs. 12.1 and 12.2).

12.1 Techniques and Tools

Pathogens such as bacteria, fungus, nematodes, viruses, pests, weeds, insects,
photoplasma, and other organisms can cause plant illnesses. Based on routine
inspections, ranchers can recognize the symptoms and indicators of a plant’s condi-
tion. Possible warning signals include overflow, a cottony mass, or an apparent mass
on the plant. Among the symptoms are galls, wilt, rots, cankers, necrosis, chlorosis,
as well as underdevelopment and overdevelopment.

Fig. 12.1 Common

stepladder for detection of Input Data base
crops image image
Defect area
classified

Training
Preprocessing image
Abnormal

Feature Classification
Extration
Normal
12 Plant Diseases Diagnosis with Artificial Intelligence (AI) 191

Fig. 12.2 System design

for detecting plant leaf Feature Health
disease extract Segmentation diseased
image

Learn mode
Data CNN
set Segmentation algorithm

Model
Test set Test model

Predicted
results

12.1 Crop Diagnosis

The CropDiagnosis mobile app’s major objective is to provide users with solutions
for executives who are causing users concern with accurate yield diagnosis and
customized application aid. In order to conduct a full survey, yield specifics,
including type, area, soil, and other factors, are gathered and presented (Fegade
and Pawar 2020). Different aspects, like the nature, appearance, and development of
the harvest, are also taken into account, and maybe a diagnosis is made. The program
requires details on finer aspects of development parameters, such as kind, area, soil,
and attributes like look, kind, and development of a plant, in order to make decisions.

12.1 Plantix App

A major aspect of the Plantix app, in addition to some other functions, is the capacity
to identify plant illnesses. The Plantix app was created by the horticultural IT
Company PEAT in Berlin. It is employed to identify soil inadequacies and faults.
Images of plants are used by the software to detect diseases. A few of these photos
are saved in an advanced cell and synced with the image in the worker for diagnosis
(Fegade and Pawar 2020). The Plantix app’s robotized crop disease distinguishing
proof is a key element. Based on images of the hazardous plants that ranchers have
supplied the app’s analysis. The app not only highlights disease symptoms but also
offers advice on how to lessen disease activity as well as useful details on how to
avoid harvest illness the next season. Additionally, the software keeps a database of
diseases so that farmers without internet connection can easily consult it.

12.1 Saillog Agrio

Saillog, an AI tool, can help farmers spot and take care of pests and diseases that
affect their crops. Saillog comes with Agrio, a user-friendly mobile app that is free to
use. Customers of this software send photos of dangerous plants using sophisticated
192 S. M. Munawar et al.

mobile devices. The disease recognition programming is completed when this set of
photographs has been broken down. Additionally, a temporary arrangement is
occasionally offered.

12.1 Future Directions

Machine learning and deep learning are increasingly being used in applications
connected to agriculture. Picture preparation procedures are used for the precise
finding and grouping of harvest illness as well as the precise location and order of the
plant diseases significant for the productive development of the crop. For their ability
to identify plant diseases and treatment options as well as for assisting farmers in
raising the profitability of their crop yields, a variety of commercially available
products are quickly gaining popularity.

Acknowledgments We are thankful to the management and principal of C. Abdul Hakeem

College (Autonomous) Melvisharam, Tamil Nadu, India; to Dr. Deepak Gupta, Maharaja Agrasen
Institute of Technology, India; and Dr. Aditya Khamparia, BabaSaheb Bhimrao Ambedkar Uni-
versity, Lucknow, India, for their encouragement, providing the necessary facilities and support in
carrying out the work.

References
Conflict of Interest The authors declare that there are no conflicts of interest.

Ananthi V (2020) Fused segmentation algorithm for the detection of nutrient deficiency in crops
using SAR images. In: Artificial intelligence techniques for satellite image analysis. Springer, pp
137–159
Baidar T (2020) Rice crop classification and yield estimation using multi-temporal sentinel-2 data: a
case study of Terrain Districts of Nepal.
Bhagat M, Kumar D, Haque I, Munda HS, Bhagat R (2020) Plant leaf disease classification using
grid search based SVM. In: 2nd International conference on data, engineering and applications
(IDEA), pp 1–6
Cui J, Zhang X, Wang W, Wang L (2020) Integration of optical and SAR remote sensing images for
crop-type mapping based on a novel object-oriented feature selection method. Int J Agric Biol
Eng 13:178–190
Das S, Sengupta S (2020) Feature extraction and disease prediction from paddy crops using data
mining techniques. In: Computational intelligence in pattern recognition. Springer, pp 155–163
Fegade TK, Pawar B (2020) Crop prediction using artificial neural network and support vector
machine. In: Data management, analytics and innovation. Springer, pp 311–324
Feng K, Tian RS (2020) Forecasting reference evapotranspiration using data mining and limited
climatic data 54. Taylor & Francis, pp 363–371
Harakannanavara SS, Rudagi JM, Puranikmath VI, Ayesha Siddiqua R, Pramodhini R (2022) Plant
leaf disease detection using computer vision and machine learning algorithms. Global Trans
Proc 3(1):305–310
Majeed Y, Zhang J, Zhang X, Fu L, Karkee M, Zhang Q et al (2020) Deep learning based
segmentation for automated training of apple trees on trellis wires. Comput Electron Agric
170:105277
12 Plant Diseases Diagnosis with Artificial Intelligence (AI) 193

Panigrahi KP, Das H, Sahoo AK, Moharana SC (2020) Maize leaf disease detection and classifi-
cation using machine learning algorithms. In: Progress in computing, analytics and networking.
Springer, pp 659–669
Sujatha R, Chatterjee JM, Jhanjhi N, Brohi SN (2021) Performance of deep learning vs machine
learning in plant leaf disease detection. Microprocess Microsyst 80:103615
Chapter 13
Analyzing the Frontier of AI-Based Plant
Disease Detection: Insights and Perspectives

Mridula Dwivedi, Babita Pandey, and Vipin Saxena

Abstract Plant diseases (PDs) are a significant risk to agriculture all over the world
and constitute a threat not only to economic security but also to nutritional safety.
The utilization of artificial intelligence (AI), to be more precise, machine learning
(ML) and computer vision, has recently emerged as a potentially useful technique for
the early and accurate detection of a PD. The prime objective of this survey is to
provide readers with an in-depth look at the cutting edge on AI-based plant disease
detection (PDD). In this chapter, we explore a number of AI- and ML-based
approaches that have the potential to assist with PDD. In addition, we have shed
light on the potential for AI-driven solutions to be utilized in agricultural contexts,
and we have identified research gaps and difficulties. This study tries to fulfill the
expectation that it will be of use to other researchers, agricultural professionals, and
policymakers in their search for disease-control strategies that are both more suc-
cessful and more permanent. This study not only identifies the AI methods that hold
the most promise for PDD but also brings to light some of the challenges and
problems that remain unresolved that could lead to additional developments in the
field.

Keywords Plant disease detection · AI · Crop · Machine learning

13.1 Introduction

Farming is essential for the global availability of food and economic prosperity.
However, this critical sector faces numerous challenges, with one of the most
formidable being the prevalence of PDs. These diseases not only lead to substantial
crop yield losses but also pose a threat to the livelihoods of millions of farmers
worldwide. Authentic and efficient diagnosis of PDs is imperative for effective
disease management, reducing crop losses, and optimizing resource utilization.

M. Dwivedi (✉) · B. Pandey · V. Saxena

Department of Computer Science, Babasaheb Bhimrao Ambedkar University, Lucknow, Uttar
Pradesh, India

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 195
A. Khamparia et al. (eds.), Microbial Data Intelligence and Computational
Techniques for Sustainable Computing, Microorganisms for Sustainability 47,
https://doi.org/10.1007/978-981-99-9621-6_13
196 M. Dwivedi et al.

Historically, the identification of PDs was predominantly carried out by visual

inspections performed by farmers or farm professionals. Nevertheless, these tech-
niques were plagued by several constraints. The tasks were seen to be significantly
time consuming, requiring a major allocation of workers, and their outcomes were
limited to a rather restricted geographical scope.
Technology plays an essential part in the farming industry, exerting a significant
impact on various agricultural enterprises and emerging farms. The involvement of
technology in agriculture, in recent times, has presented the potential to cultivate
crops in arid desert environments. Automation techniques are highly coveted instru-
ments in the agricultural business. Numerous studies support the claim that the
adoption of automation technologies in agricultural settings has the potential to
enhance crop productivity, leading to an increase in farmers’ annual earnings
(Bagde et al. 2015).
The emergence of PDs is dependent upon a simultaneous presence of three
contributing aspects: the host, the pathogen, and the environmental circumstances
that are favorable for the proliferation of the disease. The term “host” refers to the
particular crop that is being examined for disease detection. On the other hand,
“pathogens” cover the various agents that cause diseases, including fungi, viruses,
and bacteria. Additionally, environmental factors play a crucial role in determining
whether the disease will flourish or decline (Patil and Kumar 2020).
In the event that any of these components are lacking, the disease will not exhibit
any symptoms. PDs originate from two main sources: biotic and abiotic causes.
Biotic factors are derived from the impact of creatures that are alive, such as fungi,
bacteria, viruses, and nematodes. On the other hand, abiotic factors stem from
ecological elements such as temperature, humidity, soil moisture, and the general
ambient conditions.
Lately, the convergence of AI and agriculture has opened up new avenues for
addressing the challenges of PDD. AI, particularly ML and computer vision, has
demonstrated remarkable potential in automating the detection and identification of
PDs. By harnessing the power of AI, farmers and agronomists can not only identify
diseases in their early stages but also make data-driven decisions regarding treat-
ments and resource allocation, thereby reducing the environmental footprint of
agriculture.
The primary aim of this study is to offer a generalized idea to the novice
researchers about PDD with the help of AI technology. We have provided the
steps that are required for PDD. We have even discussed the best suitable existing
AI methods for PDD, depending upon the type of crop along with the accuracy of AI
method. The impact of PD on human life and environment is provided in detail along
with the challenges and open issues of PDD.
The following sections will delve into the methodologies employed in AI-based
PDD, showcase exemplary applications, discuss challenges, and offer insights into
the trajectory of this field. The remaining sections of this chapter are organized as
follows: Section 13.2 will provide a summary of related works, Sect. 13.3 general-
izes the idea of steps involved in plant disease detection, and Sect. 13.4 discusses the
AI methods best suitable for plant type with its accuracies. Sections 13.5, 13.6, and
13 Analyzing the Frontier of AI-Based Plant Disease Detection: Insights. . . 197

13.7 discuss the impact of plant diseases on human life and environment, challenges,
and open issues respectively. At last, the chapter concludes in Sect. 13.8.

13.2 Related Work

Extensive literatures have been conducted on PDD by various researchers depicting

comprehensive analysis and proposing novel methodologies which have left behind
some drawbacks as well. Plant illnesses have been seen as a substantial hazard to
global food security due to the lack of essential infrastructure in several regions
worldwide. Histogram of Oriented Gradients (HOG) and Support Vector Machines
(SVM) are employed for the purpose of predicting the presence of disease in certain
crops.
Gokulnath and Usha Devi (2020) have highlighted a significant issue character-
ized by the fast growth of agricultural goods. However, this growth is hindered by
the negative impact of diseases, pests, and weeds, which are responsible for causing
a decline in productivity levels of these items. ML techniques are employed to
identify the symptoms it produces. One shortcoming of this approach is the restricted
access to just three spectral groups, which consequently imposes limitations on the
investigation of plant state.
Thangavel et al. (2022) presented many methodologies for predicting plant
illnesses and proposed an effective approach for detecting and diagnosing infections
in plants. This approach involved the use of datasets and subsequent training
modules to enable the recognition of PD, achieving a commendable 96.5% accuracy.
A total of 32 distinct plants and illnesses were employed along with convolutional
neural network (CNN) as shown in Table 13.1.

13.3 Steps in the Detection of Plant Diseases

Due to the fact that manual diagnosis of PDs is a time-consuming process that is also
prone to errors, it is most useful in the setting of more modest-sized agricultural
operations. On the other side, one could argue that autonomous diagnosis demon-
strates a greater level of accuracy and efficiency, using less time and labor resources
than traditional methods do. As a direct result of this, a significant amount of
research has been carried out within the same framework. The phases of the PDD
process are illustrated in Fig. 13.1.
Input Image: The input image is provided to the model for detection if it is a
healthy plant or an unhealthy or a diseased one.
Image Preprocessing: The input image consists of noise which makes it inappro-
priate to work with. Preprocessing step is performed for the removal of such noise.
198 M. Dwivedi et al.

Table 13.1 Exhaustive literature survey

Author Method Drawbacks
G. Dhingra (Dhingra Image Processing (IP) Less accurate
et al. 2018)
V. Singh (Singh and Image Segmentation: Algorithm selection is difficult
Misra 2017) Genetic Algorithm
G. Shrestha (Shrestha CNN Accuracy might be improvised
et al. 2020)
R. Sujatha (Sujatha Clustering and Using IoT might be much beneficial
et al. 2021) Classification
J. Liu (Liu and Wang Supervised and Fails to maintain a balance between
2021) Unsupervised Learning accuracy and complexity
K Lin (Lin et al. 2019) VGG16 Comparison is not made properly
S. Zhang (Zhang et al. Clustering Requires clusters of same dimensions
2019)
A. Badage (Badage Edge Detection Problem choosing edge in multi-edge
2018) image
J. Singh (Singh and Regional High computation time and memory
Kaur 2019) needed

Feature Extraction: Feature extraction is carried out for choosing parameters like
shape, texture, color for using them as parameters to provide details and information
of the image. These parameters are fed into the classifier as input features.
Database: The database consists of a variety of plant images that are both healthy
and unhealthy. These images help the classifier train.
Training: The images from the database are trained in the training step for better
and much accurate classification.
Classification: The trained classifier classifies the inputted, preprocessed image as
a healthy or unhealthy one based on the image’s features.

13.4 Disease Detection AI Methods and Type of Crops

The technique known as PDD requires the investigation of a huge number of plant
species that come from a variety of locations all over the world. It is important to
note that India and China are currently serving as the primary locations where a
significant portion of the research endeavors in the subject of PDD are being
undertaken. This endeavor is being carried out on a global scale, and it involves
the cultivation of a broad variety of crops, including soybeans, apples, bananas,
avocados, citrus fruits, coffee, maize, cucumbers, millet, oil palms, rice, potatoes,
berries, wheat, and tomatoes, each of which is subjected to in-depth testing for the
detection of disease.
In the subject of PDD, a great number of in-depth research have been carried out,
the majority of which have concentrated on crops that are typically cultivated. This is
13 Analyzing the Frontier of AI-Based Plant Disease Detection: Insights. . . 199

Fig. 13.1 Plant disease detection steps

Table 13.2 Analysis of AI method depending upon crop

Author Plant Method Accuracy
M. Ji (Ji and Wu 2022) Grape DeepLabV3+ 97.75%
J. Ma (Ma et al. 2019) Cucumber CART 90.67%
G. Yang (Yang et al. 2020) Tomato LFC-Net 99.7%
R. Sreevallabhadev (Sreevallabhadev 2020) Rice CNN + SVM 96.8%
H. Ali (Ali et al. 2017) Citrus fruits Bagging tree 99.9%
W. Haider (Haider et al. 2021) Wheat CNN + DT 97.2%

shown in Table 13.2, where we have analyzed the results obtained for each type of
crop and described the visual methods that were utilized to obtain these conclusions.
For example, rice has been the focal point of several research investigations that
200 M. Dwivedi et al.

employed a wide range of methodologies, such as ML, Deep Learning (DL), and
Image Processing (IP). Chen et al. (2021) utilized the Mobile Network V2 method in
conjunction with the SE block-based attention strategy, resulting in 99.33%
accuracy.
Similarly for cucumber investigation (Ma et al. 2019), Classification and Regres-
sion Trees were adopted, resulting in 90.67% accuracy. A tomato-related research
(Yang et al. 2020) yielded 99.7% accuracy, adopting feedback network and area
suggestion network. The Bagged Tree classifier was utilized in citrus crops (Ali et al.
2017), which provided 99.9% accuracy rate. In the identification of apple diseases,
DenseNet121 model attained an accuracy of 93.71% (Zhong and Zhao 2020).
It is essential to recognize that the quality of the data has a significant influence on
the efficacy of these models, as this is a factor that cannot be ignored. As a
consequence of this, the incorporation of attention processes has been a significant
contributor to the advancement in the diagnosis of diseases affecting rice and grapes.
This demonstrates how important thorough data management and analysis are in the
field of PDD.

13.5 Impact of Plant Diseases on Human Life

and Environment

Diseases have a big effect on a plant’s health because they stop important things
from happening, like absorbing water and nutrients, making food, growing healthy,
and splitting cells. How much damage is done depends on a lot of different things,
such as how dangerous the bacteria are, how strong the host plants are, the weather,
how long the infection has been going on, and a lot more. Because of this, these
diseases can cause a wide range of signs, from mild problems to complete plant
death.
These sneaky pathogens, which are often carried by polluted plant parts, cause a
wide range of problems in plant populations. Several manifestations of the disease
include the development of lesions on both fruits and leaves, rotting of the roots and
fruits, occurrence of leaf blights, wilting, and ultimately, the demise of the plant.
Notably, the effects of plant illnesses are not limited to the world of plants. Take
the mild mottle virus, which is known to affect the immune system of humans and
cause clinical signs. Also, some PDs can reduce the amount of food available or add
harmful chemicals to the food chain, which can be bad for people’s health. Also,
putting bacteria in the soil that fight PDs, like herbicides and insecticides, which are
meant to kill bugs and diseases, can accidentally hurt people.
Due to the fact that different plants are more likely to have different bugs and
diseases, traditional farming methods often use a lot of pesticides and herbicides,
which unfortunately add to long-term pollution of the land. Even though most plant
pathogens can’t make people sick, it’s best not to eat fruits and veggies that contain
mold or are rotten from farms where herbicides are used. Also, you need to be very
13 Analyzing the Frontier of AI-Based Plant Disease Detection: Insights. . . 201

careful with food that has been affected by fungus that make toxins. Both of these
types of food pose serious health risks.
Taking out the sick parts of veggies could reduce the number of disease-causing
agents and rotting food. But it’s important to know that this method doesn’t always
get rid of all possible sources of pollution. Some fungi and poisons like to grow and
spread, so it’s important to deal with them all at once, especially in places where
there are a lot of farms with natural plant foods.

13.6 Challenges

PDD has a number of difficult obstacles, each of which calls for innovative solu-
tions. Identification of illnesses in the field in real time presents a significant
obstacle. It is vital for there to be a rapid and accurate diagnosis in the field in
order to facilitate early action and treatment of sickness. Another challenge is the
impact of climate change. As the state of our environment worsens, diseases that
were once confined to specific regions may begin to spread. It is essential to modify
diagnostic procedures in response to shifting illness patterns. It can be challenging to
preserve analytical images of a consistent and good quality. In addition, the limited
availability of various datasets for the training of AI models is a factor that delays
progress. Large and varied datasets are necessary for the development of robust and
effective algorithms for disease detection. Both scalability and computational com-
plexity provide significant challenges. In order to diagnose diseases in real time, AI
systems need to be able to effectively process huge amounts of data. Environmental
factors also contribute to a reduction in accuracy. In order to avoid producing false
positives and negatives as a result of illumination, humidity, and other conditions,
more robust AI models are required. The use of Unmanned Aerial Vehicles (UAVs)
and the İnternet of Things (IoT) for the diagnosis of illnesses has both potential and
limitations. Integration of these technologies into agricultural practices will require
extensive planning as well as the creation of supporting infrastructure. In conclusion,
the identification of diseases by crowdsourcing and public engagement is an inno-
vative approach; nonetheless, it is fraught with difficulties in terms of data quality,
privacy, and scalability. These challenges bring into focus the essential role that
teamwork and creativity play in PD control.

13.7 Open Issues

The potential breakthroughs in the field of PD identification through the utilization

of AI are highly promising and set for substantial progress. One of the primary areas
of emphasis lies in the ongoing improvement of AI algorithms in order to augment
the precision and dependability of illness detection systems. The attainment of this
objective necessitates the optimization of algorithms and the augmentation of varied
202 M. Dwivedi et al.

and comprehensive datasets, hence guaranteeing the accurate identification of

diseases.
An additional promising area of exploration involves the implementation of real-
time monitoring techniques. In this context, AI-driven systems that are equipped
with various sensors, drones, and cameras are employed to actively and attentively
examine agricultural crops. This approach facilitates the early detection of diseases,
hence enabling prompt intervention measures to be taken. These technologies have
the capacity to significantly transform the field of agriculture through their ability to
proactively mitigate the occurrence of disease outbreaks.
Moreover, the incorporation of many sensing modalities, including hyperspectral
imaging, thermal imaging, and chemical analysis, into AI systems is imminent. The
utilization of a multi-modal approach holds the potential to enhance disease detec-
tion by including a wider range of factors and providing more precise results, hence
addressing the specific requirements of different crops and settings.
Within the domain of edge computing, AI algorithms are being deployed on edge
devices, including IoT sensors and UAVs, in order to minimize latency and decrease
reliance on uninterrupted internet connectivity. This advancement enhances the
feasibility of AI-driven disease detection, particularly in underserved regions char-
acterized by restricted availability of high-speed internet connectivity.
The idea of automating therapy is a promising development, wherein AI systems
not only detect problems but also provide recommendations and occasionally inde-
pendently administer therapies, such as precise application of pesticides or herbi-
cides. The use of this approach has the potential to result in a substantial decrease in
the amount of labor and resources needed for the management of diseases.
The possibility for early disease forecasting in the future is anticipated, wherein
AI will be utilized to predict disease outbreaks. This prediction will be based on
historical data, weather patterns, and environmental factors. Consequently, farmers
will be empowered to proactively protect their crops.
The significance of data integration and sharing should be underscored as the
domain of PDD employing AI progresses. The utilization of platforms that enable
the efficient sharing of PD data among researchers, farmers, and agricultural orga-
nizations has the potential to enhance the collective knowledge and disease man-
agement endeavors.
Additional avenues encompass the amalgamation of AI with robotics to facilitate
autonomous farming, the augmentation of ML explainability to foster user trust, the
use of blockchain technology for the purpose of agricultural traceability, and the
promotion of international cooperation to address global PDs. In addition, educa-
tional and training programs strive to provide farmers and agricultural professionals
with the necessary knowledge and resources to proficiently utilize AI systems for the
purpose of detecting and managing diseases. Furthermore, it is imperative to address
ethical issues, environmental implications, and responsible deployment in order to
guarantee the sustainable and ethical integration of AI in the field of agriculture.
The field of PDD with AI is characterized by its dynamic nature and quick
evolution. It has significant promise for transforming the agricultural sector, as it
has the capacity to enhance crop productivity, mitigate environmental consequences,
13 Analyzing the Frontier of AI-Based Plant Disease Detection: Insights. . . 203

and safeguard global food supplies. Nevertheless, the realization of this transforma-
tional potential necessitates collaborative endeavors from researchers, policymakers,
and agricultural stakeholders in order to surmount obstacles.

13.8 Conclusion

PDD is a substantial barrier to overcome for those working in the agricultural sector.
As a result, it is of the utmost importance to detect plant diseases as quickly as
possible in order to lessen the impact of these problems and limit their spread within
agricultural settings. Lately, there has been a substantial number of studies that has
been carried out on a variety of approaches to PD detection, and many of these
studies have made important and ground-breaking discoveries. This research was
conducted with the intention of introducing various AI technologies and approaches
to aid farmers in achieving higher yields. In addition to this, it places an emphasis on
PDD methodologies throughout numerous literatures. In this study chapter, the
important connection between AI and PDD is investigated, and insights into its
significance, methodology, challenges, and potential future routes of investigation
are provided. With the help of this work, we have endeavored to present a funda-
mental concept for utilizing AI in the detection of PDs. This chapter details the
processes involved in PDD as well as the AI methods that are utilized for PDD. A
crucial component of this presentation is a discussion of the difficulties and oppor-
tunities that lie ahead for PDD. The impact of PDs on human life and the environ-
ment is also explored, which is something that is quite helpful for beginning
researchers.

References

Ali H, Lali MI, Nawaz MZ, Sharif M, Saleem BA (2017) Symptom based automated detection of
citrus diseases using color histogram and textural descriptors. Comput Electron Agric 138:92–
104
Badage A (2018) Crop disease detection using machine learning: Indian agriculture. Int Res J Eng
Technol 5(9):866–869
Bagde S, Patil S, Patil S, Patil P (2015) Artificial neural network based plant leaf disease detection.
Int J Comput Sci Mobile Comput 4:900–905
Chen J, Chen J, Zhang D, Nanehkaran YA, Sun Y (2021) A cognitive vision method for the
detection of plant disease images. Mach Vis Appl 32:1–18
Dhingra G, Kumar V, Joshi HD (2018) Study of digital image processing techniques for leaf disease
detection and classification. Multimed Tools Appl 77:19951–20000
Gokulnath BV, Usha Devi G (2020) A survey on plant disease prediction using machine learning
and deep learning techniques. Intel Artif 23:136–154
Haider W, Rehman AU, Durrani NM, Rehman SU (2021) A generic approach for wheat disease
classification and verification using expert opinion for knowledge-based decisions. IEEE Access
9:31104–31129
204 M. Dwivedi et al.

Ji M, Wu Z (2022) Automatic detection and severity analysis of grape black measles disease based
on deep learning and fuzzy logic. Comput Electron Agric 193:106718
Lin K, Gong L, Huang Y, Liu C, Pan J (2019) Deep learning-based segmentation and quantification
of cucumber powdery mildew using convolutional neural network. Front Plant Sci 10:422622
Liu J, Wang X (2021) Plant diseases and pests detection based on deep learning: a review. Plant
Methods 17:1–18
Ma J, Du K, Zheng F, Zhang L, Sun Z (2019) A segmentation method for processing greenhouse
vegetable foliar disease symptom images. Inform Process Agric 6:216–223
Patil RR, Kumar S (2020) A bibliometric survey on the diagnosis of plant leaf diseases using
artificial intelligence. Libr Philos Pract 2020:3987
Shrestha G, Deepsikha, Das M, Dey N (2020) Plant disease detection using CNN. In: Proceedings
of 2020 IEEE applied signal processing conference, ASPCON. IEEE, pp 109–113
Singh J, Kaur H (2019) Plant disease detection based on region-based segmentation and KNN
classifier. In: Proceedings of the international conference on ISMAC in computational vision
and bio-engineering 2018 (ISMAC-CVB), vol 30. Springer International Publishing, pp
1667–1675
Singh V, Misra AK (2017) Detection of plant leaf diseases using image segmentation and soft
computing techniques. Inform Process Agric 4:41–49
Sreevallabhadev R (2020) An improved machine learning algorithm for predicting blast disease in
paddy crop. Mater Today Proc 33:682–686
Sujatha R, Chatterjee JM, Jhanjhi NZ, Brohi SN (2021) Performance of deep learning vs machine
learning in plant leaf disease detection. Microprocess Microsyst 80:103615
Thangavel M, Gayathri PK, Sabari KR (2022) Plant leaf disease detection using deep learning. Int J
Eng Res Technol 10:3599–3605
Yang G et al (2020) Self-supervised collaborative multi-network for fine-grained visual categori-
zation of tomato diseases. IEEE Access 8:211912–211923
Zhang S, You Z, Wu X (2019) Plant disease leaf image segmentation based on superpixel clustering
and EM algorithm. Neural Comput & Applic 31:1225–1232
Zhong Y, Zhao M (2020) Research on deep learning in apple leaf disease recognition. Comput
Electron Agric 168:105146
Chapter 14
Fuzzy and Data Mining Methods
for Enhancing Plant Productivity
and Sustainability

Khalil Ahmed, Mithilesh Kumar Dubey, Devendra Kumar Pandey,

and Sartaj Singh

Abstract The agriculture sector plays a pivotal role in addressing global food
security challenges while also grappling with the imperative of sustainability. In
recent years, the use of data mining and fuzzy logic together has emerged as a potent
toolset for optimizing plant productivity and fostering sustainable agricultural prac-
tices. This chapter explores the intersection of Fuzzified reasoning and Knowledge
discovery in data in the context of plant productivity and sustainability. We delve
into the principles, methodologies, and real-world applications of these technologies
to empower agricultural stakeholders with the knowledge and tools needed to
advance sustainable farming practices.

Keywords Sustainable agriculture · Fuzzy logic · Data mining · Fuzzy rule

14.1 Introduction

The need for food and agricultural goods is increasing as the world’s population
escalates. Simultaneously, environmental concerns, resource constraints, and cli-
mate change impose unprecedented challenges on the agriculture sector (Deshpande
2017). To address these challenges, it becomes essential to optimize plant produc-
tivity while minimizing resource use and environmental impact. Fuzzy logic and
data mining are two prominent computational techniques that, when combined, offer
innovative solutions to enhance plant productivity and sustainability (Gharde et al.
2018).
Agriculture is at the nexus of global challenges in the twenty-first century,
confronted with the formidable task of simultaneously feeding an ever-expanding
global population while mitigating the profound environmental and resource sus-
tainability challenges it faces. There are still people in the globe to burgeon, the

K. Ahmed · M. K. Dubey (✉) · D. K. Pandey · S. Singh

School of Computer Application, Lovely Professional University, Phagwara, Punjab, India
e-mail: [email protected]; [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 205
A. Khamparia et al. (eds.), Microbial Data Intelligence and Computational
Techniques for Sustainable Computing, Microorganisms for Sustainability 47,
https://doi.org/10.1007/978-981-99-9621-6_14
206 K. Ahmed et al.

demand for food, Fiber, and bioenergy intensifies, putting immense pressure on
agricultural systems (Antia 2023). This necessitates a fundamental transformation in
the way we approach agriculture—a transformation that not only amplifies plant
productivity but also safeguards the planet’s delicate ecological balance.
Central to this agricultural revolution is the integration of cutting-edge technologies,
including fuzzy logic and data mining, which have emerged as a powerful synergistic
duo for optimizing plant productivity while championing sustainability (Misra 2014).

14.2 The Imperative of Sustainable Agriculture

Sustainable agriculture stands as a pivotal pillar of our future as shown in Fig. 14.1. It
encapsulates a holistic approach that seeks to meet the burgeoning demand for
agricultural products while minimizing the ecological footprint (Sharma et al. 2022a).
Key concerns in sustainable agriculture encompass the judicious use of resources
such as water and soil emissions of carbon dioxide are being reduced, the conser-
vation of biodiversity, and the promotion of resilient farming systems that can
withstand climate variability (Kumar et al. 2018). However, achieving these multi-
faceted goals is a Herculean task. It requires not only technological innovation but
also a deep understanding of the complex, interrelated factors that affect plant
growth and overall farm performance. This is where fuzzy logic and data mining
step into the limelight, offering advanced computational techniques to unravel the
intricacies of agricultural systems as shown in Table 14.1.

Fig. 14.1 Normative imperatives model for sustainable development strategy

14 Fuzzy and Data Mining Methods for Enhancing Plant Productivity. . . 207

Table 14.1 Agriculture that is both customary and sustainable

Customary Sustainable
Growing regions and using creativity and science to Procedures for least impedance with
control nature attributes
Extreme soil breakdown and technological No-till or definitely less culturing
cultivation (organic culturing)
Windy conditions and deteriorating soil Slow winds and eroded soil
Low water intrusion Water invades at a rapid rate

Table 14.2 Primary difficulties facing sustainable agriculture

Parameters Obstacles
Areas Ways to stop soil erosion, stop ecosystem dissolution, and halt habitat loss
without lowering the profitability of crops
Management Efficiently use H2O
Low inputs Efficiently use nutrients
Pest control
Human Increase yield productivity
inference Reduce the negative impact on human health
Environment Lessen the effect of environmental degradation
Catastrophe in habitat

14.2.1 Fuzzy Logic

Taming Agricultural Uncertainty Fuzzy logic, initially formulated by Lotfi Zadeh in

the 1960s, is a mathematical framework that excels in handling uncertainty, impre-
cision, and vagueness—an inherent feature of agricultural data. In agriculture,
concepts like “optimal soil moisture,” “ideal temperature range,” or “high pest
susceptibility” often elude precise numerical definitions. Fuzzy logic introduces
the notion of fuzzy sets, which enable the representation of such ambiguous concepts
using membership functions (Kumar et al. 2018). This enables agricultural stake-
holders to work with imprecise data in a structured manner.

14.2.2 Data Mining: Unearthing Insights from

Agricultural Data

In tandem with fuzzy logic, data mining techniques play an instrumental role in
modern agriculture. The agriculture sector generates a staggering volume of data,
encompassing information on climate, soil properties, crop characteristics, and more.
Data mining techniques, such as classification, clustering, and regression, serve as a
treasure trove of tools for extracting valuable knowledge from this data deluge
(Naseem et al. 2022). They enable researchers and practitioners to identify patterns,
make predictions, and optimize agricultural processes as shown in Table 14.2.
208 K. Ahmed et al.

Data mining techniques: Data mining techniques in agriculture involve the

application of various methods and algorithms to extract valuable insights and
patterns from large agricultural datasets. These techniques are crucial for making
informed decisions, optimizing resource allocation, and improving crop yields in the
farming sector. Some key data mining techniques in agriculture.

14.2.3 The Synergy: Fuzzy Logic and Data Mining.

However, the true power of these technologies comes to the fore when they are
integrated. The synergy between fuzzy logic and data mining forms a formidable
partnership that empowers agricultural stakeholders (Mabhaudhi et al. 2019). Fuzzy
logic allows for the representation of im-precise knowledge, while data mining
techniques help unearth hidden patterns and relationships in vast datasets (Klir and
Yuan 1995). Together, they pave the way for intelligent decision support systems
that can enhance crop yields, reduce resource consumption, and promote
sustainability.

14.2.4 Fuzzy Logic: A Foundation for Uncertainty Handling.

Fuzzy logic provides a mathematical framework for handling uncertainty, impreci-

sion, and vagueness in agricultural data as shown in Table 14.3 (Raorane and
Kulkarni 2012). This section introduces the fundamentals of fuzzy logic, highlight-
ing its key components:
Fuzzy sets: Fuzzy sets allow the representation of vague concepts, such as “high
temperature” or “adequate soil moisture,” using membership functions (Majumdar
et al. 2017).
Fuzzy rules: Fuzzy rules translate expert knowledge into a formal structure that
guides decision-making processes (Doğan et al. 2015).
Inference systems: Fuzzy inference systems process fuzzy inputs, apply fuzzy
rules, and generate crisp or fuzzy outputs (Naseem et al. n.d.). Fuzzy rules in
agriculture, when combined with technology and data mining, offer a powerful

Table 14.3 Fuzzy logic techniques in the agriculture domain

SDM
domain Usage Technique/method
Agriculture Precision agriculture Cross-validation techniques
Crop yield prediction ML, DL, and artificial intelligence
Crop disease detection and ML, DL, IoT, and Artificial
classification intelligence
Recommendation system for crop CNN, ANN
14 Fuzzy and Data Mining Methods for Enhancing Plant Productivity. . . 209

approach to decision-making and optimization in farming practices. Fuzzy logic is a

mathematical framework that deals with uncertainty and imprecision, making it well
suited for addressing the complex and uncertain nature of agricultural systems.
When integrated with technology and data mining, it enables farmers and agricul-
tural experts to make more informed and adaptive decisions. Let’s explore this
concept in detail: Fuzzy logic extends traditional binary (true/false) logic by
allowing for degrees of truth between 0 and 1, representing the level of membership
or certainty. Fuzzy sets describe the degree of membership of an element to a set.
Fuzzy rules consist of “if-then” statements that use fuzzy linguistic variables to
express relationships. Fuzzy rules allow for adaptive decision-making, considering
the uncertainty and variability in agricultural systems. Fuzzy logic provides trans-
parent and interpretable rules that farmers and experts can understand and trust. By
optimizing resource usage (e.g., water, pesticides, fertilizers), fuzzy logic can
improve crop yields while minimizing environmental impact. Fuzzy logic helps
mitigate risks associated with unpredictable factors like weather and pests, leading
to more reliable outcomes. Fuzzy rules in agriculture, when integrated with technol-
ogy and data mining, enable data-driven, adaptive, and efficient decision-making
(Melin and Castillo 2014). This approach leverages the power of fuzzy logic to
handle uncertainty and imprecision inherent in agricultural systems, ultimately
improving crop yields, resource management, and sustainability in farming
practices.

14.3 Data Mining Techniques for Agricultural Insights

Data mining techniques extract knowledge from vast datasets, enabling data-driven
decision-making in agriculture as shown in Fig. 14.2 (RURAL DO 2014). This
section presents an overview of data mining techniques applicable to plant produc-
tivity and sustainability.
Classification: Classify crops based on yield potential, disease susceptibility, or
other factors.

Fig. 14.2 Data mining

techniques
210 K. Ahmed et al.

14.3.1 Classification and Clustering

Group similar crops or regions to identify patterns and trends (Jamshidi et al. 2016).
Classification algorithms are employed to categorize data into distinct groups or
classes. In agriculture, this can be highly valuable: Crop Disease Identification.
Using historical data on crop health, machine learning classification models can
identify diseases or pests affecting crops and suggest appropriate interventions. Soil
Type Classification: classifying soil types based on their characteristics (e.g., texture,
pH levels) helps farmers make informed decisions about crop selection and irrigation
methods.

14.3.2 Regression

Regression analysis is a statistical way to understand the relationships between

variables. In agriculture, it can be applied to predict outcomes based on various
factors. Predict crop yields or resource requirements based on historical data (Pantazi
et al. 2019). By analyzing historical data on crop yields and factors like weather
conditions, soil quality, and fertilizer usage, regression models can predict future
crop yields. This helps farmers plan their harvest and resources effectively. Regres-
sion can be used to predict the growth rate of livestock based on factors like diet,
environment, and genetics.
Time series analysis: Studies of data series are used to understand data collected
over time, making it crucial for agriculture. Weather Forecasting: analyzing histor-
ical weather data through time series analysis helps in forecasting future weather
patterns, which is vital for crop management. Crop Growth Monitoring: continuous
monitoring of crop growth over time can help detect anomalies or trends that may
require action, such as adjusting irrigation schedules.
Association rule mining: Discover relationships between variables, such as envi-
ronmental factors and crop performance.

14.4 Integration of Fuzzy Logic and Data Mining

The synergy between fuzzy logic and data mining offers a potent approach to
enhancing agricultural sustainability:
Fuzzy clustering: Utilize fuzzy clustering to group crops based on imprecise
characteristics, facilitating targeted interventions.
Fuzzy classification: Combine fuzzy logic with classification algorithms to
account for uncertainty in crop categorization.
Fuzzy association rules: Develop fuzzy association rules to uncover nuanced
relationships between agricultural variables (Zhai et al. 2020).
14 Fuzzy and Data Mining Methods for Enhancing Plant Productivity. . . 211

14.5 Case Studies

This section presents real-world case studies that demonstrate the practical applica-
tion of fuzzy logic and data mining in agriculture.
Precision agriculture: Explain how to optimize resource allocation in precision
agriculture, reducing input waste and environmental impact as shown in Fig. 14.3.
Crop Disease Management: Investigate the use of fuzzy classification and data
mining to predict and manage crop diseases, minimizing yield loss (Gavioli et al.
2019).
This form of leaf staging, like the leaf collar approach, starts with a short initial
leaf. Leaf counting then changes, concluding with the leaf that is at least 40–50%
exposed from the whorl, rather than the highest leaf with a visible collar as shown in
Fig. 14.4. The tip of this “indicator” leaf generally “droops” or hangs down in knee-
high corn or older, thus I refer to this as the “droopy” leaf approach (Arumugam et al.
2022).
Climate adaptation: Examine the integration of fuzzy logic and data mining in
adapting agricultural practices to changing climate conditions.

Fig. 14.3 Precision

agriculture life cycle

Fig. 14.4 Cereal-yield-vs.-extreme-poverty-scatter due to proper crop management in India

(1961–2021) according to Our World Data report
212 K. Ahmed et al.

14.6 Challenges and Future Directions

While the integration of fuzzy logic and data mining shows great promise, it also
presents challenges such as data quality, model interpretability, and scalability
(Arumugam et al. 2022). This section discusses these challenges and outlines
potential future directions for research and application in the field of agricultural
sustainability.

14.6.1 Challenges

Data Quality and Availability: One of the foremost challenges in applying data
mining and fuzzy logic to agriculture is the quality and availability of data. In many
regions, data on soil properties, weather, and crop performance may be limited or
inconsistent. Ensuring data quality and accessibility remains a critical hurdle
(Narmadha et al. 2022).
Interpretability: Complex data mining models, such as deep learning algorithms,
often lack interpretability. For stakeholders in agriculture, understanding the ratio-
nale behind a decision or prediction is crucial. Balancing model complexity with
interpretability is an ongoing challenge.
Scale and Scalability: Agriculture spans a wide range of scales, from small family
farms to large agribusinesses. Scaling data mining and fuzzy logic methods to
accommodate this diversity is challenging (Sharma et al. 2022b). Methods that
work for one scale may not be directly applicable to another.
Integration with Expert Knowledge: Fuzzy logic relies on expert knowledge to
define fuzzy sets and rules. Integrating this expert knowledge seamlessly into data
mining models can be complex. Ensuring that the models respect domain expertise is
an ongoing challenge.
Resource Constraints: Sustainable agriculture often requires the optimization of
resource use, such as water, fertilizers, and pesticides. Agricultural land is finite, and
in many regions, it’s under pressure due to urbanization and degradation. As the
global population grows, the need for agricultural land increases, making efficient
land use critical. Access to water is a significant constraint, particularly in arid and
semi-arid regions. Efficient irrigation practices are essential to conserve water
resources. Labor shortages are common in agriculture, especially during peak
seasons. Finding and retaining skilled labor can be challenging. Farmers often face
financial constraints, limiting their ability to invest in modern machinery and
technologies. Access to agricultural knowledge and best practices can be limited in
some regions, hindering productivity. Developing models that account for these
constraints while maximizing productivity poses a challenge, especially when con-
sidering dynamic environmental conditions. Technology such as GPS, sensors, and
drones enables precision agriculture. Farmers can target resources like water, fertil-
izer, and pesticides precisely where they are needed, minimizing waste. Smart
14 Fuzzy and Data Mining Methods for Enhancing Plant Productivity. . . 213

irrigation systems monitor soil moisture levels and weather conditions to optimize
water usage, reducing water waste and energy costs. Farm machinery, like tractors
and harvesters, reduces labor demands, making farming more efficient. Advances in
robotics also offer potential solutions to labor shortages. Satellite and drone imagery
provide real-time data on crop health, enabling early detection of issues like pests or
diseases. This helps in timely interventions, reducing crop losses. Genetically
modified crops can be engineered to be more resilient to boosting yields and
lowering the demand for chemical pesticides, illnesses, and surrounding inputs.

14.7 Role of Data Mining in Addressing Resource

Constraints

Data mining techniques can be applied to various agricultural datasets to extract

valuable insights such as. Historical weather data can be analyzed to predict future
weather patterns, helping farmers plan planting and harvesting times, and optimize
resource allocation as shown in Table 14.4.
Soil data, including texture, pH levels, and nutrient content, can be analyzed to
determine optimal crop choices and fertilizer requirements. Analyzing historical
crop yield data in conjunction with factors like weather, soil, and farming practices
can provide insights into yield optimization. Data mining can identify patterns in
pest and disease outbreaks, allowing for early intervention and reduced chemical
use. Analyzing market trends and price data can assist farmers in making choices in
crop selection and when to sell their produce for maximum profit. Data mining can
optimize the supply chain, reducing post-harvest losses and ensuring that crops reach
the market efficiently. Machine learning models can predict resource requirements,

Table 14.4 Data analysis tasks and techniques

Data analysis Data Dependency
techniques summarization Segmentation Classification Prediction analysis
Descriptive and ✓ ✓ ✓
visualization
Correlation ✓
analysis
Cluster analysis ✓ ✓
Discriminant ✓ ✓
analysis
Regression ✓ ✓
analysis
Case-base ✓
reasoning
Decision tree ✓ ✓
Association rule ✓
214 K. Ahmed et al.

such as water and fertilizer, based on historical data, helping farmers plan and budget
more effectively.

14.8 Challenges and Considerations

The accuracy and quality of data are critical for meaningful analysis. Data collected
from various sources may require cleaning and validation. Not all farmers have
access to the latest technology, and investments can be costly. Addressing the digital
divide is crucial. Farmers must be cautious about sharing sensitive data. Proper data
security measures must be in place. Farmers and agricultural professionals need
training to effectively use technology and interpret data-driven insights. Technology
and data mining are powerful tools for addressing resource constraints in agriculture.
They enable farmers to make more informed decisions, optimize resource usage, and
increase overall productivity while considering environmental sustainability. How-
ever, it’s essential to ensure that these tools are accessible to all farmers and that data
privacy and security concerns are addressed.

14.9 Future Directions

Enhanced Data Collection and Integration: Future research should focus on improv-
ing data collection techniques, including the use of remote sensing, IoT devices, and
drones, to gather high-resolution and real-time data. Additionally, efforts to integrate
diverse data sources seamlessly will be crucial for comprehensive analysis.
Explainable AI in agriculture: The development of explainable AI models that
combine the strengths of data mining and fuzzy logic can bridge the interpretability
gap. This will make it easier for agricultural stakeholders to trust and act upon model
recommendations.
Machine learning for precision agriculture: Continued research into machine
learning techniques tailored to precision agriculture can lead to more efficient
resource management. Algorithms that adapt to changing conditions and provide
real-time recommendations will be invaluable.
Big data and cloud computing: Harnessing the power of big data analytics and
cloud computing can address scalability issues. Cloud-based platforms can provide
accessible and scalable solutions for agricultural stakeholders, regardless of their
scale of operation.
Robotic and autonomous farming: Integrating fuzzy logic and data mining into
robotic and autonomous farming systems can revolutionize agriculture. These tech-
nologies can enable automated decision-making and precision application of
resources.
14 Fuzzy and Data Mining Methods for Enhancing Plant Productivity. . . 215

Climate resilience: Developing models that help agriculture adapt to changing

climate conditions will be critical. Predictive models that consider long-term climate
trends and their impact on crop yields can aid in climate-resilient farming practices.
Decision support systems: Building user-friendly decision support systems that
incorporate fuzzy logic and data mining can empower farmers and policymakers to
make informed choices. These systems should consider local conditions and con-
straints. The challenges and future directions in applying fuzzy logic and data mining
methods to enhance plant productivity and sustainability are intertwined. Overcom-
ing challenges while exploring these future directions will require interdisciplinary
collaboration among researchers, farmers, policymakers, and technology devel-
opers. By addressing these challenges and embracing innovative directions, we
can propel agriculture toward a more sustainable and productive future.

14.10 Conclusion

The fusion of fuzzy logic and data mining holds tremendous potential for enhancing
plant productivity and promoting sustainable agricultural practices. By harnessing
the power of these computational techniques, agriculture can become more resilient,
efficient, and environmentally friendly. This chapter has provided a comprehensive
overview of the principles, methodologies, and real-world applications of fuzzy
logic and data mining in agriculture, paving the way for continued innovation in
this critical domain. In this chapter, we embark on a comprehensive exploration of
the fusion of fuzzy logic and data mining in the context of plant productivity and
sustainability. We delve into the underlying principles, methodologies, and real-
world applications of these technologies, providing a roadmap for agricultural
stakeholders to harness the potential of these cutting-edge tools. Through this
synergy, agriculture can not only meet the demands of a growing world but also
do so in a way that respects the delicate balance of our planet’s ecosystems, ensuring
a sustainable future for generations to come.

References

Antia DD (2023) Desalination of saline irrigation water using hydrophobic, metal–polymer hydro-
gels. Sustainability 15(9):7063
Arumugam K, Swathi Y, Sanchez DT, Mustafa M, Phoemchalard C, Phasinam K, Okoronkwo E
(2022) Towards applicability of machine learning techniques in agriculture and energy sector.
Mater Today: Proc 51:2260–2263
Deshpande T (2017) State of agriculture in India. PRS Legisl Res 53(8):6–7
Doğan O, Aşan H, Ayç E (2015) Use of data mining techniques in advance decision making
processes in a local firm. Eur J Business Econ 10(2):6821. https://doi.org/10.12955/ejbe.
v10i2.682
216 K. Ahmed et al.

Gavioli A, de Souza EG, Bazzi CL, Schenatto K, Betzek NM (2019) Identification of management
zones in precision agriculture: an evaluation of alternative cluster analysis methods. Biosyst Eng
181:86–102
Gharde Y, Singh PK, Dubey RP, Gupta PK (2018) Assessment of yield and economic losses in
agriculture due to weeds in India. Crop Prot 107:12–18
Jamshidi P, Sharifloo A, Pahl C, Arabnejad H, Metzger A, Estrada G (2016) Fuzzy self-learning
controllers for elasticity management in dynamic cloud architectures. In: 2016 12th Interna-
tional ACM SIGSOFT conference on quality of software architectures (QoSA). IEEE, pp 70–79
Klir G, Yuan B (1995) Fuzzy sets and fuzzy logic, vol 4. Prentice Hall, New Jersey, pp 1–12
Kumar V, Jat HS, Sharma PC, Gathala MK, Malik RK, Kamboj BR, Yadav AK, Ladha JK,
Raman A, Sharma DK, McDonald A (2018) Can productivity and profitability be enhanced
in intensively managed cereal systems while reducing the environmental footprint of produc-
tion? Assessing sustainable intensification options in the breadbasket of India. Agric Ecosyst
Environ 252:132–147
Mabhaudhi T, Chimonyo VGP, Hlahla S, Massawe F, Mayes S, Nhamo L, Modi AT (2019)
Prospects of or-phan crops in climate change. Planta 250:695–708
Majumdar J, Naraseeyappa S, Ankalaki S (2017) Analysis of agriculture data using data mining
techniques: application of big data. J Big Data 4(1):20
Melin P, Castillo O (2014) A review on type-2 fuzzy logic applications in clustering, classification
and pattern recognition. Appl Soft Comput 21:568–577
Misra AK (2014) Climate change and challenges of water and food security. Int J Sustain Built
Environ 3(1):153–165
Narmadha R, Latchoumi TP, Jayanthiladevi A, Yookesh TL, Mary SP (2022) A fuzzy-based
framework for an agriculture recommender system using membership function. Appl Soft
Comput: Tech Appl:207–223
Naseem M, Singh V, Ahmed K, Mahroof M, Ahamad G, Abbasi E (2022) Architecture of automatic
irrigation system in Hilly area using wireless sensor net-work: a review. In: 2022 2nd Interna-
tional conference on emerging frontiers in electrical and electronic technologies (ICEFEET).
IEEE, pp 1–6
Naseem M, Alam M, Ahmad K, Singh V, Mahroof M, Ahamad G (n.d.) Machine learning
approaches for automatic irrigation system in Hilly areas using wireless sensor networks
Pantazi XE, Moshou D, Bochtis D (2019) Intelligent data mining and fusion systems in agriculture.
Academic Press
Raorane AA, Kulkarni RV (2012) Data mining: an effective tool for yield estimation in the
agricultural sector. Int J Emerg Trends Technol Comput Sci 1(2):1–4
RURAL DO (2014) IRLA2014
Sharma V, Tripathi AK, Mittal H (2022a) Technological revolutions in smart farming: current
trends, challenges and future directions. Comput Electron Agric:107217
Sharma A, Abrahamian P, Carvalho R, Choudhary M, Paret ML, Vallad GE, Jones JB (2022b)
Future of bacterial disease management in crop production. Annu Rev Phytopathol 60:259–282
Zhai Z, Martínez JF, Beltran V, Martínez NL (2020) Decision support systems for agriculture 4.0:
survey and challenges. Comput Electron Agric 170:105256
Chapter 15
Plant Disease Diagnosis with Artificial
Intelligence (AI)

Muhammad Naveed , Muhammad Majeed, Khizra Jabeen, Nimra Hanif,

Rida Naveed, Sania Saleem, and Nida Khan

Abstract Plant diseases, which are unseen but deadly, endanger our crops and the
food security of nations. However, optimism stems from the convergence of pow-
erful artificial intelligence (AI) approaches, each of which plays a distinct role in the
protection of our domains. A change in the way farming is practiced at the moment
could be embodied by an unwavering application of artificial intelligence and its
subsets in agriculture. A farmer may accomplish more with fewer resources thanks
to AI-powered farming solutions, which also improve quality and ensure speedy
GTM (go-to-market) strategies for crops. Direct use of AI (artificial intelligence) or
machine intelligence in the agricultural industry could represent a paradigm shift in
the way farming is now carried out. Deep learning, driven by neural networks, has
transformed how we perceive and diagnose many diseases. Deep learning over-
comes the constraints of existing approaches by autonomously extracting detailed
visual information, providing greater precision and efficiency in recognizing plant
diseases. Convolutional neural networks (CNNs), a subset of deep learning, have
emerged as powerful tools, with elaborate network structures and local receptive
fields that enable them to interpret complex visual input, making them indispensable
in the field of image recognition. Machine learning approaches, such as support
vector machine (SVM) and artificial neural network (ANN) classifiers, have also

M. Naveed · K. Jabeen · N. Hanif · R. Naveed

Department of Biotechnology, Faculty of Science and Technology, University of Central
Punjab, Lahore, Punjab, Pakistan
e-mail: [email protected]
M. Majeed (✉)
Department of Botany, University of Gujrat, Gujrat, Pakistan
e-mail: [email protected]
S. Saleem
Department of Plant Sciences, Quaid-i-Azam University, Islamabad, Pakistan
e-mail: [email protected]
N. Khan
Department of Botany, University of Science and Technology Bannu, Bannu, Khyber
Pakhtunkhwa, Pakistan

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 217
A. Khamparia et al. (eds.), Microbial Data Intelligence and Computational
Techniques for Sustainable Computing, Microorganisms for Sustainability 47,
https://doi.org/10.1007/978-981-99-9621-6_15
218 M. Naveed et al.

stepped up to the plate, automating the diagnosis of plant diseases with remarkable
precision. Deep learning-capable robotics and machine intelligence have had a
profoundly disruptive and enabling impact on industry, governments, and society.
They are also having an impact on more general trends in international sustainability.
Weather patterns, soil composition, and disease trends all tell their own tales,
providing forecast insights and personalized preventative steps to safeguard the
harvest. A symphony of IoT devices orchestrates vigilance across smart farms. By
capturing the afflicted plant sections, farmers may quickly and correctly identify
illnesses and find remedies using a mobile app through AI advancements. The most
recent artificial intelligence (AI) algorithms for cloud-based image processing enable
real-time diagnosis. Artificial intelligence and its thorough learning capabilities have
developed into a crucial strategy for addressing a range of farming-related
difficulties.

Keywords Artificial intelligence · Plant diseases · Remarkable precision

15.1 Introduction

Agriculture has always been considered the most essential need of livelihoods
throughout the ecosystem. It plays a vital role in supporting every type of organism
in establishing habitats, producing foods, or utilizing the raw materials to produce
good-quality industrial products, etc. Agriculture not only defines its significance in
natural shelters and foods but has a direct link with the income of a human being and
also of its country (Majeed et al. 2020a). Building a strong economy of a country
requires agricultural development followed by the establishment of industrial
processing to increase the GDP value of a country. It makes up half of the employed
labor force, accounts for around 24% of the GDP, and is the main source of foreign
exchange profits (Haq et al. 2022b). The entire rural and urban population’s feed
depends on agriculture. As agriculture is linked to the quality, production, and
growth of plants, it is necessary to simply protect and increase their production, so
as to live (Talaviya et al. 2020).
Plant diseases affect the quality of vegetables, organic goods, vegetables, and
cereals. Organisms which cause infectious diseases in plants, referred to as the “plant
pathogens,” include fungi, viroid, protozoa, oomycetes, parasitic weeds and plants,
nematodes, phytoplasmas, bacteria (Aziz et al. 2023a), virus-like organisms, and
viruses (Naveed et al. 2022a; Omara et al. 2023). In less developed nations, where
access to disease-control measures is limited, they are subjected to a yearly loss of
30–50% of vital crops which often leads to hunger and starvation (Aggarwal et al.
2022).
Plant disease identification or diagnosis is one of the major challenges in agri-
culture and a key component affecting the outcome of cultivating crops (Arif et al.
2021). If not detected before, plant pathogens can decrease the crop production rate
from 10 to 95%. Identification through the naked eye can be a tedious task for local
farmers as it requires professional observatory skills and expertise in plant pathology
15 Plant Disease Diagnosis with Artificial Intelligence (AI) 219

(Talaviya et al. 2020). Additionally, if a rare disease attacks a field, farmers seek
specialist help to make a precise and timely diagnosis, which inevitably results in
higher treatment expenses (Arshad et al. 2022). To avoid such complications, early
diagnosis of various plant diseases is one of the many farming aspects where
computerization in agriculture has made tremendous strides. Almost every country
has shifted its emphasis toward mechanizing agriculture in order to achieve accuracy
and precision in identifying plant diseases (Orchi et al. 2021). There is a vast array of
technology available to us now, including high-quality digital cameras with excel-
lent focus and resolution as well as cameras built into even our mobile phones. These
technologies are currently employed to collect datasets about plants
(Chithambarathanu and Jeyakumar 2023).
The development of artificial intelligence (AI) is based on the idea of prior
learning experiences. Artificial intelligence (AI) is playing a crucial role in replacing
traditional methods with quick, accurate, computerized, cost-effective, and, most
importantly, precise ways of identifying plant sickness (Majeed et al. 2021a). The
use of totally computer-based image managing innovation in rural designing
research produces amazing results about the capability of a programmed method
for identifying diseases and their treatments in plants so that we could protect them at
a very early stage before the destruction is done (de Oliveira and de Silva 2023).
Through early identification, prompt treatment, and reduced disease occurrence,
automatic detection techniques help farmers improve crop quality. Image processing
is utilized to quantify the size of the diseased area and identify color differences in
the affected area (Kamdar et al. 2021).
Machine learning (ML) techniques like backpropagation, artificial neural net-
works (ANN), and convolutional neural networks (CNNs) are automating the use of
machines and the development of cutting-edge inventions (Nawaz et al. 2020). The
ML-based techniques are easier to implement and do not need a large amount of
training data, but they are slow because of complicated preprocessing and rely on the
expertise of experienced human professionals for feature extraction and optimal
feature selection needed to accomplish classification (Tariq et al. 2023). ML-based
models were also introduced for the classification and identification of plant diseases
such as decision tree (DT), random forest (RF), K-nearest neighbors (KNN), and
support vector machine (SVM) (Kukadiya and Meva 2023).
The convolutional neural network (CNN) is a well-known DL model that dem-
onstrated efficient pattern recognition performance and is frequently used for
detecting early plant leaf diseases. In recent studies, CNN has mostly been employed
for diagnosing and classifying crop plant diseases (Hassan et al. 2022a). It has
convolutional layers, which are sets of image channels that have been integrated to
create images or spotlight maps (Kothari 2018). CNN optimizes the weights and
channel parameters in the protected-up layers to create highlights that are becoming
for solving problems in contrast to conventional device/algorithm learning method-
ologies that depend on hand-crafted highlights (Shoaib et al. 2023). Due to the
efficient feature representation used in these methods, crop-related classification
tasks have demonstrated encouraging results. Computer vision’s mature CNN archi-
tectures, such as AlexNet, GoogLeNet, VGGNet, ResNet, EfficientNet, MobileNet,
220 M. Naveed et al.

and DenseNet, are extensively used in existing plant disease categorization methods
(Albahli and Masood 2022).
Approaches based on deep learning (DL) and AI are widely used in many
industries, including agriculture (Ahmad et al. 2023). These methods avoid laborious
image preprocessing and have a smaller memory footprint since they automatically
synthesize discriminative features from the input samples (Albattah et al. 2022).
Therefore, we must use technology for great use and should keep progressing in
these fields of research; otherwise, if we do not focus on these problems today, it will
not be good for the future of humanity.

15.1.1 AI Technologies in Plant Diseases Diagnosis

15.1.1.1 Deep Learning and Neural Network

The first concept of the deep learning method originated from a paper by Hinton
et al. published in the Journal of Science. Deep learning uses neural networks to
analyze and learn from data, and several hidden layers act like artificial neurons to
extract data features. The artificial neuron is employed to first capture basic features
and subsequently merge these basic features to derive more advanced, abstract
features; this approach has the potential to significantly mitigate the basic problems.
A growing number of scientists are becoming interested in deep learning since it
addresses the problem that traditional methods rely on artificially manufactured
characteristics. Systems of recommendation, pattern recognition, speech recogni-
tion, and natural language processing now use it successfully (Ghosh and Roy
2021).
Conventional image classification and recognition techniques that depend on
manual design components make it difficult to obtain deep and complex picture
feature information. And the deep learning technique has the potential to eliminate
this barrier (Albahli and Masood 2022). It can do autonomous learning on the
original image right away to learn about numerous image feature stages, such as
low- to high-level semantic elements. Conventional plant disease and pathogen
identification algorithms mostly use manually generated image recognition charac-
teristics, which are complex and rely on luck and expertise; the original image’s
features cannot be automatically learned and extracted by these algorithms (Mahlein
2016). On the other hand, this approach is able to automatically extract features from
huge amounts of data without the need for any operatic system (Majeed et al. 2022a).
It consists of numerous coatings and has high unsupervised learning and feature
expression abilities. It can also autonomously extract picture features for visual
categorization and identification. Hence, deep learning can make a significant
contribution to the field of recognizing horticulture disorders in images (Mohanty
et al. 2016).
Many renowned deep neural network models have been produced as a result of
deep learning techniques like deep convolutional neural networks (CNNs), which is
15 Plant Disease Diagnosis with Artificial Intelligence (AI) 221

the most commonly employed methodology today. Compared to earlier manual

design feature extraction methods in picture recognition, using these deep neural
network models to automate feature extraction from high-dimensional feature data-
bases offers several advantages (Waheed et al. 2022). Moreover, with an expansion
of developmental data as well as advancements in computational capabilities, deep
neural networks continue to enhance their ability to characterize and understand
patterns. Currently, the surge in this approach is impacting both research and
manufacturing departments, with deep neural network models consistently
outperforming their traditional counterparts; over the past few years, the dominant
deep learning framework has been the deep convolutional neural network (Khoja
et al. 2022).

15.1.2 Convolutional Neural Network (CNN)

CNNs are experts at conducting convolution operations and have complex network
architecture. Neural network model, as illustrated in Fig. 15.1, is made up of a
convolution layer, an input layer, a pooling layer, an output layer, and a full
connection layer. There is no requirement for a full connection in one model,
where the convolution layer and the pooling layer alternate several times when the
neurons of the convolution layer are linked to the neurons of the pooling layer. A
model that’s frequently used in the DL field is CNN. This is due to CNN’s large
model capacity and ability to process complicated information as a result of its
fundamental structural features, which gives CNN an edge in picture identification
(Guerrero-Ibañez and Reyes-Muñoz 2023). Meanwhile, as a result of CNN’s success
in visual perception-related tasks, deep learning is growing in popularity. The
fundamental advantage of the convolution neural network is its local receptive
field, often known as the convolution core. The very initial step in this network is
to obtain some essential features information then after it, the neurons are sent into
the pooling layer, which retrieves the details twice. Now these pooling strategies
have been utilized to perform different computed functions (Biswal and Chestnut
2022). The data moves on to the fully connected layer, where the neurons create

Fig. 15.1 The structure of CNN

222 M. Naveed et al.

complete connections with the neurons in the highest layer, after passing through
many convolutional and pooling layers. In the final least, the output results from
classifying the full-connection layer data using the SoftMax technique are given to
the output layer as output results (Albattah et al. 2022).

15.2 Machine Learning Method

Identification of plant diseases is crucial to a successful agricultural system. The

majority of the time, a farmer can see disease symptoms in plants with the naked eye,
but this requires regular observation. However, this procedure is costlier and occa-
sionally less precise in big plantations. Farmers in other countries, like India, may be
required to show the specimen to specialists, which adds time and costs (Kukadiya
and Meva 2023; Thakur and Mittal 2020).

15.2.1 Disease Detection System for General Plant

Plant illnesses can be discovered by inspecting the plant’s leaf, stem, and root.
Digital image processing may be used to identify damaged leaves, stems, fruits,
and flowers, as well as the form and color of the afflicted region (Majeed et al.
2020b). It mainly consists of four basic steps elaborated in Fig. 15.2.

Fig. 15.2 Technique for

detecting plant diseases
15 Plant Disease Diagnosis with Artificial Intelligence (AI) 223

15.2.2 Data Acquisition

Different types of automated cameras are used to capture images to recognize any
illness in plants or crops.

15.2.3 Dataset Annotation

To construct a domain-specific dataset for taking images of various classifications.

15.2.4 Data Processing

Prior to further processing, acquired images must go through data-scrubbing phases

to improve certain picture properties. The segmentation process is employed to
divide the plant image into distinct segments, facilitating the extraction of disease-
affected regions in the plant’s stems, fruits, or leaves from the background.

15.2.5 Feature Extraction

In this step, plant physiology is determined by shape, color, height, and texture by
using various methods like gray-level co-occurrence matrix (GLCM), Blendvision,
and machine intelligence. In the end, plant disease can be classified via any tech-
nique (Khan et al. 2022).

15.3 Classification Techniques

15.3.1 Support Vector Machine (SVM) Classifier

The SVM classifier is the most commonly employed method of machine learning
technologies to classify and detect any plant pathogen and plant infection. The
detection of leaf damage produced by canker and anthracnose infections is required
for the diagnosis of illnesses affecting citrus trees, such as grapefruit, lemons, limes,
and oranges (Tassadduq et al. 2022). The experimental outcome produced a true
acceptance rate of 95%. Detection of grape plant infections achieved an average
accuracy rate of 88.89%. The identification of chimera and anthracnose infections in
oil palm leaves reached 97% and 95% accuracy rates, respectively (Singh et al.
2021). The identification of potato crop infections, specifically late blight and early
224 M. Naveed et al.

blight, was conducted using over 300 publicly accessible images, achieving an
accuracy rate of 95% (Liu and Wang 2021).
Created a supervised method for identifiying ailments in tea plants. Using SVM
classifiers, three distinct diseases characterized by fewer features were successfully
identified, achieving an accuracy rate of 90% through the developed method. A
substantial dataset was employed in soybean cultivation to detect three distinct
diseases, and the reported mean classification reliability for these diseases stood at
approximately 90% (Kumar et al. 2021).

15.3.2 Artificial Neural Network Classifier

The use of artificial neural networks (ANN), to diagnose patterns, has been explored
in previous studies concerning plant disease detection. An evaluation of a proposed
method for recognizing plant diseases was conducted, employing the feed-forward
backpropagation algorithm, which demonstrated strong performance with a preci-
sion rate of approximately 93%. They examined the solution’s effectiveness in
addressing diseases that impact plants (Vasudevan and Karthick 2023). To improve
the precision of recognizing fungal-induced illnesses in cucumber plants, a model
was created. A study conducted on identifying the groundnut plant disease com-
monly known as leaf spot using neural network backpropagation is proposed
(Majeed et al. 2021b). The empirical data demonstrate that they diagnosed four
categories of illnesses from 100 sample diseased leaf photos with 97.41% accuracy.

15.3.3 Fizzy Classifier

The method to determine the presence of infection in wheat crop photos using Fuzzy
Classifier in a related study on Fuzzy Classifier in plant disease detection. This
method has been tested on a dataset of healthy and unhealthy leaves. The categori-
zation of healthy and unhealthy leaves was discovered with an accuracy of 88%, and
illness detection with an accuracy of 56% (Lee et al. 2017). Table 15.1 summarizes
and provides the comparison of several machine learning classifiers employed in
identifying plant diseases.
Table 15.1 Summarizes and provides the comparison of several machine learning
classifiers employed in identifying plant diseases.

15.4 Advantages and Challenges

AI-powered plant disease diagnosis is transforming agriculture by providing accu-

rate, timely, and sustainable solutions to handle plant diseases.
15 Plant Disease Diagnosis with Artificial Intelligence (AI) 225

Table 15.1 Comparative analysis of classification techniques

Techniques classification Crops culture Diseases Accuracy rate
SVM classifier Citrus 2 95%
Grape 2 88.89%
Oil palm 2 97% and 95%
Potato 2 90%
Tea 3 93%
ANN classifier Cucumber 2 Improved accuracy
Pomegranate 4 90%
Groundnut 4 97.41%
Fizzy classifier Wheat 1 88%

15.4.1 Early Detection of Plant Diseases

AI-based systems are able to identify plant diseases in their very earliest stages,
normally before any outward signs show up. In most viral attacks, the symptoms
appear when either it’s too late or it has become difficult to save the crops from
spreading the disease (Ahmad et al. 2023). Early detection reduces the spread of
illnesses and enables prompt action. The segmentation of the plant leaf images into
surface areas like background, diseased area, and non-diseased area of the leaf is
ensured by image sensing and analysis. The sick or infected area is then removed and
sent to the lab for further analysis.

15.4.2 Rapid Diagnosis

AI systems are able to quickly process enormous amounts of data and images to
produce results. This speed is essential for making decisions quickly and taking the
appropriate steps to stop the spread of illnesses. The detection of diseases through
the naked eye and by observation requires a great level of skills and professionalism,
which can only be done by hiring a plant pathologist or expert. This process usually
takes up a lot of time. By using deep learning through image-based plant disease
techniques, expeditious diagnosis results can be obtained, and the spreading of
disease can be ceased (Shoaib et al. 2023).

15.4.3 Accurate Estimations

Large datasets can be analyzed by AI algorithms, which can spot tiny trends that can
be challenging for human experts to notice. As a result, disease diagnosis has high
rates of accuracy, which lowers the possibility of misdiagnosis. Real-time estima-
tions can be made during the cultivation phase by creating a field map and
226 M. Naveed et al.

identifying the places where the crops need water, fertilizer, and pesticides using
high-detail photos from drone and copter systems (Waheed et al. 2023).

15.4.4 Cost-Effectiveness

In the long run, using AI for disease diagnosis may be cost effective. It saves time
and money by reducing the need for manual labor and the reliance on specialists for
each diagnosis. Hiring a plant expert or pathologist is not preferred as it requires
exclusive equipment and techniques, which are mostly not affordable (Das et al.
2022).

15.4.5 Global Collaboration and Accessibility

As more data becomes accessible, AI models can continuously learn and increase
their diagnostic precision. The system will continue to be successful in recognizing
new and emerging diseases thanks to its versatility. AI-based timely disease man-
agement can boost crop production. Farmers can grow healthier crops and maximize
their harvests by preventing or controlling disease outbreaks. Even small-scale and
resource-constrained farmers can access these technologies to enhance their crop
management practices as AI systems become more popular and user-friendly.
AI-based disease diagnosis can promote international cooperation in plant health
management and research. Researchers and specialists from many fields can
exchange information and insights to produce more successful disease management
measures (Aziz et al. 2023b).

15.4.6 Continuous Monitoring

AI-powered systems can be set up in the field and used for remote plant health
monitoring. This makes it possible for farmers to monitor their crops and get instant
notifications when a disease outbreak occurs. Remote sensing (RS) techniques along
with hyperspectral imaging and 3D laser scanning are crucial to constructing crop
metrics over thousands of acres of cultivable land (Majeed et al. 2023).

15.4.7 Improved Yield and Crop Quality

AI can assist farmers in selectively applying remedies to damaged plants by pre-

cisely recognizing them, hence minimizing the overall use of pesticides and
15 Plant Disease Diagnosis with Artificial Intelligence (AI) 227

fungicides. This encourages the use of more environmentally responsible and

sustainable farming techniques. AI-based timely disease management can boost
crop production. Farmers can grow healthier crops and maximize their harvests by
preventing or controlling disease outbreaks (Majeed et al. 2022b). AI-capable
equipment that can automate irrigation and increase overall production are aware
of historical climate patterns, soil condition, and the kind of crops that should be
planted. Irrigation uses up close to 70% of the world’s freshwater supply; automation
can help farmers manage their water problems while also saving water (Singh et al.
2021).

15.4.8 Data-Driven Insights

AI systems produce useful data that can be utilized to track disease dynamics,
comprehend disease dynamics, and guide data-driven crop management decisions.
Drone-based photos can assist with field scanning, crop monitoring, and other tasks.
Farmers can connect them with IOT and PC vision technology to ensure rapid
actions. These feeds may continuously alarm farmers about the climate.

15.4.9 Robotics in Agriculture

Agribot, often known as Agbot, is a farming robot. It helps the farmer boost the
crop’s productivity and lessens the amount of manual labor he or she needs to
do. We may anticipate that these agricultural robots will perform tilling, sowing,
harvesting, and many other farm tasks on their own in the following generations.
Indeed, these agricultural robots will take care of weeding, pest control, and disease
management (Yağ and Altan 2022).

15.4.10 Integration with IoT

A wide range of sectors and businesses, including manufacturing, health, commu-

nications, energy, and agriculture, will be impacted by the “Internet of Things
(IoT).” The goal of the Internet of Things (IoT) use in agriculture is to equip farmers
with decision-making instruments and automation technologies that seamlessly
combine goods, information, and services for improved productivity, quality, and
profit (Thakur and Mittal 2020). A range of plant types and infections can be easily
accommodated by scaling AI models. They are adaptable to various climates and
farming methods, making them useful agricultural tools (Bashir et al. 2023). Drone-
based photos can assist with field scanning, crop monitoring, and other tasks.
228 M. Naveed et al.

Farmers can connect them with IoT and PC vision technology to ensure rapid
actions. These feeds may continuously alarm farmers about the climate.

15.4.11 Challenges and Limitations

AI-driven plant disease diagnosis has almost revolutionized the agriculture industry,
but there are some challenges and limitations which need to be addressed for the
future betterment and help mankind. These challenges are mentioned below along
with their reason for limitations (Salman et al. 2023).

15.4.12 Data Quality and Quantity

To enhance AI accuracy, we need extensive and high-quality datasets of diseased

and healthy plants. It required acquiring comprehensive datasets of various plant
diseases, which can be challenging, leading to potential biases in AI models.
Accurate labeling of images and data is time consuming and requires expertise,
making it a bottleneck in model training (Hussain et al. 2023).

15.4.13 Scalability

Scaling AI solutions to accommodate the needs of large agricultural operations is a

challenge that needs to be addressed. Implementing AI solutions in large agricultural
settings may require significant infrastructure upgrades and investment. Scaling AI
systems can be expensive, limiting access for smaller farmers or regions with limited
resources.

15.4.14 Ethical Considerations

AI adoption in agriculture must consider ethical concerns, such as data privacy and
AI bias. Collecting and sharing agricultural data for AI analysis raise concerns about
data privacy and ownership. AI algorithms may exhibit bias if training data is not
diverse, leading to unfair recommendations or decisions (Naveed et al. 2022b).
15 Plant Disease Diagnosis with Artificial Intelligence (AI) 229

15.4.15 Complexity of Plant Diseases

The complexity of plant disease is a major challenge researchers are facing these
days. Plant diseases can manifest differently based on factors like location, climate,
and plant variety, making it challenging to create universally applicable models. AI
models may struggle to identify diseases when multiple infections occur
simultaneously.

15.5 Future Prospects and Implications

AI is revolutionizing many industries like agriculture, food, and textiles. In recent

years, AI has been making significant trends in the field of plant disease diagnosis.
This chapter explores the current landscape of AI-driven plant disease diagnosis, its
future prospects, and the implications it holds for agriculture. Accurate and timely
diagnosis of plant diseases are crucial for ensuring food security and optimizing crop
yields (Majeed et al. 2022c). Traditional methods of disease detection are often time
consuming and less precise. Here, AI steps in to bridge the gap by offering more
efficient and accurate solutions. AI is being used for tasks like crop monitoring, soil
analysis, and even autonomous farming equipment. We will delve into the fusion of
AI techniques and the role of robotics and drones.

15.5.1 Fusion of AI Techniques

AI-powered cameras and sensors can identify disease symptoms by analyzing plant
images. This technology aids in early detection. Consider a farmer in the middle of a
big cornfield examining if any of the plants have a fungal infection that isn’t yet
obvious to the human eye simply by using their smartphone, which is loaded with an
AI-powered plant disease detection software. They take a snapshot of the corn
leaves, and the app analyzes them in seconds, finding minor discolorations and
patterns that indicate illness. The farmer receives quick diagnostic and treatment
suggestions (Naveed et al. 2022c), allowing them to act and avoid future spread.
Data analytics tools process vast amounts of agricultural data, identifying disease
trends and providing insights for preventive measures. Farmers may forecast disease
outbreaks and offer particular preventative actions by examining past data. For
example, they may recommend modifying irrigation schedules based on previous
disease trends to reduce the risk of fungal infections.
Internet of Things (IoT) devices can collect real-time data from the field, allowing
for immediate action when disease symptoms are detected. A network of IoT sensors
continually monitors the health of crops on a smart farm. Throughout the fields, soil
moisture monitors, meteorological stations, and disease detection sensors are
230 M. Naveed et al.

carefully positioned. These devices continually gather and transmit real-time data to
a centralized control system. An alarm is promptly delivered to the farmer’s
smartphone when a sensor detects unexpected moisture levels or disease-related
characteristics. This timely warning enables the farmer to analyze the problem and
take appropriate measures (Hassan et al. 2022b).

15.5.2 Robotics and Drones

Drones equipped with high-resolution cameras can monitor large fields efficiently,
detecting early signs of disease and nutrient deficiencies (Ahmed and Yadav 2023).
A high-tech drone equipped with a high-resolution camera takes flight in a big
vineyard. It flies across rows of grapevines, photographing the leaves and vines in
great detail. The photos are sent to a computer that is loaded with AI algorithms. The
AI examines the photos for tiny discolorations or anomalies that might indicate
grapevine disease. If any problems are discovered, the technology generates a map
identifying the afflicted regions, allowing the vineyard manager to precisely focus
treatment and avoid additional damage (Haq et al. 2022).
Robots equipped with AI algorithms can autonomously perform tasks like
weeding, planting, and harvesting, ensuring healthier crops (Yağ and Altan 2022).
Autonomous robots armed with AI algorithms patrol the fields of a vegetable farm.
These robots are taught to recognize weeds and differentiate them from cultivated
plants (Albattah et al. 2022). They employ computer vision to detect the shape and
features of the crops they meet as they travel through the rows. When a weed is
detected, the robot accurately administers herbicide to the undesirable plants while
leaving the crops alone. Not only does this automated crop management lessen the
need for pesticides, but it also encourages healthier, weed-free crops.

15.6 Conclusion

Agriculture, the heartbeat of our world, sustains life in more ways than one. Beyond
providing sustenance and raw materials, it shapes economies, livelihoods, and
ecosystems. However, the very foundation of agriculture, our plants, faces an
unrelenting challenge—disease. Plant diseases jeopardize the quality and quantity
of our crops, and in regions with limited resources, they often lead to hunger and
hardship. Timely and accurate diagnosis of these diseases is paramount, and this is
where technology, particularly artificial intelligence (AI) and machine learning, has
emerged as a beacon of hope. The chapter, realm AI-powered plant disease diagno-
sis, exploring its transformative potential, methodologies, and implications for
agriculture and the world at large.
In the grand tapestry of agriculture’s future, AI stands as a pivotal thread,
weaving together innovation, efficiency, and sustainability. While we celebrate its
15 Plant Disease Diagnosis with Artificial Intelligence (AI) 231

transformative potential in the diagnosis of plant diseases and its promise to shape a
brighter future for farming, we must also acknowledge the tapestry’s intricate
patterns, marked by the challenges and limitations that need our attention.

References

Aggarwal S, Suchithra M, Chandramouli N, Sarada M, Verma A, Vetrithangam D, Pant B,

Ambachew Adugna B (2022) Rice disease detection using artificial intelligence and machine
learning techniques to improvise agro-business. Sci Program 2022:1757888. https://doi.org/10.
1155/2022/1757888
Ahmad A, Saraswat D, El Gamal A (2023) A survey on using deep learning techniques for plant
disease diagnosis and recommendations for development of appropriate tools. Smart Agric
Technol 3:100083. https://doi.org/10.1016/j.atech.2022.100083
Ahmed I, Yadav PK (2023) Plant disease detection using machine learning approaches. Expert Syst
40(5):e13136. https://doi.org/10.1111/exsy.13136
Albahli S, Masood M (2022) Efficient attention-based CNN network (EANet) for multi-class maize
crop disease classification [Original Research]. Front Plant Sci 13:1003152. https://doi.org/10.
3389/fpls.2022.1003152
Albattah W, Javed A, Nawaz M, Masood M, Albahli S (2022) Artificial intelligence-based drone
system for multiclass plant disease detection using an improved efficient convolutional neural
network [Original Research]. Front Plant Sci 13:808380. https://doi.org/10.3389/fpls.2022.
808380
Arif U, Bhatti KH, Ajaib M, Wagay NA, Majeed M, Zeb J, Hameed A, Kiani J (2021) Ethnobo-
tanical indigenous knowledge of Tehsil Charhoi, District Kotli, Azad Jammu and Kashmir,
Pakistan. Ethnobot Res Appl 22:1–24. https://doi.org/10.32859/ERA.22.50.1-24
Arshad F, Waheed M, Harun N, Fatima K, Khan BA, Fatima K, Abbas Z, Jabeen S, Majeed M
(2022) Indigenous farmer’s perception about fodder and foraging species of semi-arid lowlands
of Pakistan: a case study of District Kasur, Pakistan. Taiwania 67(4):510–523
Aziz A, Anwar MM, Majeed M, Fatima S, Mehdi SS, Mangrio WM, Elbouzidi A, Abdullah M,
Shaukat S, Zahid N, Mahmoud EA, Casini R, Yessoufou K, Elansary HO (2023a) Quantifying
landscape and social amenities as ecosystem services in rapidly changing peri-urban landscape.
Land 12(2):477
Aziz T, Naveed M, Jabeen K, Shabbir MA, Sarwar A, Zhennai Y, Alharbi M, Alshammari A,
Alasmari AF (2023b) Integrated genome based evaluation of safety and probiotic characteristics
of Lactiplantibacillus plantarum YW11 isolated from Tibetan kefir. Front Microbiol 14:
1157615. https://doi.org/10.3389/fmicb.2023.1157615
Bashir SM, Altaf M, Hussain T, Umair M, Majeed M, Mangrio WM, Khan AM, Gulshan AB,
Hamed MH, Ashraf S, Amjad MS, Bussmann RW, Abbasi AM, Casini R, Alataway A, Dewidar
AZ, Al-Yafrsi M, Amin MH, Elansary HO (2023) Vernacular taxonomy, cultural and
ethnopharmacological applications of avian and mammalian species in the vicinity of Ayubia
National Park, Himalayan Region. Biology 12:4
Biswal B, Chestnut S (2022) DeepTrac: applying artificial intelligence in plant disease detection.
SoutheastCon 2022
Chithambarathanu M, Jeyakumar MK (2023) Survey on crop pest detection using deep learning and
machine learning approaches. Multimed Tools Appl 82:42277. https://doi.org/10.1007/s11042-
023-15221-3
Das S, Pattanayak S, Behera PR (2022) Application of machine learning: a recent advancement in
plant diseases detection. J Plant Prot Res 62(2):122–135. https://doi.org/10.24425/jppr.2022.
141360
232 M. Naveed et al.

de Oliveira RC, de Silva RDS (2023) Artificial intelligence in agriculture: benefits, challenges, and
trends. Appl Sci 13(13):7405. https://www.mdpi.com/2076-3417/13/13/7405
Ghosh A, Roy P (2021) AI based automated model for plant disease detection, a deep learning
approach. In: Dutta P, Mandal JK, Mukhopadhyay S (eds) Computational intelligence in
communications and business analytics. Springer, Cham
Guerrero-Ibañez A, Reyes-Muñoz A (2023) Monitoring tomato leaf disease through convolutional
neural networks. Electronics 12(1):229. https://www.mdpi.com/2079-9292/12/1/229
Haq SM, Tariq A, Li Q, Yaqoob U, Majeed M, Hassan M, Fatima S, Kumar M, Bussmann RW,
Moazzam MFU, Aslam M (2022) Influence of edaphic properties in determining forest com-
munity patterns of the Zabarwan Mountain range in the Kashmir Himalayas. Forests 13:8
Haq SM, Yaqoob U, Majeed M, Amjad MS, Hassan M, Ahmad R, Morales-de la Nuez A (2022b)
Quantitative ethnoveterinary study on plant resource utilization by indigenous communities in
high-altitude regions. Front Vet Sci 9:94404
Hassan M, Haq SM, Ahmad R, Majeed M, Sahito HA, Shirani M, Mubeen I, Aziz MA, Pieroni A,
Bussmann RW, Alataway A, Dewidar AZ, Al-Yafrsi M, Elansary HO, Yessoufou K (2022a)
Traditional use of wild and domestic fauna among different ethnic groups in the Western
Himalayas? Cross cultural analysis. Animals 12:17
Hassan M, Haq SM, Majeed M, Umair M, Sahito HA, Shirani M, Waheed M, Aziz R, Ahmad R,
Bussmann RW, Alataway A, Dewidar AZ, El-Abedin TKZ, Al-Yafrsi M, Elansary HO,
Yessoufou K (2022b) Traditional food and medicine: ethno-traditional usage of fish fauna
across the valley of Kashmir: a Western Himalayan region. Diversity 14:6
Hussain S, Raza A, Abdo HG, Mubeen M, Tariq A, Nasim W, Majeed M, Almohamad H, Al
Dughairi AA (2023) Relation of land surface temperature with different vegetation indices using
multi-temporal remote sensing data in Sahiwal region, Pakistan. Geosci Lett 10(1):33
Kamdar JH, Jasani MD, Jasani JD, Praba J, Georrge JJ (2021) Chapter 11. Artificial intelligence for
plant disease detection: past, present, and future. In: Jyotir Moy C, Abhishek K, Pramod
Singh R, Vishal J (eds) Internet of things and machine learning in agriculture. De Gruyter, pp
223–238. https://doi.org/10.1515/9783110691276-011
Khan AM, Li Q, Saqib Z, Khan N, Habib T, Khalid N, Majeed M, Tariq A (2022) MaxEnt
modelling and impact of climate change on habitat suitability variations of economically
important Chilgoza pine (Pinus gerardiana Wall.) in South Asia. Forests 13:5
Khoja AA, Haq SM, Majeed M, Hassan M, Waheed M, Yaqoob U, Bussmann RW, Alataway A,
Dewidar AZ, Al-Yafrsi M, Elansary HO, Yessoufou K, Zaman W (2022) Diversity, ecological
and traditional knowledge of pteridophytes in the Western Himalayas. Diversity 14:8
Kothari J (2018) Plant disease identification using artificial intelligence: machine learning approach.
Int J Innov Res Comput Commun Eng 7:11082–11085
Kukadiya H, Meva DD (2023) Machine learning in agriculture for crop diseases identification: a
survey. Int J Res GRANTHAALAYAH 11:5099. https://doi.org/10.29121/granthaalayah.v11.
i3.2023.5099
Kumar S, Jain A, Shukla AP, Singh S, Raja R, Rani S, Harshitha G, AlZain MA, Masud M (2021) A
comparative analysis of machine learning algorithms for detection of organic and nonorganic
cotton diseases. Math Probl Eng 2021:1790171. https://doi.org/10.1155/2021/1790171
Lee SH, Chan CS, Mayo SJ, Remagnino P (2017) How deep learning extracts and learns leaf
features for plant classification. Pattern Recogn 71:1–13. https://doi.org/10.1016/j.patcog.2017.
05.015
Liu J, Wang X (2021) Plant diseases and pests detection based on deep learning: a review. Plant
Methods 17(1):22. https://doi.org/10.1186/s13007-021-00722-9
Mahlein A-K (2016) Plant disease detection by imaging sensors—parallels and specific demands
for precision agriculture and plant phenotyping. Plant Dis 100(2):241–251. https://doi.org/10.
1094/pdis-03-15-0340-fe
Majeed M, Bhatti KH, Amjad MS, Abbasi AM, Rashid A, Nawaz F, Ahmad KS (2020a) Ethno-
veterinary practices of Poaceae taxa in Punjab, Pakistan
15 Plant Disease Diagnosis with Artificial Intelligence (AI) 233

Majeed M, Bhatti KH, Amjad MS, Abbasi M, Id RWB, Nawaz F, Rashid A, Mehmood A, Id MM,
Khan WM, Id SA (2020b) Ethno-veterinary uses of Poaceae in Punjab, Pakistan. PLoS One
15(11):e0241705. https://doi.org/10.1371/journal.pone.0241705
Majeed M, Bhatti KH, Pieroni A, Sõukand R, Bussmann RW, Khan AM, Chaudhari SK, Aziz MA,
Amjad MS (2021a) Gathered wild food plants among diverse religious groups in Jhelum
District, Punjab, Pakistan. Foods 10:3
Majeed M, Tariq A, Anwar MM, Khan AM, Arshad F, Mumtaz F, Farhan M, Zhang L, Zafar A,
Aziz M, Abbasi S, Rahman G, Hussain S, Waheed M, Fatima K, Shaukat S (2021b) Monitoring
of land use? And cover change and potential causal factors of climate change in Jhelum District,
Punjab, Pakistan, through GIS and multi-temporal satellite data. Land 10:10
Majeed M, Khan AM, Habib T, Anwar MM, Sahito HA, Khan N, Ali K (2022a) Vegetation
analysis and environmental indicators of an arid tropical forest ecosystem of Pakistan. Ecol
Indic 142:109291. https://doi.org/10.1016/j.ecolind.2022.109291
Majeed M, Lu L, Haq SM, Waheed M, Sahito HA, Fatima S, Aziz R, Bussmann RW, Tariq A,
Ullah I, Aslam M (2022b) Spatiotemporal distribution patterns of climbers along an abiotic
gradient in Jhelum District, Punjab, Pakistan. Forests 13:8
Majeed M, Tariq A, Haq SM, Waheed M, Anwar MM, Li Q, Aslam M, Abbasi S, Mousa BG, Jamil
A (2022c) A detailed ecological exploration of the distribution patterns of wild Poaceae from the
Jhelum District (Punjab), Pakistan. Sustainability (Switzerland) 14:7
Majeed M, Lu L, Anwar MM, Tariq A, Qin S, El-Hefnawy ME, El-Sharnouby M, Li Q, Alasmari A
(2023) Prediction of flash flood susceptibility using integrating analytic hierarchy process
(AHP) and frequency ratio (FR) algorithms. Front Environ Sci 10:1037547. https://doi.org/10.
3389/fenvs.2022.1037547
Mohanty SP, Hughes DP, Salathé M (2016) Using deep learning for image-based plant disease
detection [methods]. Front Plant Sci 7:01419. https://doi.org/10.3389/fpls.2016.01419
Naveed M, Jabeen K, Naz R, Mughal MS, Rabaan AA, Bakhrebah MA, Alhoshani FM,
Aljeldah M, Shammari BRA, Alissa M, Sabour AA, Alaeq RA, Alshiekheid MA, Garout M,
Almogbel MS, Halwani MA, Turkistani SA, Ahmed N (2022a) Regulation of host immune
response against i proteins via computational mRNA vaccine design through transcriptional
modification. Microorganisms 10(8):10081621. https://doi.org/10.3390/
microorganisms10081621
Naveed M, Makhdoom SI, Ali U, Jabeen K, Aziz T, Khan AA, Jamil S, Shahzad M, Alharbi M,
Alshammari A (2022b) Immunoinformatics approach to design multi-epitope-based vaccine
against Machupo virus taking viral nucleocapsid as a potential candidate. Vaccines (Basel)
10(10):1732. https://doi.org/10.3390/vaccines10101732
Naveed M, Mughal MS, Jabeen K, Aziz T, Naz S, Nazir N, Shahzad M, Alharbi M, Alshammari A,
Sadhu SS (2022c) Evaluation of the whole proteome to design a novel mRNA-based vaccine
against multidrug-resistant Serratia marcescens. Front Microbiol 13:960285. https://doi.org/10.
3389/fmicb.2022.960285
Nawaz M, Khan TS, Mudassar RHA, Kausar M, Usman A, Fatima T, Ahmad R, Ahmad J (2020)
Plant disease detection using internet of thing (IoT). Int J Adv Comput Sci Appl 11:5
Omara J, Talavera E, Otim D, Turcza D, Ofumbi E, Owomugisha G (2023) A field-based
recommender system for crop disease detection using machine learning [original research].
Front Artif Intell 6:1010804. https://doi.org/10.3389/frai.2023.1010804
Orchi H, Sadik M, Khaldoun M (2021) On using artificial intelligence and the internet of things for
crop disease detection: a contemporary survey. Agriculture 12:9. https://doi.org/10.3390/
agriculture12010009
Salman Z, Muhammad A, Piran MJ, Han D (2023) Crop-saving with AI: latest trends in deep
learning techniques for plant pathology [review]. Front Plant Sci 14:4709. https://doi.org/10.
3389/fpls.2023.1224709
Shoaib M, Shah B, Ei-Sappagh S, Ali A, Ullah A, Alenezi F, Gechev T, Hussain T, Ali F (2023) An
advanced deep learning models-based plant disease detection: a review of recent research
[review]. Front Plant Sci 14:1158933. https://doi.org/10.3389/fpls.2023.1158933
234 M. Naveed et al.

Singh T, Kumar K, Bedi SS (2021) A review on artificial intelligence techniques for disease
recognition in plants. IOP Conf Ser Mater Sci Eng 1022(1):012032. https://doi.org/10.1088/
1757-899X/1022/1/012032
Talaviya T, Shah D, Patel N, Yagnik H, Shah M (2020) Implementation of artificial intelligence in
agriculture for optimisation of irrigation and application of pesticides and herbicides. Artif Intell
Agric 4:58–73. https://doi.org/10.1016/j.aiia.2020.04.002
Tariq A, Mumtaz F, Majeed M, Zeng X (2023) Spatio-temporal assessment of land use land cover
based on trajectories and cellular automata Markov modelling and its impact on land surface
temperature of Lahore district Pakistan. Environ Monit Assess 195:1
Tassadduq SS, Akhtar S, Waheed M, Bangash N, Nayab DE, Majeed M, Abbasi S, Muhammad M,
Alataway A, Dewidar AZ, Elansary HO, Yessoufou K (2022) Ecological distribution patterns of
wild grasses and abiotic factors. Sustainability (Switzerland) 14:18
Thakur T, Mittal A (2020) Real time IoT application for classification of crop diseases using
machine learning in cloud environment. Int J Innov Sci Mod Eng 6:1–4. https://doi.org/10.
35940/ijisme.D1186.016420
Vasudevan N, Karthick T (2023) A hybrid approach for plant disease detection using E-GAN and
CapsNet. Comput Syst Sci Eng 46(1):337–356. http://www.techscience.com/csse/v46n1/51322
Waheed M, Arshad F, Majeed M, Fatima S, Mukhtar N, Aziz R, Mangrio WM, Almohamad H,
Dughairi AA, Al-Mutiry M, Abdo HG (2022) Community structure and distribution pattern of
Woody vegetation in response to soil properties in semi-arid Lowland District Kasur Punjab,
Pakistan. Land 11:12
Waheed M, Arshad F, Majeed M, Haq SM, Aziz R, Bussmann RW, Ali K, Subhan F, Jones DA,
Zaitouny A (2023) Potential distribution of a noxious weed (Solanum viarum Du-nal), current
status, and future invasion risk based on MaxEnt modeling. In: Geology, ecology, and land-
scapes. Taylor & Francis, p 1. https://doi.org/10.1080/24749508.2023.2179752
Yağ İ, Altan A (2022) Artificial intelligence-based robust hybrid algorithm design and implemen-
tation for real-time detection of plant diseases in agricultural environments. Biology 11(12):
1732. https://www.mdpi.com/2079-7737/11/12/1732
Chapter 16
Sustainable AI-Driven Applications
for Plant Care and Treatment

Muhammad Naveed , Nafeesa Zahid, Ibtihaj Fatima, Ayesha Saleem,

Muhammad Majeed , Amina Abid, Khushbakht Javed, Rehmana Wazir,
and Amina Qasim

Abstract Technology has emerged as a formidable ally in the endeavor, with

artificial intelligence (AI) leading the charge in revolutionizing farming practices
via the notion of precision agriculture. Since the dawn of the civilization, there has
always been a basic human need in the field of agriculture as plants were a primary
source of food. The plant disease poses a serious threat to food security because it
legitimately affects harvest yield, which reduces the growth of a crop yield and leads
to significant loss and consequent financial losses. As a result, there is a demand for
quick and effective plant disease detection and evaluation strategies. AI intervention
in agriculture is assisting farmers in regaining their farming efficiency and reducing
adverse environmental influences. By substituting traditional methods with more
effective ones, it is bringing about a revolution in agriculture. Plant disease is one of
the most important and critical challenges that affects the agriculture and its trading
system. Plant diseases can adversely affect the growth of plants and, in severe cases,
might lead to the plant death. AI-driven technologies have the capability not only to
detect the disease but also to assess its toxicity and precisely classify the specific type

M. Naveed · A. Saleem · A. Abid · K. Javed · R. Wazir

Department of Biotechnology, Faculty of Science and Technology, University of Central
Punjab, Lahore, Punjab, Pakistan
e-mail: [email protected]
N. Zahid
Department of Botany, University of Kotli, Kotli, Azad Jammu and Kashmir, Pakistan
e-mail: [email protected]
I. Fatima
Department of Botany, University of Education, Lahore, Punjab, Pakistan
M. Majeed (✉)
Department of Botany, University of Gujrat, Gujrat, Pakistan
e-mail: [email protected]; https://www.researchgate.net/profile/Muhammad-Majeed-18
A. Qasim
Department of Botany, Minhaj University Lahore, Lahore, Pakistan
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 235
A. Khamparia et al. (eds.), Microbial Data Intelligence and Computational
Techniques for Sustainable Computing, Microorganisms for Sustainability 47,
https://doi.org/10.1007/978-981-99-9621-6_16
236 M. Naveed et al.

of disease detected in a given plant sample. It investigates AI’s revolutionary role in

improving farming practices through precision agriculture. Precision agriculture is
based on data-driven decision-making. The importance of AI extends beyond data
collection to data processing and interpretation. AI algorithms analyze the acquired
data to uncover patterns, correlations, and anomalies that the human eye may not see.

Keywords Plant diseases · Artificial intelligence · Environmental influences

16.1 Introduction

Regardless of the developmental stages of countries, agriculture system has always

been serving as the backbone of economy (Arzani and Ashraf 2017). The signifi-
cance of agriculture is evident in several domains as approximately 70% of the
population rely on the cultivation of crops for their needs. This depicts the impor-
tance of agriculture as being the most vital source that could stand the chance in the
era of rapidly growing population (Sekaran et al. 2021). These diseases can occur at
various stages of plant development such as seedling, seed development, and its
growth (Majeed et al. 2020a). When plants get affected with diseases, they undergo a
range of mechanical, morphological, and biochemical changes (Van Esse et al.
2020). Moreover, plant stress can be categorized into two primary types: biotic
stress, which involves interactions with living organisms in a way that affect plant
growth, such as viruses, bacteria, and fungi. The other type is abiotic stress,
including non-living or environmental factors (Haq et al. 2022). As the world’s
population is increasing, which limits the agricultural land resources, it becomes
difficult to ensure the efficient production of agricultural products. For the produc-
tion of agricultural products, detection of disease at an early stage, fertilization,
spraying and detection of unwanted weed is important (Mosa et al. 2017). According
to the research, plant disease could cause 20–40% of losses to the crops yield %,
thereby reducing market availability for potential buyers. Delayed diagnosis of
pathogenic, viral, or pest-related plant diseases necessitates the application of higher
doses of pesticides, ultimately compromising crop quality (Rashid et al. 2019). Early
and accurate identification of plant diseases not only enhances the overall quality of
agricultural products but also minimizes the need for chemical sprays such as
fungicides and herbicides. Since agriculture is essential, there is need for techniques
that enhance the agriculture methods in terms of monitoring, planting, harvesting,
and detecting plant diseases (Mohamed et al. 2021).
However, visual inspection based on the traditional method is used for detecting
plant diseases in small-scale agriculture fields (Majeed et al. 2022a). However, it is
difficult to use these methods in large-scale agriculture as this method is continuous
and tiring. This challenge becomes even more difficult when diagnosing plants with
similar leaves for both species and diseases, often resulting in visual errors
(Martinelli et al. 2015). To address these issues, diverse image processing and
artificial intelligence (AI) approaches, including deep learning and machine learning
16 Sustainable AI-Driven Applications for Plant Care and Treatment 237

Fig. 16.1 Role of artificial intelligence (AI) in agriculture system

(ML), are used for real-time plant disease classification (Fig. 16.1) (Rakhmatulin
et al. 2021).
The academic literature in this field had rapid growth from 2000 to 2015.
Subsequently, there was a great impact of industry on different agriculture practices;
hence there is a gradually increasing trend since 2017 and substantial 255.7%
increase in the number of Scopus-indexed articles and reviews published between
2019 and 2020 (Ullah et al. 2022) (Fig. 16.2). This growth strongly indicates a
pronounced scholarly interest in exploring the applications of AI within the context
of sustainable agriculture practices.
The detection methods of plant disease can be categorized into two main groups
based on their complexity: machine learning and deep learning (Jackulin and
Murugavalli 2022). Basic machine learning techniques such as Support Vector
Machine (SVM), K-Nearest Neighbor (KNN), random forest (RF), and Naïve
Bayes (NB) rely on specifically designed features, necessitating the recognition of
patterns. These specific features are Histogram of Oriented Gradient (HOG), Hue
Saturation Value (HSV), Red-Green-Blue (RGB) color features, and Linear Binary
Pattern (LBP) (Zhang et al. 2020). In machine learning, more data is required to
achieve satisfactory results. Firstly, the data is collected and characterized according
to the classes of disease. Subsequently, for the image to be processed, dataset is
created for further extraction. Machine learning has the algorithms which are capable
of recognizing any change in features through comparison and thus classifying the
238 M. Naveed et al.

Numbers of Papers Published

1995-2000

2001-2005

2006-2010

2010-2015

2015-2020

0 1000 2000 3000 4000 5000 6000 7000

IoT and WSN Based Research IoT and Machine Learning Based Research

Fig. 16.2 Yearwise number of publications based on the AI agricultural practices

output data as healthy or diseased (Meng et al. 2020). Conversely, deep learning
techniques such as Convolutional Neural Networks (CNNs) have found extensive
application in research focused on plant disease detection. While the Support Vector
Machine (SVM) was the predominant machine learning approach for an extended
period, there was a notable shift around 2015 when Convolutional Neural Networks
(CNNs) replaced SVM as the most widely adopted machine learning technique for
disease detection (Tugrul et al. 2022). There are various types of CNN such as
Inception-V3, AlexNet, ResNet50, and VGG16 (Xiao et al. 2018). However, deep
machine learning needs large set of data, which is often considered a challenge.
Nowadays, CNN has been used as the art of the model for the detection of plant
diseases, especially since this task requires dealing with image data applications
(Waheed et al. 2022). However, these technological approaches could be used for
the detection of plant diseases. Thus, these methods can be viewed as innovative
means of detecting diseases, as they rely on image-processing techniques rather than
destructive serology and molecular methods (Alzubaidi et al. 2021).

16.1.1 Role of AI in Optimizing Farming Practices Through

Precision Agriculture

Precision agriculture collects, analyzes, and interprets data using AI-driven solu-
tions, allowing farmers to make educated decisions that optimize resource utiliza-
tion, boost production, and encourage sustainable agricultural practices (Linaza et al.
2021). With its capacity to handle massive volumes of data and extract useful
16 Sustainable AI-Driven Applications for Plant Care and Treatment 239

insights, AI serves as the backbone of this strategy (Khoja et al. 2022). One of AI’s
most important contributions is satellite imaging, drones, and Internet of Things
(IoT) devices, which are used to collect real-time data on crop health, soil moisture,
temperature, and other factors. These technologies enable farmers to monitor their
fields in unprecedented detail by providing high-resolution pictures and remote
sensing capabilities. AI systems give a thorough grasp of field circumstances by
continually recording and analyzing data, resulting in proactive and accurate
responses (Singh et al. 2021).
These insights are essential in anticipating and avoiding possible problems. For
example, using historical and real-time data, AI can forecast disease outbreaks,
insect infestations, and yield changes (Coulibaly et al. 2022). Farmers can use this
forecasting capacity to respond early, employing tailored remedies and minimizing
crop yield losses. Furthermore, based on data analysis, AI can identify appropriate
planting density, irrigation rates, and fertilizer application, ensuring that resources
are dispersed efficiently across the field (Tassadduq et al. 2022). Another incredible
use of AI in precision agriculture is yield prediction. AI algorithms can reliably
anticipate agricultural yields by analyzing a variety of parameters such as weather
patterns, soil quality, and planting strategies. This forecast enables farmers to make
smart decisions about harvesting schedules, storage, and selling, resulting in
increased profitability (Sudha 2023; Mathenge et al. 2022). The various AI applica-
tions for plant care and treatment used in different agriculture and cultivating fields
are shown in Table 16.1.
AI-powered solutions have a significant influence on resource management. For
example, automated irrigation systems modify water applications depending on real-
time soil moisture levels and weather forecasts. This accuracy reduces water waste
and increases water-use efficiency, which is crucial in water-stressed areas. Simi-
larly, the importance of AI in weed and pest management cannot be emphasized
(Rathore 2017). Artificial intelligence-powered image identification systems identify
and categorize weeds, pests, and illnesses in real time, allowing for focused treat-
ments and lowering dependency on broad-spectrum pesticides. Precision agriculture
encourages sustainable agricultural practices that limit environmental damage by
minimizing the excess use of water, fertilizers, and pesticides (Majeed et al. 2020b).
Furthermore, AI’s capacity to optimize resource utilization is consistent with the
ideas of the circular economy, in which inputs are utilized more efficiently, waste is
eliminated, and the total environmental impact is decreased (Agrawal et al. 2021).
However, it is crucial to recognize that the use of AI in agriculture is fraught with
difficulties. For small-scale farmers, the initial investment in technology, training,
and infrastructure might be intimidating (Minkoff-Zern 2019). Furthermore, data
privacy and security issues must be addressed to guarantee that sensitive agricultural
data is not affected. Collaboration among technology developers, agricultural spe-
cialists, politicians, and farmers is critical to overcoming these obstacles and
establishing an ecosystem that promotes the wider use of AI-powered precision
agriculture (Gardezi et al. 2023). The use of AI in optimizing agricultural practices
through precision agriculture is a transformational force with the potential to influ-
ence the future of agriculture. The combination of technology and agriculture has the
240 M. Naveed et al.

Table 16.1 Various applications of AI in agriculture for plant care and treatment
Applications Description References
Soil analysis Based on climate, geography, and generally stable soil proper- Karlen et al.
ties (such as soil texture, depth, and mineralogy), land man- (2001)
agement has long-term potential. This application assists
farmers in better understanding the potential of their property as
well as the possibility for climate variation change and mitiga-
tion methods
Farm manager The farm manager app assists farmers in determining which Rehman et al.
tactics to use before planting begins. Without using your phone, (2022)
this programmer views, organizes, and modifies any informa-
tion about your field, such as yield, planting, and spraying
conditions
Pest Village tree offers smart pest management solutions by gather- Sun et al.
management ing pest incidence information from farms. Furthermore, it uses (2021)
a crowdsourcing technique, distributing photographs and loca-
tion data to other farmers who may be affected
Agrippa By using the eFarmer Application, farmers may produce elec- Karetsos
tronic maps of their fields, record a history of crops grown in the et al. (2014)
field (e.g., planting, fertilizing, harvesting, warehouses, petrol
stations), and track the position of things in the field (e.g., soil
sample for agrochemical laboratories)
Semios Network coverage, orchard pests, frost, diseases, and irrigation Rehman et al.
are all covered. As part of the monitoring services, event alerts (2022)
are sent out in real time
Fertilizer EcoFert aids in fertilizer management, allowing it to be utilized Rehman et al.
management to its greatest capacity. It determines the ideal fertilizer combi- (2022)
nation to cover the required nutrient suspension and takes into
account the needs of various yields. Furthermore, it takes into
account the cost of fertilizer, depending on current market prices

potential to feed the world’s rising population while also assuring appropriate
management of our planet’s resources (Majeed et al. 2021a). As AI advances, its
impact on precision agriculture will definitely expand, providing novel solutions to
the difficulties of a changing world.

16.1.2 Agriculture Satellite Imagery: A Technological

Revolution for Sustainable Farming

The potential for monitoring and management is at the heart of satellite imagery’s
importance in agriculture. Farmers can now monitor vast areas of land with unprec-
edented clarity and accuracy, allowing them to detect early signs of crop stress,
disease, insect infestations, and nutritional deficiencies (Arshad et al. 2022). This
proactive technique allows for early intervention, which reduces production losses
and eliminates the need for unnecessary chemical inputs (Lamb et al. 2008). Satellite
16 Sustainable AI-Driven Applications for Plant Care and Treatment 241

imaging enables farmers to make educated decisions regarding irrigation, fertiliza-

tion, and pest management tactics by giving vital information on crop health and
development. This improves resource efficiency and production, which is critical as
the world’s population continues to grow (Sinha et al. 2023). Satellite imagery’s
predictive capacity strengthens its importance in agriculture. This capacity to predict
harvest quantities has a dramatic impact on supply chain management, allowing for
improved coordination among farmers, distributors, and consumers (Raj et al. 2022).
Furthermore, satellite imaging data-driven insights enhance precision agriculture
practices. Farmers may use resources like water, fertilizer, and pesticides more
efficiently by separating fields into zones based on changes in soil type and vege-
tation. This will not only reduce waste but also have a lower environmental effect,
which is consistent with the spirit of sustainable farming (Jadhav et al. 2023).
Satellite photography also enables farmers to better address water management
concerns. Monitoring soil moisture levels and crop water stress aids in the optimi-
zation of irrigation practices, ensuring that water resources are utilized efficiently
(Khan et al. 2022a). This is especially important in water-stressed areas, since it
allows for agricultural cultivation while minimizing water waste. Furthermore, using
satellite data with weather forecasting improves farmers’ capacity to adapt to
extreme weather occurrences (Alzubaidi et al. 2021). Farmers can take preemptive
actions to safeguard their crops and minimize possible losses if they get timely
information about imminent droughts, strong rains, or frost. Satellite photography
assists in monitoring changes in land cover, deforestation, and urban expansion,
providing insights into the changing terrain. Such data is crucial to legislators and
urban planners because it helps them make decisions regarding land use, environ-
mental protection, and sustainable development. Furthermore, satellite photography
aids agricultural research and development. These data may be used by researchers
to investigate production patterns and measure the influence of various agricultural
practices (Tiwari et al. 2023).

16.1.3 Use of Sensors and Data Analytics

Sensors are at the cutting edge of this revolutionary journey, acting as the digital
world’s eyes and ears. These devices collect information from the physical environ-
ment, such as temperature, humidity, soil moisture, air quality, and energy use
(Liang and Shah 2023). Sensors put in fields, for example, give real-time information
about soil conditions and crop health, allowing farmers to modify irrigation and
fertilization schedules. However, the true value of these sensors comes in their
capacity to generate vast amounts of data. Advanced analytical approaches are
necessary to extract relevant insights from this abundance of data. This is where
data analytics can help (Tariq et al. 2023). Data analytics is the process of
processing, interpreting, and visualizing acquired data using modern algorithms
and statistical approaches. In the context of resource allocation, data analytics
helps decision-makers to spot patterns, trends, and anomalies that might otherwise
242 M. Naveed et al.

Table 16.2 IoT sensor types used in agriculture

Sr. Application in
no. Sensor type Sensor Function agriculture References
1 Optical Photodiode Use of light to Determines the Wai et al.
measure soil soil’s clay, (2021)
properties organic matter,
and moisture
content
2 Mechanical Tensiometer Take action by Detects the root Peixoto
using probes to force employed in et al.
detect soil water absorption (2019)
compaction
3 Electro- Ion-selective elec- Electrodes are Detecting nitro- Kim et al.
mechanical trodes and used to detect cer- gen, potassium, (2013)
ion-selective fields tain ions in soil and phosphorus
affect transistor (NPK) levels in
sensor soils
4 Di-electric Electrodes for fre- To assess moisture To measure water Kashyap
soil quency or time levels content from soil and
moisture domain Kumar
reflectometry (2021)
5 Airflow Measurements can To measure soil Classifies soil Pawar and
be taken at a fixed permeability kinds, moisture Deosarkar
location using a levels, and soil (2023)
designated method structure/
compaction
6 Location Global positioning Provides latitude, The GPS system Nayfeh
system longitude, and alti- allows for exact et al.
tude information location (2023)
as well as altitude

go undetected using traditional approaches. In healthcare, for example, data analyt-

ics may be used to analyze patient information to forecast disease outbreaks and
allocate medical resources appropriately (Kuppusamy et al. 2021). These cutting-
edge approaches of AI make the farming sector more adaptable. Soil moisture and
fertility can be monitored by biosensors. Past data of weather fluctuation with
non-linear dependencies can be predicted by neural networks. Therefore, econom-
ically important commodities like rice, wheat, and maize can be sowed at appropriate
times by using the knowledge of AI. Farms produce thousands of data points on
temperature, soil, water level, and weather on a daily basis. These data are used by
AI and ML models for greater yield (Table 16.2). Some crops need an adequate
amount of water, and water at a greater scale became tough, so the AI irrigation
system was introduced.
16 Sustainable AI-Driven Applications for Plant Care and Treatment 243

16.2 Drones in Agriculture: Planting the Seeds of a

High-Tech Future

The use of technology into agriculture has resulted in dramatic developments, with
the drone being one of the most inventive technologies to emerge. Unmanned aerial
vehicles, or drones, have found wide and effective applications in a variety of
industries, including agriculture. Drones have revolutionized farming practices by
capturing high-resolution pictures, collecting data, and providing insights from
above, ushering in a new era of accuracy, efficiency, and sustainability (Khan and
Shahriyar 2023).

16.2.1 Precision Farming

Drones have become an integral part of precision agriculture, allowing farmers to

manage their crops. Drones, which are outfitted with modern cameras and sensors,
may take comprehensive photos and data, providing insights into numerous aspects
of crop health and growth (Bashir et al. 2023). These discoveries span from
identifying nutrient shortages, illnesses, and insect infestations to determining irri-
gation requirements and monitoring plant stress. Farmers may make reliable deci-
sions regarding resource allocation, waste reduction, and yield potential using this
information (Maddikunta et al. 2021).

16.2.2 Crop Monitoring and Management

Drones provide effective crop monitoring over wide regions. It may acquire data that
the human eye cannot see by using imaging technology such as multispectral and
infrared cameras. Farmers can monitor tiny changes in plant health and soil condi-
tions, allowing for preventive treatments. The capacity to detect problems early and
administer focused therapies leads to increased production and less reliance on
broad-spectrum chemicals (Budiharto et al. 2019).

16.2.3 Mapping and Surveying

Drones excel in producing precise and up-to-date maps of agricultural lands. They
can make accurate topographical maps, determine changes in soil type and elevation,
and even construct 3D representations of the ground using aerial surveys (Majeed
et al. 2022b). These maps help in the construction of efficient drainage systems, the
optimization of irrigation practices, and the uniform use of resources. Drone
244 M. Naveed et al.

mapping capabilities also include monitoring changes in land cover, which aids in
land use planning and management (Daponte et al. 2019).

16.2.4 Yield Prediction and Assessment

Drones help in yield prediction and assessment by collecting data on plant density,
canopy cover, and flowering patterns. When this data is paired with modern analyt-
ics, growers may more correctly anticipate possible crops. They also aid in post-
harvest inspections, providing information on yield variances across different por-
tions of a crop. This information is useful for assessing the effectiveness of various
planting and management practices (Ben Ayed and Hanana 2021).

16.2.5 Environmental Stewardship

Drones can help promote sustainable farming practices. They can reduce the envi-
ronmental effect of fertilizers and pesticides by permitting accurate resource appli-
cation. Furthermore, their capacity to monitor soil moisture levels and plant health
assists in effective water consumption and decreases agriculture’s ecological imprint
(Dutta and Mitra 2021).

16.2.6 Internet of Things (IoT) in Agriculture: Real-Time

Data Collection for Better Farming

The Internet of Things (IoT) has ushered in a new era of connectivity and data-driven
decision-making across several industries, and agriculture has seen significant ben-
efits from this technological revolution. The research article published in Agronomy
investigated the potential of Internet of Things (IoT) technologies to play a pivotal
role in the evolution of smart agriculture in the twenty-first century. Illustrative
instances of IoT applications within agriculture include farm management, control-
ling greenhouse environments, monitoring animals and herds, tracking emissions,
managing irrigation systems, operating autonomous machinery, and deploying
drones (Fig. 16.3). These gadgets allow precision agriculture, decrease resource
waste, and contribute to sustainable agricultural practices by permitting real-time
data collection and analysis (Majeed et al. 2021b).
16 Sustainable AI-Driven Applications for Plant Care and Treatment 245

Drones

IOT Airblast
Tractor

Sensors
Sensors

Data Data
Receiving Analytics

Fig. 16.3 The IoT-based smart agriculture monitoring system

16.2.7 Weather Monitoring and Forecasting

The weather has a significant impact on agricultural outputs. Temperature, humidity,

wind speed, and precipitation data are collected by IoT weather stations. This
information not only is important for on-farm decision-making but also helps with
accurate weather predictions. Farmers can predict natural weather disasters and
adjust their operations accordingly, safeguarding their crops and minimizing poten-
tial losses (Khan et al. 2022b).

16.2.8 Livestock Management

IoT devices are also used in animal husbandry. Smart collars with sensors can
monitor animal health and behavior. Tracking movement patterns, spotting symp-
toms of disease, and even forecasting when animals are going to give birth are all
part of this. These gadgets improve overall herd management and contribute to
animal welfare by offering insights on animal well-being (Hassan et al. 2022a).
246 M. Naveed et al.

16.3 Predictive Modeling for Yield Optimization

Recently, AI has become a widely used technology in the agricultural sector. It is

employed to produce healthier crops, minimize losses due to insects, monitor soil
quality for optimal growing conditions, and analyze data for farmers. As a result, AI
activities have contributed to an increase in efficiency throughout the food supply
chain. Crop yield prediction and optimization through AI tools have become a
valuable technology for modern agriculture. AI tools collect weather data, historical
patterns, and growth models to provide more precise knowledge to farmers and
agricultural stakeholders.

16.3.1 DSSAT AI Growth Models

Agricultural decision-makers required excessive amounts of knowledge to under-

stand the possible outcomes of their resolution. This helps them organize plans and
policies to achieve their goals. DSSAT (decision support system for agrotechnology
transfer) model was developed to find crop production, sources, and risks linked to
crop production. This microcomputer software package has a stimulation model of
soil, weather database, crops, and soil strategy evolution programs integrated with
the main shell program. It helps decision-makers to make efficient decisions by
minimizing human resources used for complex analyses. It stimulates the growth
and development of plants.
The CERES-Wheat model has been designed for growth stimulation. This model
considers some key factors like plantation date, temperature, availability of water,
and nitrogen level to predict the yield of wheat and optimize cultivation ways.
CERES models simulated the yield of grain with different levels of root mean square
error. This model has also been used for maize and rice for growth stimulation.

16.3.2 Crop Enhancement Through Predictive Modeling

In precise agriculture, yield prediction remains a major concern. It is important for

decision-makers at the regional and national stages. A precise prediction model can
provide a helping hand for a farmer to make decisions about what and when to grow.
Various models have been introduced and validated as well. The crop yield depends
on a number of factors like climate, weather, soil conditions, fertilizer use, and
variety of seeds, so it requires more than one dataset for yield prediction. This shows
that yield prediction comprises many complex steps. Nowadays, actual yield can be
estimated through different models, but still there is a need for better performance
prediction models. Different features are used to build predictive models. In agri-
culture, crops with high market value are mostly preferred; the right choice of crop is
16 Sustainable AI-Driven Applications for Plant Care and Treatment 247

based on many factors. Data mining technologies play an important role in identi-
fying the best crop choice or hybrid seed choice. A crop with a high growth
probability can be determined by algorithms. Therefore, an AI forecasting approach
for crops addresses issues such as crop selection, necessitating the implementation of
an efficient decision-making method for optimal crop selection. Optimized decision-
making methodology can be used to increase crop yield. In this approach, datasets
are collected and then preprocessed to normalize the Z-score, and adaptive shearlet
method is applied to extract the features. The discrete hybrid Deep Belief Network
utilized the VGG NET algorithm to find out the best crop for growth. This approach
can be utilized in three different datasets. Many studies have revealed that models for
crop production prediction have been developed at the country level. A study was
conducted in Saudi Arabia in which AI was used for modeling and predicting yield
to increase food security (Arshad et al. 2022). They use artificial neural networks
(ANNs) and multilayer perceptron (MLP) models to precisely predict the yield of
crops, temperature, insecticides, and level of rainfall, depending on environmental
information. Temperature fluctuation, rainfall level, and insecticide effects on yield
can be successfully evaluated by using AI. Future values of different crop yields like
potatoes, rice, wheat, and sorghum can be predicted by using the MLP model. AI
approaches continually check soil nutrient levels and compare them with previous
data. It uses different datasets to analyze the environmental effects of dosage and a
suitable number of fertilizers for high-yield production. This will promote the
eco-friendly agriculture system. With the passage of time, there is a rapid change
in climate which has had a negative impact on agriculture, like difficulties in
deciding which crop should be sowed. AI makes it simple to understand the effects
of weather, seasonal sunshine, wind, and rain on the plant cycle. Farmers can get
information from weather predictions as they examine the plant seeding time (Arif
et al. 2021). Exact fertilization can be developed by clever seeding technology. AI
tools may help farmers with time management and increase quality. Nowadays,
drones collect data and monitor farm conditions. AI improves yield through soil and
crop monitoring by collecting the data through drones, the internet, and field satellite
image. These data are analyzed by AI tools to provide the most appropriate answer
(Waheed et al. 2023). ML helps the farmer to assess the data of crops, check weather
alteration, and efficiently manage the farm. AI also provides a sustainable way of
water usage to prevent water shortage. AI helps find the optimal place for a
plantation depending on geographical factors and soil chemistry. AI uses an ML
algorithm to check the seed quality by scanning it and comparing it with a healthy
one. With the help of this technology, farmers can make well-informed choices and
lower the likelihood of crop failure, which ultimately increases agricultural output
and profitability (Fig. 16.4).
248 M. Naveed et al.

Fig. 16.4 Management and monitoring of plant care through AI-based machine and deep learning

16.4 Disease Detection and Treatment

Any developing country can be regarded as having its foundation in the agricultural
sector. Therefore, farmers are required to have access to the most substantial
technologies and practices in order to get the highest yield possible from their
crops (Majeed et al. 2023). The diagnosis of various plant diseases is one of the
many farming aspects where information technology in agriculture has made tre-
mendous progress. Plant disease identification is one of the major challenges in
agriculture and has a considerable impact on crop production. Artificial intelligence
can be a major resource in combating crop diseases because of its capacity to identify
problems, diagnose diseases, and determine their causes as well as their potential
treatments (Mohanty et al. 2016). Computers are now able to learn without being
specifically programmed because of machine learning, which is conceptually similar
to how people learn. The use of algorithms that use machine learning to create
models out of known data is called learning or training (Zhou 2021).

16.4.1 Machine Learning for Crop Disease Management

For the detection and diagnosis of plant diseases, numerous studies have been
conducted using traditional machine learning techniques like random forests, artifi-
cial neural networks, fuzzy logic, K-means method, Support Vector Machines
(SVM), and Convolutional Neural Networks (CNNs). For instance, in a study of
tomato leaf disease classification by Tan et al. a total of 105 color features were
16 Sustainable AI-Driven Applications for Plant Care and Treatment 249

recovered using color histograms and color moment methods, while 52 texture
features were extracted using a gray level co-occurrence matrix (GLCM) and local
binary pattern (LBP) technique. For ten different forms of tomato leaf disease, the
efficacy of KNN, SVM, and random forest (RF) machine learning algorithms, as
well as AlexNet, VGG16, ResNet34, EfficientNet-b0, and MobileNetV2 architec-
tural deep learning methods, was investigated (Tan et al. 2021). Utilizing contem-
porary, innovative imaging and information technologies for disease detection is
essential to precision agriculture. These smart, non-intrusive techniques use almost
real-time observations to guard from crop damage brought by plant diseases.

16.4.2 Deep Learning Models to Extract Features

Another artificial intelligence approach is to utilize deep learning models, where

researchers initially used a combination of deep learning techniques and conven-
tional classifiers to classify plant diseases and pests. They extracted image features
using pertained CNN models, which they then fed into machine learning classifiers
like SVM to categorize (Yalcin and Razavi 2016). The efficiency of this strategy was
demonstrated by experimental data. Others developed CNN-based Meta topologies
with various feature extractors that could distinguish between healthy and diseased
plants. Another study by Hassan et al. used features generated from a deep CNN
model and input into an SVM classifier to identify and categorize several types of
rice illnesses with 97.5% accuracy (Hasan et al. 2019).

16.4.3 Two-Stage Detection Network (Faster R-CNN)

The Faster R-CNN two-stage detection network’s fundamental procedure involves

using the backbone network to retrieve the input image’s feature map. The anchor
box confidence is then calculated using RPN’s (Region Proposal Network) sugges-
tions. The network is given the feature map of the proposed area, and the initial
detection results are adjusted to produce the results for the placement and categori-
zation of the lesions. To increase the identification of plant diseases and pests,
common techniques improve on the backbone structure, feature map, anchor ratio,
ROIpooling, and loss function (Liu and Wang 2021). For the identification of tomato
diseases, beetroot leaf spot disease, rice blast, bacterial blight, sheath blight, and
grape leaf diseases, several researches have used Faster R-CNN, resulting in high
detection accuracy and speed improvements (Fuentes et al. 2019).
250 M. Naveed et al.

16.4.4 AI and Remote Sensing Applications for Plant Health

Monitoring and Treatment

Plant diseases can be caused by bacteria, fungi, nematodes, viruses, pests, weeds,
insects, and other pathogens. Farmers can identify the symptoms and signs of a
plant’s condition based on routine inspections. The following are some examples of
tools and techniques that can be harnessed by farmers to revolutionize crop disease
detection and management. An intelligent expert system with a fuzzy logic-based
decision-making algorithm has been developed to diagnose crop diseases (Hassan
et al. 2022b) (Table 16.3). Twenty-one prevalent diseases of wheat and cotton can be
detected by the method. Fuzzy logic has been used as the backend decision-making
engine for the proposed architecture, and a front-end Android application has been
created using the jFuzzyLite library. For the evaluation, various aggregation and
defuzzification techniques have been used (Toseef and Khan 2018). The most
effective mix of these strategies is advised for application at a practical scale based
on the experimental findings.

Table 16.3 IoT technologies, its applications, and advantages in smart agriculture
Sr. Application in Advantages in
no. IoT technology agriculture agriculture system References
1 Wire senseless network Sensors used in moni- Management and data Jondhale
(WSN): sensor nodes toring physical collection gathered et al.
having radio communi- parameters from sensors (2022)
cation capabilities
2 Cloud computing: Provides processing Management and easy Sadeeq
internet-based resources and data data collection gathered et al.
computing shared to computer on from cloud computing (2021)
demand services such as cloud
storage and fields maps
3 Protocols of communi- These protocols pro- Management and data de Kfouri
cation: it is the back- vide exchange of data collection of data col- et al.
bone of the IoT within the network in a lected from cloud com- (2019)
systems, which pro- number of exchange puting services,
vides connectivity data formats sensors, cloud storage,
etc.
4 Data analytics: the pro- Have access to multiple Market trends, correla- Hariri
cess of analyzing and forms of data types tions, preferences, and et al.
examining large other important (2019)
datasets information
5 Embedded system: a The system executes Profitability and sus- Pereira
system consists of both particular functions, tainability increase by et al.
software and hardware including management, the reduction of pro- (2019)
controlling, and moni- duction cost
toring of different
activities
16 Sustainable AI-Driven Applications for Plant Care and Treatment 251

16.4.4.1 Plantix Application

The ability to diagnose plant diseases is an important aspect of the Plantix app in
addition to certain other features. Berlin-based horticultural IT business PEAT
developed the Plantix app. It is used to locate deficiencies and flaws in soil. The
software analyzes plant images to identify a disease. A variety of these images are
saved in a sophisticated cell and are coordinated with the image in the worker’s
diagnosis (Fujita et al. 2016).

16.4.4.2 Stress Detection

Plant stress produced by conditions such as drought, nutrient inadequacy, or pollu-

tion may be assessed using remote sensing and AI technology. Timely intervention
based on stress indicators promotes optimal resource allocation, such as water and
fertilizers, ultimately enhancing plant health (Khang et al. 2023).

16.4.4.3 Yield Prediction

AI algorithms anticipate agricultural yields using historical data and remote sensing
inputs. This knowledge aids in efficient planning, allowing farmers to properly
handle harvest logistics and marketing plans (Wang and Gamon 2019).

16.4.4.4 Weed Control

Remote sensing, combined with AI, can differentiate between crops and weeds,
enabling more targeted herbicide delivery. This method reduces the environmental
effect while also lowering the expenditures associated with needless treatments
(Iqbal et al. 2020).

16.4.4.5 Nutrient Management

Precise fertilization and efficient nutrient management are essential in agriculture.

Recently, nutrient management decision support technologies have been developed
for modern agricultural systems using machine learning (ML). It is feasible to
combine and interpret several bits of knowledge that have never been investigated
for this purpose before the era of nutrient management (Ennaji et al. 2023). The
texture, soil type, organic matter, exchangeable nutrients, pH, total inorganic or
organic carbon, and the total soil capacity to hold exchangeable cations (CEC) are all
considered in nutrient management algorithms (Qin et al. 2018). The amount of
nitrogen lost during wet conditions significantly influences a soil’s ability to retain
252 M. Naveed et al.

water beyond its saturation point, typically corresponding to the depth of the water
table. Deep learning was employed in a study (Condori et al. 2017) to identify maize
with insufficient nitrogen levels. Four CNN models were pre-trained for this purpose
using applied transfer learning, with growth stages (V4, V6, and R1) taken into
consideration. Following digitization of the leaf samples, 384 pictures representing
the 16 nitrogen treatments taken into consideration by the study were produced.
Regarding each growth stage, the results revealed a considerable variation. How-
ever, when compared to conventional text-based methods, CNNs formed using RGB
images produced excellent results (on average 61.8%) as opposed to these methods
(Bashir et al. 2023).

16.4.4.6 Smart Irrigation

Considering the crop security and satisfaction of farmers, accurate predictions of

irrigation needs and crop yields are essential. The planned predictions render a
substantial contribution to raising agricultural yields and reducing production
costs. The traditional methods are unable to meet the expanding demand as they
use pesticides in greater quantity, which affects the oil content. Challenges in
agriculture, such as crop pests, limited storage capacity, pesticide regulation, irriga-
tion management, and water resource control, are a few fields that cause agricultural
difficulties which can be resolved by artificial intelligence (Sinwar et al. 2020).
Therefore, it is essential to use digital technologies at several points along the
agricultural supply chain, including the automation of farm equipment, use of
instruments and information from satellites from a distance, machine learning and
artificial intelligence for more effective crop monitoring, water management, and
traceability of agricultural food products (Majeed et al. 2022c). The sensors are
linked to economical Arduino-based devices to store the collected data and run
analysis algorithms to forecast the crop’s water needs at a specific moment. Smart
irrigation systems ensure precise delivery of the required amount of water to each
plant, thereby ensuring their overall health. Regular monitoring is essential for
sprinklers and drip modules to prevent any potential malfunctions. The
non-uniformity of the land and crop types is the fundamental challenge in the
construction of a smart irrigation system. Using machine learning approaches
(PLSR and ANFIS) and sensors, Navarro-Helln et al. proposed a smart irrigation
decision support system (SIDSS) for calculating the weekly irrigation needs. The
technology was tried on citrus plants in Spain’s southeast. The system’s performance
was evaluated using expert human assessment (Hinnell et al. 2010). Their approach
revealed a decrease of 22% in irrigation needs relative to weather values from the
prior year (i.e., irrigation needs for 2015 have been projected using data from 2014).
Various researchers have demonstrated how automation and AI could improve
yield growth. There is a dire need to install smart irrigation systems that can irrigate
larger areas with minimal water use in order to cope up with incurring water scarcity.
Gathering information regarding the level of moisture in the soil, the water content of
plants, the atmosphere’s humidity, the temperature, etc., is necessary for smart
16 Sustainable AI-Driven Applications for Plant Care and Treatment 253

irrigation. A Neural Network-based irrigation system (Neuro-Drip) has been

presented by Hinnell et al. Sensors for measuring humidity, temperature, and soil
moisture can all be used to collect data (Hinnell et al. 2010).

16.4.4.7 Climate Resilience

Remote sensing and artificial intelligence (AI) help to improve climate resilience by
detecting changes in plant patterns over time. These tools aid in the early detection of
alterations induced by climate change, hence assisting adaption methods (Khan et al.
2022a). In the near future it is believed that there will be a rise in the use of artificial
intelligence models around the world given the recent decades’ tremendous advance-
ment of the field. Applications and integration have become more and more impres-
sive in regard to agricultural developments including smart irrigation, soil and
nutrient management, and yield enhancement. When applied correctly for
decision-making or support, ML algorithms have great potential in any field of
work. However, due to complicated definition concepts and computational pro-
cedures, the application of AI in this subject has been restricted despite its potential.

16.5 Conclusion

Before the advent of artificial intelligence (AI), agricultural systems lacked effective
irrigation and proper monitoring of plant diseases, especially considering challenges
related to varying plant heights and extreme weather conditions. These problems
have been resolved with the help of AI technology. Now, soil moisture levels can be
detected via remote sensors, and performance can be enhanced through automated
irrigation systems aided by GPS. Agriculture-related applications of machine learn-
ing and deep learning are becoming increasingly prevalent. Precision farming
employing artificial intelligence in agriculture increases crop output and quality
while utilizing scarce resources. Furthermore, remote sensing employs cutting-
edge techniques that help ranchers keep an eye on their crops without having to
physically monitor the land. Many businesses today anticipate the expansion of
agriculture with AI support. Remote sensing and artificial intelligence reclassify the
traditional farming model by redefining common agricultural trends. Several com-
mercially available goods are becoming increasingly well acknowledged as distinct
over time due to the implementation of artificial intelligence revolutionary applica-
tions that can help identify and manage crop diseases to increase crop yield and
quality. The use of artificial intelligence in agriculture is rapidly developing with
numerous complex tactics and is predicted to bring about an extensive change in the
near future.
254 M. Naveed et al.

References

Agrawal R, Wankhede VA, Kumar A, Luthra S, Majumdar A, Kazancoglu Y (2021) An explor-

atory state-of-the-art review of artificial intelligence applications in circular economy using
structural topic modeling. Oper Manag Res 15:609–626
Alzubaidi L, Zhang J, Humaidi AJ, Al-Dujaili A, Duan Y, Al-Shamma O, Farhan L (2021) Review
of deep learning: concepts, CNN architectures, challenges, applications, future directions. J Big
Data 8:1–74
Arif U, Bhatti KH, Ajaib M, Wagay NA, Majeed M, Zeb J, Hameed A, Kiani J (2021) Ethnobo-
tanical indigenous knowledge of Tehsil Charhoi, District Kotli, Azad Jammu and Kashmir,
Pakistan. Ethnobot Res Appl 22:1–24. https://doi.org/10.32859/ERA.22.50.1-24
Arshad F, Waheed M, Harun N, Fatima K, Khan BA, Fatima K, Abbas Z, Jabeen S, Majeed M
(2022) Indigenous farmer perception about fodder and foraging species of semi-arid lowlands of
Pakistan: a case study of District Kasur, Pakistan. Taiwania 67:4
Arzani A, Ashraf M (2017) Cultivated ancient wheats (Triticum spp.): a potential source of health-
beneficial food products. Compr Rev Food Sci Food Saf 16(3):477–488
Bashir SM, Altaf M, Hussain T, Umair M, Majeed M, Mangrio WM, Khan AM, Gulshan AB,
Hamed MH, Ashraf S, Amjad MS, Bussmann RW, Abbasi AM, Casini R, Alataway A, Dewidar
AZ, Al-Yafrsi M, Amin MH, Elansary HO (2023) Vernacular taxonomy, cultural and
ethnopharmacological applications of avian and mammalian species in the vicinity of Ayubia
National Park, Himalayan Region. Biology 12:4
Ben Ayed R, Hanana M (2021) Artificial intelligence to improve the food and agriculture sector. J
Food Qual 2021:1–7
Budiharto W, Chowanda A, Gunawan AAS, Irwansyah E, Suroso JS (2019) A review and progress
of research on autonomous drone in agriculture, delivering items and geographical information
systems (GIS). In: Paper presented at the 2019 2nd world symposium on communication
engineering (WSCE)
Condori RHM, Romualdo LM, Bruno OM, de Cerqueira Luz PH (2017) Comparison between
traditional texture methods and deep learning descriptors for detection of nitrogen deficiency in
maize crops. In: Paper presented at the 2017 workshop of computer vision (WVC)
Coulibaly S, Kamsu-Foguem B, Kamissoko D, Traore D (2022) Deep learning for precision
agriculture: a bibliometric analysis. Intell Syst Appl 16:200102
Daponte P, De Vito L, Glielmo L, Iannelli L, Liuzza D, Picariello F, Silano G (2019) A review on
the use of drones for precision agriculture. In: Paper presented at the IOP conference series:
earth and environmental science
de Kfouri GO, Gonçalves DG, Dutra BV, de Alencastro JF, de Caldas Filho FL, e Martins LM, de
Sousa Jr RT (2019) Design of a distributed HIDS for IoT backbone components. In: Paper
presented at the FedCSIS (Communication Papers)
Dutta PK, Mitra S (2021) Application of agricultural drones and IoT to understand food supply
chain during post COVID-19, pp 67–87
Ennaji O, Vergütz L, El Allali A (2023) Machine learning in nutrient management: a review. Artif
Intell Agric 9:1–11. https://doi.org/10.1016/j.aiia.2023.06.001
Fuentes A, Yoon S, Kim SC, Park DS (2019) A robust deep-learning-based detector for real-time
tomato plant diseases and pests recognition. Sens Agric 1(17):153
Fujita E, Kawasaki Y, Uga H, Kagiwada S, Iyatomi H (2016) Basic investigation on a robust and
practical plant diagnostic system. In: Paper presented at the 2016 15th IEEE international
conference on machine learning and applications (ICMLA)
Gardezi M, Joshi B, Rizzo DM, Ryan M, Prutzer E, Brugler S, Dadkhah A (2023) Artificial
intelligence in farming: challenges and opportunities for building trust. Agron J 2023:21353
Haq SM, Yaqoob U, Majeed M, Amjad MS, Hassan M, Ahmad R, Morales-de la Nuez A (2022)
Quantitative ethnoveterinary study on plant resource utilization by indigenous communities in
high-altitude regions. Front Vet Sci 9:94404
16 Sustainable AI-Driven Applications for Plant Care and Treatment 255

Hariri RH, Fredericks EM, Bowers KM (2019) Uncertainty in big data analytics: survey, opportu-
nities, and challenges. J Big Data 6(1):1–16
Hasan MJ, Mahbub S, Alom MS, Nasim MA (2019) Rice disease identification and classification
by integrating support vector machine with deep convolutional neural network. In: Paper
presented at the 2019 1st international conference on advances in science, engineering and
robotics technology (ICASERT)
Hassan M, Haq SM, Ahmad R, Majeed M, Sahito HA, Shirani M, Mubeen I, Aziz MA, Pieroni A,
Bussmann RW, Alataway A, Dewidar AZ, Al-Yafrsi M, Elansary HO, Yessoufou K (2022a)
Traditional use of wild and domestic fauna among different ethnic groups in the Western
Himalayas? Cross cultural analysis. Animals 12:17
Hassan M, Haq SM, Majeed M, Umair M, Sahito HA, Shirani M, Waheed M, Aziz R, Ahmad R,
Bussmann RW, Alataway A, Dewidar AZ, El-Abedin TKZ, Al-Yafrsi M, Elansary HO,
Yessoufou K (2022b) Traditional food and medicine: ethno-traditional usage of fish fauna
across the valley of Kashmir: a Western Himalayan region. Diversity 14:6
Hinnell A, Lazarovitch N, Furman A, Poulton M, Warrick A (2010) Neuro-drip: estimation of
subsurface wetting patterns for drip irrigation using neural networks. Irrig Sci 28:535–544
Iqbal N, Khaliq A, Cheema ZA (2020) Weed control through allelopathic crop water extracts and
S-metolachlor in cotton. Inform Process Agric 7(1):165–172
Jackulin C, Murugavalli S (2022) A comprehensive review on detection of plant disease using
machine learning and deep learning approaches. Meas Sens 24:100441
Jadhav P, Kachave V, Mane A, Joshi K (2023) Crop detection using satellite image processing.
I-Managers J Image Process 10(2):50
Jondhale SR, Maheswar R, Lloret J, Jondhale SR, Maheswar R, Lloret J (2022) Fundamentals of
wireless sensor networks. Received signal strength based target localization and tracking using
wireless sensor networks. Springer, pp 1–19
Karetsos S, Costopoulou C, Sideridis A (2014) Developing a smartphone app for m-government in
agriculture. J Agric Inform 5(1):129
Karlen D, Andrews S, Doran J (2001) Soil quality: current concepts and applications. Elsevier
Kashyap B, Kumar RJ (2021) Sensing methodologies in agriculture for soil moisture and nutrient
monitoring. IEEE Access 9:14095–14121
Khan A, Shahriyar AK (2023) Optimizing onion crop management: a smart agriculture framework
with IoT sensors and cloud technology. Appl Res Artif Intell Cloud Comput 6(1):49–67
Khan AM, Li Q, Saqib Z, Khan N, Habib T, Khalid N, Majeed M, Tariq A (2022a) MaxEnt
modelling and impact of climate change on habitat suitability variations of economically
important Chilgoza pine (Pinus gerardiana Wall.) in South Asia. Forests 13:5
Khan MHU, Wang S, Wang J, Ahmar S, Saeed S, Khan SU, Feng X (2022b) Applications of
artificial intelligence in climate-resilient smart-crop breeding. Int J Mol Sci 23(19):11156
Khang A, Rath KC, Panda S, Sree PK, Panda SK (2023) Revolutionizing agriculture: exploring
advanced technologies for plant protection in the agriculture sector. In: Handbook of research on
AI-equipped IoT applications in high-tech agriculture. IGI Global, pp 1–22
Khoja AA, Haq SM, Majeed M, Hassan M, Waheed M, Yaqoob U, Bussmann RW, Alataway A,
Dewidar AZ, Al-Yafrsi M, Elansary HO, Yessoufou K, Zaman W (2022) Diversity, ecological
and traditional knowledge of pteridophytes in the Western Himalayas. Diversity 14:8
Kim H, Sudduth K, Hummel JW, Drummond S (2013) Validation testing of a soil macronutrient
sensing system. Trans ASABE 56(1):23–31
Kuppusamy P, Shanmugananthan S, Tomar P (2021) Emerging technological model to sustainable
agriculture. In: Artificial intelligence and IoT-based technologies for sustainable farming and
smart agriculture. IGI Global, pp 101–122
Lamb DW, Frazier P, Adams P (2008) Improving pathways to adoption: putting the right P’s in
precision agriculture. Comput Electron Agric 61(1):4–9
Liang C, Shah T (2023) IoT in agriculture: the future of precision monitoring and data-driven
farming. Eigenpub Rev Sci Technol 7(1):85–104
256 M. Naveed et al.

Linaza MT, Posada J, Bund J, Eisert P, Quartulli M, Döllner J, Moysiadis T (2021) Data-driven
artificial intelligence applications for sustainable precision agriculture. Agronomy 11(6):1227
Liu J, Wang X (2021) Plant diseases and pests detection based on deep learning: a review. Plant
Methods 17:1–18
Maddikunta PKR, Hakak S, Alazab M, Bhattacharya S, Gadekallu TR, Khan WZ, Pham Q-V
(2021) Unmanned aerial vehicles in smart agriculture: applications, requirements, and chal-
lenges. IEEE Sens J 21(16):17608–17619
Majeed M, Bhatti KH, Amjad MS, Abbasi AM, Rashid A, Nawaz F , Ahmad KS (2020a) Ethno-
veterinary practices of Poaceae taxa in Punjab, Pakistan
Majeed M, Bhatti KH, Amjad MS, Abbasi M, Id RWB, Nawaz F, Rashid A, Mehmood A, Id MM,
Khan WM, Id SA (2020b) Ethno-veterinary uses of Poaceae in Punjab, Pakistan. PLoS One 15:
e0241705. https://doi.org/10.1371/journal.pone.0241705
Majeed M, Bhatti KH, Pieroni A, Sukand R, Bussmann RW, Khan AM, Chaudhari SK, Aziz MA,
Amjad MS (2021a) Gathered wild food plants among diverse religious groups in Jhelum
District, Punjab, Pakistan. Foods 10:3
Majeed M, Tariq A, Anwar MM, Khan AM, Arshad F, Mumtaz F, Farhan M, Zhang L, Zafar A,
Aziz M, Abbasi S, Rahman G, Hussain S, Waheed M, Fatima K, Shaukat S (2021b) Monitoring
of land use? And cover change and potential causal factors of climate change in Jhelum District,
Punjab, Pakistan, through GIS and multi-temporal satellite data. Land 10:10
Majeed M, Khan AM, Habib T, Anwar MM, Sahito HA, Khan N, Ali K (2022a) Vegetation
analysis and environmental indicators of an arid tropical forest ecosystem of Pakistan. Ecol
Indic 142:109291. https://doi.org/10.1016/j.ecolind.2022.109291
Majeed M, Lu L, Haq SM, Waheed M, Sahito HA, Fatima S, Aziz R, Bussmann RW, Tariq A,
Ullah I, Aslam M (2022b) Spatiotemporal distribution patterns of climbers along an abiotic
gradient in Jhelum District, Punjab, Pakistan. Forests 13:8
Majeed M, Tariq A, Haq SM, Waheed M, Anwar MM, Li Q, Aslam M, Abbasi S, Mousa BG, Jamil
A (2022c) A detailed ecological exploration of the distribution patterns of wild Poaceae from the
Jhelum District (Punjab), Pakistan. Sustainability (Switzerland) 14:7
Majeed M, Lu L, Anwar MM, Tariq A, Qin S, El-Hefnawy ME, El-Sharnouby M, Li Q, Alasmari A
(2023) Prediction of flash flood susceptibility using integrating analytic hierarchy process
(AHP) and frequency ratio (FR) algorithms. Front Environ Sci 10:1037547. https://doi.org/10.
3389/fenvs.2022.1037547
Martinelli F, Scalenghe R, Davino S, Panno S, Scuderi G, Ruisi P, Goulart LR (2015) Advanced
methods of plant disease detection. A review. Agron Sustain Dev 35:1–25
Mathenge M, Sonneveld BG, Broerse JE (2022) Application of GIS in agriculture in promoting
evidence-informed decision making for improving agriculture sustainability: a systematic
review. Sustainability 14(16):9974
Meng T, Jing X, Yan Z, Pedrycz W (2020) A survey on machine learning for data fusion. Inf Fusion
57:115–129
Minkoff-Zern L-A (2019) The new American farmer: immigration, race, and the struggle for
sustainability. MIT Press
Mohamed ES, Belal A, Abd-Elmabod SK, El-Shirbeny MA, Gad A, Zahran MB (2021) Smart
farming for improving agricultural management. Egypt J Remote Sens Space Sci 24(3):971–981
Mohanty SP, Hughes DP, Salathé M (2016) Using deep learning for image-based plant disease
detection. Front Plant Sci 7:1419
Mosa KA, Ismail A, Helmy M, Mosa KA, Ismail A, Helmy M (2017) Introduction to plant stresses.
Springer, pp 1–19
Nayfeh M, Li Y, Al Shamaileh K, Devabhaktuni V, Kaabouch N (2023) Machine learning
modeling of GPS features with applications to UAV location spoofing detection and classifica-
tion. Comput Secur 126:103085
Pawar A, Deosarkar SJ (2023) IoT-based smart agriculture: an exhaustive study. Springer, pp 1–14
16 Sustainable AI-Driven Applications for Plant Care and Treatment 257

Peixoto DS, Silva BM, de Oliveira GC, Moreira SG, da Silva F, Curi N (2019) A soil compaction
diagnosis method for occasional tillage recommendation under continuous no tillage system in
Brazil. Soil Tillage Res 194:104307
Pereira RI, Jucá SC, Carvalho PC (2019) IoT embedded systems network and sensors signal
conditioning applied to decentralized photovoltaic plants. Measurement 142:195–212
Qin Z, Myers DB, Ransom CJ, Kitchen NR, Liang SZ, Camberato JJ, Franzen DW (2018)
Application of machine learning methodologies for predicting corn economic optimal nitrogen
rate. Agron J 110(6):2596–2607
Raj EFI, Appadurai M, Athiappan K (2022) Precision farming in modern agriculture. In: Smart
agriculture automation using advanced technologies: data analytics and machine learning, cloud
architecture, automation and IoT. Springer, pp 61–87
Rakhmatulin I, Kamilaris A, Andreasen C (2021) Deep neural networks to detect weeds from crops
in agricultural environments in real-time: a review. Remote Sens 13(21):4486
Rashid S, Getnet K, Lemma S (2019) Maize value chain potential in Ethiopia: constraints and
opportunities for enhancing the system. Gates Open Res 3(354):354
Rathore B (2017) Sustainable fashion marketing: AI-powered solutions for effective promotions.
Int J New Media Stud Int Peer Rev Sch Indexed J 4(2):70–80
Rehman A, Saba T, Kashif M, Fati SM, Bahaj SA, Chaudhry H (2022) A revisit of internet of things
technologies for monitoring and control strategies in smart agriculture. Agronomy 12(1):127
Sadeeq MM, Abdulkareem NM, Zeebaree SR, Ahmed DM, Sami AS, Zebari RR (2021) IoT and
Cloud computing issues, challenges and opportunities: a review. Qubahan Acad J 1(2):1–7
Sekaran U, Lai L, Ussiri DA, Kumar S, Clay S (2021) Role of integrated crop-livestock systems in
improving agriculture production and addressing food security—a review. J Agric Food Agric
5:100190
Singh RK, Berkvens R, Weyn M (2021) AgriFusion: an architecture for IoT and emerging
technologies based on a precision agriculture survey. IEEE Access 9:136253–136283
Sinha D, Maurya AK, Abdi G, Majeed M, Agarwal R, Mukherjee R, Ganguly S, Aziz R, Bhatia M,
Majgaonkar A, Seal S, Das M, Banerjee S, Chowdhury S, Adeyemi SB, Chen JT (2023)
Integrated genomic selection for accelerating breeding programs of climate-smart cereals.
Genes 14(7):1484
Sinwar D, Dhaka VS, Sharma MK, Rani G (2020) AI-based yield prediction and smart irrigation.
Internet Things Anal Agric 2:155–180
Sudha T (2023) Artificial intelligence in human resource management. Future trends, break-
throughs and innovation in HRM
Sun Y, Ding W, Shu L, Li K, Zhang Y, Zhou Z, Han G (2021) On enabling mobile crowd sensing
for data collection in smart agriculture: a vision. IEEE Syst J 16(1):132–143
Tan L, Lu J, Jiang H (2021) Tomato leaf diseases classification based on leaf images: a comparison
between classical machine learning and deep learning methods. AgriEngineering 3(3):542–558
Tariq A, Mumtaz F, Majeed M, Zeng X (2023) Spatio-temporal assessment of land use land cover
based on trajectories and cellular automata Markov modelling and its impact on land surface
temperature of Lahore district Pakistan. Environ Monit Assess 195:1
Tassadduq SS, Akhtar S, Waheed M, Bangash N, Nayab DE, Majeed M, Abbasi S, Muhammad M,
Alataway A, Dewidar AZ, Elansary HO, Yessoufou K (2022) Ecological distribution patterns of
wild grasses and abiotic factors. Sustainability (Switzerland) 14:18
Tiwari AK, Mishra H, Nishad DC, Pandey A (2023) Sustainable water management in agriculture:
irrigation techniques and water conservation. Dr. Ajay B. Jadhao 53
Toseef M, Khan MJ (2018) An intelligent mobile application for diagnosis of crop diseases in
Pakistan using fuzzy inference system. Comput Electron Agric 153:1–11. https://doi.org/10.
1016/j.compag.2018.07.034
Tugrul B, Elfatimi E, Eryigit R (2022) Convolutional neural networks in detection of plant leaf
diseases: a review. Agriculture 12(8):1192
258 M. Naveed et al.

Ullah I, Aslam B, Shah SHIA, Tariq A, Qin S, Majeed M, Havenith HB (2022) An integrated
approach of machine learning, remote sensing, and GIS data for the landslide susceptibility
mapping. Land 11:8
Van Esse HP, Reuber TL, van der Does D (2020) Genetic modification to improve disease
resistance in crops. New Phytol 225(1):70–86
Waheed M, Arshad F, Majeed M, Fatima S, Mukhtar N, Aziz R, Mangrio WM, Almohamad H,
Dughairi AA, Al-Mutiry M, Abdo HG (2022) Community structure and distribution pattern of
woody vegetation in response to soil properties in semi-arid lowland District Kasur Punjab,
Pakistan. Land 11:12
Waheed M, Arshad F, Majeed M, Haq SM, Aziz R, Bussmann RW, Ali K, Subhan F, Jones DA,
Zaitouny A (2023) Potential distribution of a noxious weed (Solanum viarum Du-nal), current
status, and future invasion risk based on MaxEnt modeling. In: Geology, ecology, and land-
scapes. Taylor & Francis, p 1. https://doi.org/10.1080/24749508.2023.2179752
Wai MHX, Huong A, Ngu X (2021) Soil moisture level prediction using optical technique and
artificial neural network. Int J Electr Comput Eng 11(2):1752–1760
Wang R, Gamon JA (2019) Remote sensing of terrestrial plant biodiversity. Remote Sens Environ
231:111218
Xiao L, Bahri Y, Sohl-Dickstein J, Schoenholz S, Pennington J (2018) Dynamical isometry and a
mean field theory of CNNS: how to train 10,000-layer vanilla convolutional neural networks. In:
Paper presented at the international conference on machine learning
Yalcin H, Razavi S (2016) Plant classification using convolutional neural networks. In: Paper
presented at the 2016 fifth international conference on Agro-Geoinformatics (Agro-
Geoinformatics)
Zhang J, Yin Z, Chen P, Nichele SJIF (2020) Emotion recognition using multi-modal data and
machine learning techniques: a tutorial and review. Inf Fusion 59:103–126
Zhou Z-H (2021) Machine learning. Springer Nature
Chapter 17
Use Cases and Future Aspects of Intelligent
Techniques in Microbial Data Analysis

Muhammad Naveed , Zaibun-nisa Memon, Muhammad Abdullah,

Syeda Izma Makhdoom, Arooj Azeem, Sarmad Mehmood,
Maida Salahuddin, Zeerwah Rajpoot, and Muhammad Majeed

Abstract Microbes, including bacteria, archaea, fungi, and viruses, are fundamental
to our ecosystems, health, and industries. Microbial data analysis has become
indispensable in understanding their roles and interactions. In this era of big data,
advanced techniques, such as high-throughput sequencing, metagenomics, and
bioinformatics, have accelerated microbial research. This chapter explores the sig-
nificance of intelligent techniques, particularly machine learning and artificial intel-
ligence, in revolutionizing microbial data analysis. The aim of this chapter is to
showcase the pivotal role of intelligent techniques in microbial data analysis across
diverse domains, from ecology and public health to biotechnology. We delve into
case studies that highlight the practical applications of these techniques and the
transformative impact they have had on microbial research. Several case studies are
presented, illustrating the applications of intelligent techniques in microbial
research. These include predicting disease risk through gut microbiome analysis,
antibiotic resistance prediction, environmental microbiology for ecosystem manage-
ment, bioprocess optimization in biotechnology, and AI-powered antibiotic suscep-
tibility testing. Each case study demonstrates how intelligent techniques have
enhanced data analysis, prediction, and decision-making in their respective domains.

M. Naveed · S. I. Makhdoom · A. Azeem · S. Mehmood · M. Salahuddin · Z. Rajpoot

Department of Biotechnology, Faculty of Science and Technology, University of Central
Punjab, Lahore, Punjab, Pakistan
e-mail: [email protected]
Z.-n. Memon
Department of Zoology, Shah Abdul Latif University, Khairpur Mirs, Sindh, Pakistan
e-mail: [email protected]
M. Abdullah
Biodiversity Park, Cholistan Institute of Desert Studies (CIDS), The Islamia University of
Bahawalpur, Bahawalpur, Pakistan
e-mail: [email protected]
M. Majeed (✉)
Department of Botany, University of Gujrat, Gujrat, Pakistan
e-mail: [email protected]; https://www.researchgate.net/profile/Muhammad-Majeed-18

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 259
A. Khamparia et al. (eds.), Microbial Data Intelligence and Computational
Techniques for Sustainable Computing, Microorganisms for Sustainability 47,
https://doi.org/10.1007/978-981-99-9621-6_17
260 M. Naveed et al.

Microbial data analysis, driven by intelligent techniques, has ushered in a new era of
understanding and harnessing the power of microorganisms. The future of microbial
data analysis holds immense promise, with emerging trends including the integration
of omics data, explainable AI, personalized microbiome analysis, and the develop-
ment of ethical and regulatory frameworks. Collaborative research and data sharing
are expected to further advance our understanding of the microbial world, offering
solutions to some of the most critical challenges of our time.

Keywords Microbiome · Intelligent · Data analysis · Bioprocess · Optimization ·

Personalized

17.1 Introduction

Microbes, the small, frequently invisible organisms that live everywhere on our
globe, have a significant impact on how the world is now. Numerous ecosystems,
human health, and industrial processes depend on these microscopic organisms,
which also include bacteria, archaea, fungus, viruses, and other microbes. Data
analysis has become a significant tool in the understanding and use of microorgan-
isms. The significance of microbial data analysis in the age of big data cannot be
emphasized (Majeed et al. 2021a).
In order to research microbial communities, their roles, and interactions, scientists
use a wide variety of tools and procedures, which collectively make up microbial
data analysis. Thanks to technological advancements like high-throughput sequenc-
ing, metagenomics, metatranscriptomics, and bioinformatics, this discipline has
quickly developed. Drug development requires examination of microbial data,
especially to comprehend how the microbiome affects drug metabolism and effec-
tiveness (Tirtawijaya et al. 2018).
Ecological processes can be better understood by examining the diversity and
distribution of microbes, which are essential parts of ecosystems. Analysis of
microbial data assists in tracking microbial community changes brought by envi-
ronmental disturbances and offers insightful information for conservation efforts
(Khoja et al. 2022). Our bodies are home to microbes, and they have a significant
impact on our health. Understanding the human microbiome, which is linked to
several health issues such as obesity, autoimmune illnesses, and mental health
disorders, requires analysis of microbial data. Microbial data analysis is also neces-
sary for probiotic creation and personalized treatment (Belkaid and Hand 2014).
Analyzing microbial data is essential for monitoring and controlling infectious
disease epidemics. It helps in tracking the transmission of pathogens, locating the
origin of infections, and creating efficient therapies. Microbes are sensitive environ-
mental change indicators. Analysis of microbial data can be used to determine how
pollution, land use, and climate change affect microbial communities, assisting in the
development of environmental policies (Haberbeck et al. 2017).
Science continues to learn new things about microbes every day. Through the use
of data analysis, scientists are now able to discover new species, examine their
17 Use Cases and Future Aspects of Intelligent Techniques in Microbial. . . 261

genetic makeup, and learn more about their special skills. Given that many medica-
tions are made from naturally occurring chemicals generated by bacteria, examina-
tion of microbial data is crucial in the drug development process. In this research,
mining microbial genomes for new therapeutic compounds is a promising strategy
(Arif et al. 2021).
Microbial data analysis is a cornerstone of contemporary science and technology,
with applications in everything from industry and environmental protection to
ecology and medicine. The enormous volume of data produced by microbial inves-
tigations offers a wealth of knowledge just waiting to be discovered. Our capacity to
harness the power of these microscopic creatures for the benefit of people and the
environment grows as our knowledge of the microbial world expands. Therefore, the
value of microbial data analysis cannot be overstated, and further developments in
this area hold the promise of tackling some of the most important problems of our
day (Waheed et al. 2023).

17.1.1 Role of Intelligent Techniques

The subject of microbial data analysis has undergone a revolution thanks to intelli-
gent techniques, which now provide strong tools and procedures to solve the riddles
of microbial communities. It is becoming more and more difficult for researchers to
manually extract significant insights from the huge information generated by high-
throughput sequencing, metagenomics, and other cutting-edge techniques used in
microbial data analysis. This is where intelligent methods have intervened to
improve the effectiveness and precision of microbiological data analysis, such as
machine learning and artificial intelligence (AI; Ullah et al. 2022).
For instance, machine learning algorithms have been crucial in taxonomy classi-
fication, enabling researchers to quickly detect and categorize microorganisms.
These methods are able to identify minor patterns and similarities across microbial
sequences that may be difficult for people to notice manually by training algorithms
on labeled datasets. This greatly expedites the identification of existing species and
even makes it possible to find new ones (Qu et al. 2019).
Additionally, intelligent approaches excel at predictive modeling, enabling
researchers to foresee microbial interactions and behaviors. For instance, in ecolog-
ical studies, machine learning models can forecast how changes in environmental
conditions would affect the dynamics of microbial communities, assisting with the
preservation of biodiversity and the evaluation of the health of ecosystems (Lucas
2020). Predictive modeling based on microbiome data in the healthcare sector assists
in predicting disease risks, treatment outcomes, and tailored medicine interventions
based on a person’s microbial composition (Majeed et al. 2020a).
When tackling the enormous complexity of microbiological data, artificial intel-
ligence (AI) is also a factor. With the help of AI-driven data integration technologies,
information from diverse sources, including environmental metadata,
metatranscriptomics, and genomic data, may be unified to give a comprehensive
262 M. Naveed et al.

picture of microbial communities. Understanding how bacteria react to environmen-

tal changes is vital in areas like bioremediation and agriculture and is made possible
by this integrated approach (Sahayasheela et al. 2022).

17.2 Microbial Data Analysis

Microbes, the microscopic organisms that inhabit various environments, play a

crucial role as sensitive indicators of ecosystem changes (Falkowski et al. 2008).
Their incredible diversity and adaptability make them valuable sensors for detecting
shifts in environmental conditions. Microbes can swiftly respond to alterations in
chemical and physical surroundings, providing researchers with precise data. In this
commentary, we explore the immense potential of microbial data analysis, drawing
an intriguing parallel with the olfactory system. Just as populations of sensory
neurons in our noses respond to specific odors, microbial communities can be
considered as nature’s sensors, capable of reporting environmental changes with
remarkable fidelity (Haq et al. 2022).
Moreover, this commentary emphasizes the role of machine learning in unlocking
the insights hidden within microbial data. Machine learning techniques hold the
promise of deciphering microbial responses to environmental fluctuations, enhanc-
ing our understanding of ecosystems and aiding in the detection of significant
environmental changes (Bzdok and Ioannidis 2019).

17.2.1 Using AI for Public Health Surveillance

The twenty-first century has witnessed an unprecedented surge in data collected

from public health surveillance. This exponential growth can be attributed to
advancements in information technology and the establishment of comprehensive
data collection systems (Bates et al. 2014). In this context, the utilization of artificial
intelligence (AI) emerges as a potent solution for addressing the challenges posed by
this deluge of data. AI offers the potential to revolutionize disease monitoring,
making it more reliable and efficient (Majeed et al. 2021b).
By harnessing AI, we can not only manage the vast volumes of infectious disease
data effectively but also enable proactive surveillance and projection of disease
outbreaks (Kamel Boulos et al. 2011). Robust data management platforms, in
conjunction with AI methods, have the potential to provide valuable support to
government agencies, healthcare providers, and medical professionals in their efforts
to respond to diseases swiftly and with precision (Khan et al. 2022).
17 Use Cases and Future Aspects of Intelligent Techniques in Microbial. . . 263

17.2.2 Microbial Networks: Nature’s Intelligent Systems

Living organisms, from the tiniest microbes to complex multicellular organisms like
humans, persist and thrive due to intricate interactions within dynamic, environment-
responsive networks (Jeong et al. 2000). These networks operate across multiple
scales and dimensions, facilitating communication and adaptation. While we often
associate “intelligence” with the human brain, this commentary suggests that similar
traits may exist in the intricate macromolecular networks of microbes (Fig. 17.1).
Despite lacking brains, microbes exhibit intelligent characteristics such as mem-
ory, anticipation, adaptation, and reflection. The commentary explores how micro-
bial networks demonstrate these traits and how they may confer advantages in their
respective environments. It proposes that if we broaden our definition of “intelli-
gence” to exclude terms like “human” and “brain,” all forms of life, including
microbes, display some or all characteristics consistent with intelligence (Guttenberg
et al. 2020). Additionally, it highlights how advances in genome-wide data analysis
offer valuable insights into microbial intelligence, which may pave the way for
synthetic biology applications and the creation of intelligent molecular networks.

(A) Microbial Data Analysis (B) Biases

1.Compositional bias
Bacteria Fungi Virome Absolute Abundance Relative Abundance
Control Treatment Control Treatment

Bacteria A

Bacteria B

2.Sparse count matrix

16s rRNA amplification

(Bacteria)
0 50 100

18S rRNA amplification 3. Spurious indirect edges

(Fungi)
Indirect Edge
Taxa 1 Taxa 2

Shotgun Metagenomics OTU/ASV Relative

High-throughput
(Bacteria, Fungi, clustering/ Abundance
sequencing
Virome) Binning
(shotgun)
Environmental Factor/Taxa3

Fig. 17.1 Overview of (a) taxonomic profiling and (b) biases in microbial co-occurrence networks
(Matchado et al. 2021)
264 M. Naveed et al.

17.2.3 Microbial Data Analysis for Disease Association

The human microbiota, comprising a diverse array of microorganisms, exhibits

remarkable variability within individuals. These microorganisms residing in the
human body play a pivotal role in maintaining health and potentially contributing
to disease (Human Microbiome Project Consortium 2012). However, understanding
the intricate relationships between microbes and diseases remains a complex
challenge.
To address this challenge, the commentary introduces a novel approach for
predicting disease-microbe associations. This method involves integrating disease
and microbe networks and employing advanced network topology analysis. By
doing so, researchers can prioritize candidate microbes and unveil potential
disease-causing microorganisms. This not only enhances our comprehension of
microbial interactions and mechanisms but also offers clinical solutions for under-
standing disease mechanisms, diagnosis, and therapeutic interventions (Zhou et al.
2018).

17.2.4 AI-Powered Antibiotic Susceptibility Testing

Antimicrobial resistance stands as a significant global health threat, primarily driven

by the misuse of antibiotics (Laxminarayan et al. 2013). Traditional methods like
disk diffusion antibiotic susceptibility testing (AST) are widely employed but face
criticism due to their complexity and susceptibility to inter-operator variability. In
response, the commentary introduces a groundbreaking solution—an AI-based
smartphone application for offline AST analysis (Lutgring et al. 2020).
This innovative application leverages the power of artificial intelligence to
capture images of antibiotic susceptibility tests using the smartphone’s camera. It
offers a user-friendly interface for analysis and incorporates an embedded expert
system to validate data coherence and provide interpreted results. The result?
Remarkable accuracy, with high agreement rates against hospital-standard systems
and a significant reduction in inter-operator variability (Tang et al. 2021).
Perhaps most importantly, this application is well suited for resource-limited
settings, potentially revolutionizing patients’ access to antibiotic susceptibility test-
ing worldwide. It showcases the tremendous potential of AI to address global health
challenges and improve healthcare outcomes (Lam et al. 2022).
17 Use Cases and Future Aspects of Intelligent Techniques in Microbial. . . 265

17.3 Use Cases of Intelligent Techniques

17.3.1 Microbiome Studies

Microbiome studies have emerged as a critical field of research with applications

ranging from human health to environmental ecology. These studies involve ana-
lyzing the diverse communities of microorganisms that inhabit different ecosystems.
Here, we emphasize the pivotal role of artificial intelligence (AI) and machine
learning in advancing our understanding of these complex microbial communities
(Majeed et al. 2022a).
Microorganisms in the human microbiome play a critical role in skin health and
the development of various skin diseases. The complexity of microbiome data has
historically challenged traditional analysis methods. However, recent advancements
in artificial intelligence (AI) and machine learning are revolutionizing microbiome
analysis in dermatology, enabling more accurate disease diagnosis and prognosis
(Waheed et al. 2022).

17.3.1.1 Predicting Disease Risk Through Gut Microbiome Analysis

In a study by Pasolli and colleagues they harnessed the power of AI and machine
learning to analyze the gut microbiome data of over 150,000 individuals from
diverse geographical locations, age groups, and lifestyles. They developed predic-
tive models that could identify individuals at higher risk for specific diseases, such as
type 2 diabetes and inflammatory bowel disease, based on their gut microbiome
composition. This research highlights how AI-driven microbiome analysis can
contribute to personalized disease risk assessment and early intervention (Majeed
et al. 2020b).

17.3.1.2 Microbial Biomarkers for Disease Prediction

In a study published in Nature, researchers used machine learning to analyze the gut
microbiomes of patients with IBD. They applied supervised learning to identify
microbial signatures associated with disease status. By training their model on a
dataset comprising healthy individuals and IBD patients, the researchers discovered
specific microbial taxa and functional genes that were strongly correlated with IBD.
This study showcased how intelligent techniques can reveal microbial biomarkers
for complex diseases (Tassadduq et al. 2022).
266 M. Naveed et al.

17.3.1.3 Targeted Microbiome Interventions

Zmora and colleagues used machine learning to examine how individual variations
in gut microbiome composition influence responses to probiotics. By analyzing host
and microbiome features, they developed predictive models to identify individuals
who would benefit from specific probiotic interventions. This personalized approach
to microbiome modulation illustrates the potential of AI in tailoring microbial
therapies for improved health outcomes (Majeed et al. 2021c).

17.3.1.4 Discovering Novel Microbial Functions

Forslund and colleagues applied machine learning to analyze metagenomic data

from individuals with type 2 diabetes before and after metformin treatment. Their
study revealed shifts in the gut microbiota associated with both the disease and
treatment. Machine learning algorithms helped identify specific microbial functions
influenced by metformin, shedding light on the interplay between host metabolism,
drug treatment, and the gut microbiome. This research showcases how AI can
unravel novel microbial functions and their roles in health and disease (Hassan
et al. 2022a).

17.3.1.5 Functional Metagenomics in Environmental Microbiomes

In a study published in Nature Methods, scientists applied machine learning to

predict the functional potential of microbial communities in an aquatic ecosystem.
They used metagenomic data to train their model, which could infer the presence of
metabolic pathways and functional genes within the microbiome. This approach
allowed researchers to gain insights into how these microbial communities were
contributing to nutrient cycling and energy flow in the ecosystem (Khan et al. 2022).

17.3.1.6 Recognition of Fungi with Convolutional Neural Networks

(CNN)

Convolutional neural networks (CNN) are a specialized form of machine learning

that can analyze and classify images based on their unique features. In dermatology,
CNN can be applied to diagnose skin conditions by annotating images with
corresponding medical records and pathological results, generating standardized
data for analysis. This approach has proven particularly effective in diagnosing
fungal infections, such as onychomycosis, which is the most common nail disease
caused by fungi (Hassan et al. 2022b). Traditional clinical diagnosis methods for
onychomycosis involve microscopy and fungal culture, which are time consuming
and susceptible to false negatives. CNN models, such as VGG16 and InceptionV3,
17 Use Cases and Future Aspects of Intelligent Techniques in Microbial. . . 267

have demonstrated superior accuracy, sensitivity, and specificity compared to tradi-

tional methods. CNN can also complement manual microscopic examinations,
improving diagnostic accuracy and efficiency (Bashir et al. 2023).

17.3.1.7 Combined Application of Microbial Genome Sequencing

and Machine Learning

Machine learning can analyze sequencing data to identify patterns in pathogen

genomes, enabling the subtyping of pathogens for precise treatment. This approach
has been applied in the diagnosis of syphilis, a disease caused by Treponema
pallidum. Serological tests used in clinical diagnosis lack the ability to determine
the specific strain of the pathogen. By collecting and sequencing T. pallidum
samples from patients worldwide and applying machine learning techniques,
researchers successfully classified spirochete branches, providing crucial informa-
tion for personalized treatment decisions. Additionally, machine learning can ana-
lyze 16S sequencing results to reveal differences in microbial species and
composition ratios between patients and healthy individuals. This approach has
been used to diagnose diseases like human papillomavirus (HPV) infection and
predict HPV types. It can also assess skin status and predict skin diseases based on
microbial composition. Metagenomic sequencing, another technique, has been used
to diagnose acne by analyzing changes in skin microbiota (Sinha et al. 2023).

17.3.1.8 AI and Skin Diseases in the Context of the Gut–Skin Axis

The gut microbiota plays a vital role in maintaining overall health, and disruptions in
the gut–skin axis have been linked to various skin diseases, including atopic
dermatitis, psoriasis, vitiligo, and acne vulgaris. Research has shown that changes
in skin microbes are related to the progression of vitiligo, and machine learning
techniques can help diagnose and predict disease progression based on skin
microbiome change. Similarly, machine learning has been used to analyze the gut
microbiomes of patients with atopic dermatitis, revealing characteristic microbial
signatures associated with the disease. These findings suggest that gut microbiota
analysis can not only predict skin diseases but also distinguish between disease
subtypes, enhancing diagnostic accuracy and guiding treatment decisions (Majeed
et al. 2023).
268 M. Naveed et al.

17.3.2 Antibiotic Resistance Prediction

Predicting antibiotic resistance patterns in microorganisms is crucial for effective

treatment strategies. Case studies demonstrate the role of intelligent techniques in
antibiotic resistance prediction.

17.3.2.1 Genomic Analysis of Antibiotic Resistance in Pathogens

In a clinical setting, a team of researchers used machine learning to predict antibiotic

resistance in a collection of bacterial isolates. By analyzing the genetic profiles of
these bacteria, the team trained a model to predict the resistance of each isolate to
multiple antibiotics. The model demonstrated high accuracy in predicting antibiotic
susceptibility, aiding clinicians in selecting the most effective treatments for infected
patients (Sinha et al. 2023).

17.3.2.2 Real-Time Antibiotic Resistance Prediction

In a hospital setting, a real-time antibiotic resistance prediction system was

implemented using artificial intelligence. This system integrated patient-specific
data, such as antibiotic exposure history and genetic information of infecting bacte-
ria, into a machine learning model. Clinicians received instant predictions of antibi-
otic resistance, enabling them to adjust treatment plans promptly. This case study
showcased the potential of intelligent techniques in improving patient outcomes.
Microbial data analysis, powered by intelligent techniques, has revolutionized
our understanding of microbiological ecosystems and their impact on various fields.
In this section, we will delve into specific use cases where these intelligent tech-
niques have been pivotal, backed by real-world examples and scholarly references
(Jamil et al. 2022).

17.3.2.3 Tracking Resistance Trends

Intelligent techniques also aid in tracking antibiotic resistance trends. A study by Hu

et al. (2020) used machine learning to analyze resistance data from multiple
healthcare facilities. By identifying emerging resistance patterns, they contributed
to more informed antibiotic stewardship practices, minimizing the risk of multidrug-
resistant infections. The main focus of the study involved the application of
advanced machine learning models to the prepared dataset. These models were
designed to identify hidden patterns and trends in antibiotic resistance. Supervised
learning techniques, such as logistic regression and decision trees, were used to
predict the likelihood of antibiotic resistance for specific antibiotics and bacterial
17 Use Cases and Future Aspects of Intelligent Techniques in Microbial. . . 269

Table 17.1 Applications of machine learning and AI in microbial and disease analysis
Title Significance Tools used References
Predicting disease risk Personalized disease risk AI, machine learn- Pasolli
through gut microbiome assessment and early ing, gut microbiome et al.
analysis intervention data (2019)
Microbial biomarkers for Identification of microbial Machine learning, Lloyd-
disease prediction signatures for complex gut microbiome data Price et al.
diseases (2019)
Targeted microbiome Tailoring microbial therapies Machine learning, Zmora
interventions for improved health outcomes host and microbiome et al.
features (2018)
Discovering novel micro- Uncovering novel microbial Machine learning, Forslund
bial functions functions and their roles in metagenomic data et al.
health and disease (2015)
Functional metagenomics Insights into microbial con- Machine learning,
in environmental tributions to nutrient cycling metagenomic data
microbiomes and energy flow in
ecosystems
Recognition of fungi with Accurate diagnosis of fungal Convolutional neural
convolutional neural net- infections in dermatology networks (CNN),
works (CNN) dermatological
images
Combined application of Precision diagnosis of patho- Machine learning,
microbial genome gens and assessment of skin genomic and 16S
sequencing and machine status sequencing data
learning
AI and skin diseases in the Prediction and distinction of Machine learning,
context of the gut–skin skin diseases based on gut gut microbiome data
Axis microbiota
Genomic analysis of anti- Accurate prediction of antibi- Machine learning,
biotic resistance in otic resistance in bacterial genetic profiles
pathogens isolates
Real-time antibiotic resis- Prompt adjustment of treat- Artificial intelligence
tance prediction ment plans for patients (AI), patient data,
genetic information
Tracking resistance trends Informed antibiotic steward- Machine learning, Hu et al.
ship practices to minimize resistance data (2020)
drug resistance

strains. Unsupervised learning methods, such as clustering and principal component

analysis, were employed to uncover hidden relationships among bacterial strains,
resistance profiles, and patient attributes as shown in Table 17.1.
270 M. Naveed et al.

17.3.3 Environmental Microbiology

Environmental microbiology is a multidisciplinary field that delves into the intricate

relationships between microorganisms and their natural habitats, from terrestrial
soils to aquatic environments, and even extreme ecosystems like hydrothermal
vents. The microorganisms in these environments play pivotal roles in nutrient
cycling, degradation of organic matter, and the overall health of ecosystems. They
impact global processes such as climate change, biogeochemical cycling, and
pollution remediation (Majeed et al. 2022b).
Intelligent techniques, which encompass a variety of data-driven and computa-
tional approaches, have become indispensable tools in environmental microbiology.
These techniques have revolutionized the way we analyze and predict microbial
dynamics, providing unprecedented insights into the hidden world of microorgan-
isms in natural ecosystems. Given below is the significance of intelligent techniques
in understanding and managing our environment, illustrated through case studies
and practical applications.

17.3.3.1 Machine Learning Empowers Microbial Community Analysis

in Soil Ecosystems

In a research conducted by Zhang et al. (2021), machine learning emerged as a

powerful tool for unraveling the complex relationships between soil microbial
communities and their environment. Traditional methods for analyzing soil micro-
organisms are labor intensive and time consuming, making large-scale studies
challenging. In this study, high-throughput sequencing technology generated exten-
sive microbial data, which machine learning efficiently processed. The application of
machine learning techniques, such as Random Forest and Gradient Boosting
Machines, allowed researchers to identify the key environmental factors influencing
soil microbial communities in Chinese tea plantations, quantify the impacts of land
management practices, and discover specific microbial biomarkers for different soil
conditions and management practices. This case study illustrates the critical impor-
tance of machine learning in enabling data-driven decisions for sustainable agricul-
ture and land management, ultimately contributing to informed and environmentally
responsible soil ecosystem conservation and optimization (Majeed et al. 2023).

17.3.3.2 Advancements in Predicting Nutrient Cycling in Aquatic

Environments

Yao et al. (2021) conducted a research that emerged as a groundbreaking tool for
predicting nutrient concentrations and cycling rates in polluted river ecosystems,
addressing challenges posed by the complexity and variability of aquatic environ-
ments. By harnessing the power of intelligence techniques such as Random Forest
17 Use Cases and Future Aspects of Intelligent Techniques in Microbial. . . 271

and Gradient Boosting Machines, researchers accurately forecasted nutrient concen-

trations based on water quality and weather data and predicted nutrient cycling rates
considering pollutant inputs. These accurate and real-time predictions hold para-
mount importance for managing and preserving water quality, preventing eutrophi-
cation, and enabling timely interventions to safeguard aquatic ecosystems. This
study highlights how machine learning enhances our capacity to proactively manage
and conserve vital aquatic resources, highlighting its crucial role in ensuring the
sustainable management of our planet’s waterways (Hassan et al. 2022a).

17.3.3.3 Machine Learning and AI Revolutionize Bioremediation

of Contaminated Sites

In a recent study by Smith et al. (2022), the integration of artificial intelligence

(AI) and machine learning emerged as a game-changing solution for optimizing the
bioremediation of highly contaminated industrial sites. Traditional approaches to
bioremediation often prove time consuming and resource intensive, but AI and
machine learning techniques, including Artificial Neural Networks (ANNs) and
Genetic Algorithms, enabled precise modeling of complex interactions between
environmental factors and microbial responses. This allowed researchers to predict
optimal conditions for pollutant degradation, design targeted bioremediation strate-
gies, and even identify genetic modifications to enhance native microbial species’
pollutant-degrading capabilities. The transformative impact of AI and machine
learning in this study not only expedites cleanup processes but also minimizes the
ecological impact of contaminants, exemplifying their vital role in advancing sus-
tainable and environmentally responsible bioremediation efforts for industrial con-
taminated sites.

17.3.4 Biotechnology and Bioprocess Optimization

Biotechnology stands at the forefront of scientific innovation, utilizing the remark-

able capabilities of microorganisms to address a myriad of challenges and to drive
progress in diverse sectors, including biofuel production, pharmaceuticals, and
bioprocessing. Within this dynamic field, the optimization of bioprocesses is of
paramount importance to maximize efficiency, reduce costs, and minimize environ-
mental impact. In this context, intelligent techniques such as artificial intelligence
(AI) and machine learning have emerged as indispensable tools. This section
explores the pivotal role of intelligent techniques in biotechnology and bioprocess
optimization, elucidated through case studies that underscore their significance in the
industrial and pharmaceutical sectors (Majeed et al. 2023).
272 M. Naveed et al.

17.3.4.1 Optimization of Bioethanol Fermentation

In a groundbreaking study by Chen et al. (2022), machine learning and artificial

intelligence emerged as transformative tools for optimizing the bioethanol fermen-
tation process, addressing challenges associated with intricate parameter interac-
tions. By efficiently processing extensive datasets, machine learning models,
including Artificial Neural Networks (ANNs) and Genetic Algorithms, accurately
predicted optimal fermentation conditions for bioethanol production. This precision
enabled researchers to fine-tune the process, resulting in higher bioethanol yields,
reduced production costs, and minimized environmental impact, ultimately advanc-
ing the sustainability and economic viability of bioethanol production. This case
study indicates the pivotal role of data-driven decision-making in revolutionizing
bioenergy research and positioning bioethanol as a renewable and environmentally
responsible alternative to fossil fuels, offering a path to a more sustainable energy
future (Arif et al. 2021).

17.3.4.2 Accelerating Drug Discovery Through Metabolic Engineering

In a study by Sharma et al. (2022), AI emerged as a potent tool for accelerating drug
discovery through metabolic engineering, addressing the complexities of enhancing
secondary metabolite production in Streptomyces strains. By analyzing extensive
omics data, including genomics, transcriptomics, and metabolomics, machine learn-
ing techniques such as Random Forest Regression and Genetic Algorithm Optimi-
zation accurately identified genetic modifications that led to significantly increased
secondary metabolite yields. This precise strain design not only expedited drug
discovery processes but also reduced development time and costs, potentially
making life-saving medications more accessible and cost effective (Sinha et al.
2023). The study underscores the transformative impact of data-driven decision-
making in revolutionizing the pharmaceutical industry, offering a promising path to
more efficient drug development, and improved global healthcare outcomes through
innovative secondary metabolite production in microbial strains as shown in
Table 17.2.

17.4 Challenges in Microbial Data Analysis

17.4.1 Data Quality and Quantity

Ensuring data quality and quantity is a foundational challenge in microbial data

analysis with AI technology. The microbial world’s diversity and complexity
demand comprehensive, high-quality datasets for meaningful analysis (Thompson
et al. 2017). Challenges arise from various sources, including environmental
17 Use Cases and Future Aspects of Intelligent Techniques in Microbial. . . 273

Table 17.2 Applications of machine learning and AI in environmental and microbial sciences
Title of case study Importance Tools used References
Machine learning Machine learning revolu- High-throughput sequenc- Zhang
empowers microbial tionizes soil microbial com- ing, machine learning et al.
community analysis in munity analysis, enabling (random Forest, gradient (2021)
soil ecosystems data-driven decisions for boosting machines)
sustainable agriculture and
land management
Advancements in Machine learning plays a Machine learning (random Yao et al.
predicting nutrient crucial role in real-time Forest, gradient boosting (2021)
cycling in aquatic nutrient prediction for water machines)
environments quality management and
ecological conservation
Machine learning and AI and machine learning Artificial intelligence (arti- Smith
AI revolutionize bio- optimize bioremediation, ficial neural networks, et al.
remediation of con- reducing cleanup time, genetic algorithms) (2022)
taminated sites costs, and ecological
impacts in contaminated
industrial sites
Optimization of Machine learning enhances Machine learning (artifi- Chen et al.
bioethanol bioethanol production effi- cial neural networks, (2022)
fermentation ciency, reducing costs, and genetic algorithms)
environmental impact,
advancing sustainable
energy solutions
Accelerating drug dis- Machine learning expedites Machine learning (random Sharma
covery through meta- secondary metabolite pro- Forest regression, genetic et al.
bolic engineering duction in microbial strains, algorithms) (2022)
potentially improving
global healthcare outcomes

sampling, clinical specimens, and genomic sequencing. Contamination, sequencing

errors, and biases can affect data integrity, making robust data preprocessing and
quality control crucial (Karstens et al. 2019).
Moreover, AI models, especially deep learning approaches, require substantial
quantities of data for effective training (Esteva et al. 2019). Collecting diverse
microbial samples for model training is resource intensive and time consuming.
Researchers are addressing these challenges through advancements in data collec-
tion methods, bioinformatics, and collaboration across research domains (Pasolli
et al. 2019).

17.4.2 Interpretability

Interpreting AI models used in microbial data analysis is pivotal but challenging.

Often considered “black boxes,” these models lack transparency in explaining their
decision-making processes. In microbial research, understanding how AI models
274 M. Naveed et al.

reach conclusions about microbial interactions, disease associations, or other

insights is essential (Holzinger et al. 2017).
Recent research has focused on developing interpretable AI models and methods
for microbial data analysis. Explainable AI (XAI) techniques, such as feature
importance analysis and rule-based models, aim to enhance transparency and trust
in AI-generated findings (Ribeiro et al. 2016). These approaches enable researchers
to glean insights from AI models while preserving interpretability.

17.4.3 Ethical and Privacy Concerns

AI’s use in microbial data analysis raises ethical and privacy concerns, especially
when handling human-associated microbial data. Protecting individual privacy is
paramount, necessitating robust data governance and consent frameworks (Cabili
et al. 2020). Ensuring that data is anonymized and de-identified is crucial to mitigate
privacy risks (Brinegar et al. 2021). Ethical considerations extend to issues like data
ownership, informed consent, and responsible data sharing practices (Nebert et al.
2020). Striking a balance between scientific advancement through AI-driven analy-
sis and safeguarding individual privacy remains an ongoing challenge.

17.4.4 Validation and Generalization

Validating AI models in microbial data analysis is complex. Models trained on

specific datasets may excel on that data but struggle to generalize to new environ-
ments or populations. Achieving robust model generalization is crucial for their
utility across diverse microbial ecosystems.
Furthermore, rigorous validation against ground truth data, experimental results,
or clinical outcomes is essential to ensure the reliability of AI-generated insights.
This validation process is particularly critical in clinical settings where AI-driven
decisions impact patient care. Researchers are actively working on solutions to these
challenges, employing techniques like cross-validation and external validation
datasets to assess model performance and generalization.
Addressing these challenges is paramount for realizing the full potential of AI in
microbial data analysis. The latest research and interdisciplinary collaborations are
advancing methods and best practices to overcome these obstacles, facilitating
breakthroughs in microbiology, biotechnology, and healthcare.
17 Use Cases and Future Aspects of Intelligent Techniques in Microbial. . . 275

17.5 Future Aspects and Trends

Artificial intelligence (AI) has revolutionized disease diagnosis, ushering in a new

era of healthcare that is increasingly accurate, efficient, and personalized. With the
rise in the usage of AI, several promising future aspects and trends are poised to
shape the landscape of disease diagnosis. Microbial data analysis is a dynamic field,
and several exciting future aspects and trends are shaping its evolution.

17.5.1 Integration of Omics Data

Microbial data analysis is increasingly embracing multi-omics data integration. This

involves combining data from various “omics” levels, such as genomics,
metagenomics, transcriptomics, proteomics, and metabolomics. Integration enables
a more holistic view of microbial communities and their functions. For example,
integrating metagenomic and meta transcriptomic data can reveal which microbial
genes are actively expressed under different conditions (Franzosa et al. 2018). This
approach offers a deeper understanding of microbial communities’ roles in health,
disease, and environmental processes.

17.5.2 Explainable AI in Microbial Data Analysis

As AI techniques become more pervasive in microbial data analysis, the need for
interpretability and transparency grows. Explainable AI (XAI) methods are gaining
importance to provide insights into how AI models make decisions. In microbiome
research, XAI can help researchers understand the features and patterns that drive
predictions or classifications, enhancing trust and facilitating the translation of
AI-driven findings into actionable insights (Ribeiro et al. 2016).

17.5.3 Personalized Microbiome Analysis

The concept of personalized medicine is extending to the microbiome. Microbiome

data can be used to tailor healthcare interventions to individual patients. For instance,
by analyzing an individual’s microbiome, healthcare providers can prescribe per-
sonalized dietary recommendations, probiotics, or antibiotics to treat conditions like
irritable bowel syndrome (Jeffery et al. 2012). This trend is likely to grow as the link
between the microbiome and various health factors becomes clearer.
276 M. Naveed et al.

17.5.4 Ethical and Regulatory Frameworks

The responsible use of microbiome data is a pressing concern. Ethical consider-

ations, including informed consent and data privacy, are paramount. The develop-
ment of ethical guidelines and regulatory frameworks for microbiome data
collection, sharing, and analysis is essential (Haeuser and Dawson 2020).
Researchers and policymakers must work together to establish ethical best practices
that respect individual rights and ensure the secure handling of sensitive microbiome
data (Waheed et al. 2023).

17.5.5 Collaborative Research and Data Sharing

Collaboration and data sharing initiatives are crucial for advancing microbial data
analysis. The establishment of global microbiome data repositories and open science
initiatives will foster data sharing, enabling researchers worldwide to access, ana-
lyze, and build upon each other’s work. Collaboration can lead to more comprehen-
sive datasets and insights into the microbiome’s role in diverse ecosystems and
health conditions (Knight et al. 2018).
Incorporating these future aspects and trends into microbial data analysis will
contribute to a more comprehensive understanding of microbial communities and
their implications for human health, environmental sustainability, and biotechnolog-
ical advancements. Researchers and practitioners in this field should be vigilant in
staying current with these developments to make the most of intelligent techniques
in microbiome research.

17.6 Conclusion

Microbial data analysis, empowered by intelligent techniques such as machine

learning and artificial intelligence, has ushered in a new era of understanding and
harnessing the power of microorganisms. From environmental microbiology to
biotechnology and healthcare, the applications of these techniques are vast and
transformative. As we conclude this exploration, it is evident that the future of
microbial data analysis holds immense promise and potential. The significance of
microbial data analysis in various fields, including ecology, public health, and
biotechnology, cannot be overstated. Intelligent techniques have enabled us to
delve deeper into the microbial world, uncovering hidden patterns, predicting
behaviors, and accelerating scientific discoveries. From predicting disease risks
based on gut microbiome composition to revolutionizing antibiotic resistance pre-
diction and optimizing bioprocesses, intelligent techniques have become indispens-
able tools for researchers and practitioners alike. However, several challenges
17 Use Cases and Future Aspects of Intelligent Techniques in Microbial. . . 277

remain in the field of microbial data analysis, including data quality and quantity,
interpretability of machine learning models, ethical and privacy concerns, and the
need for robust validation and generalization. Addressing these challenges will be
crucial in ensuring the responsible and ethical use of intelligent techniques in
microbial research. Looking ahead, the future of microbial data analysis is bright.
Emerging trends such as the integration of omics data, the development of explain-
able AI for microbial research, personalized microbiome analysis, and the establish-
ment of ethical and regulatory frameworks will shape the landscape of this field.
Collaborative research and data sharing will also play a pivotal role in advancing our
understanding of microorganisms and their impact on the world.

References

Arif U, Bhatti KH, Ajaib M, Wagay NA, Majeed M, Zeb J, Hameed A, Kiani J (2021) Ethnobo-
tanical indigenous knowledge of Tehsil Charhoi, District Kotli, Azad Jammu and Kashmir,
Pakistan. Ethnobot Res Appl 22:1–24. https://doi.org/10.32859/ERA.22.50.1-24
Bashir SM, Altaf M, Hussain T, Umair M, Majeed M, Mangrio WM, Khan AM, Gulshan AB,
Hamed MH, Ashraf S, Amjad MS, Bussmann RW, Abbasi AM, Casini R, Alataway A, Dewidar
AZ, Al-Yafrsi M, Amin MH, Elansary HO (2023) Vernacular taxonomy, cultural and
ethnopharmacological applications of avian and mammalian species in the vicinity of Ayubia
National Park, Himalayan Region. Biology 12:4
Bates DW, Saria S, Ohno-Machado L, Shah A, Escobar G (2014) Big data in health care: using
analytics to identify and manage high-risk and high-cost patients. Health Aff 33(7):1123–1131.
https://doi.org/10.1377/hlthaff.2014.0041
Belkaid Y, Hand TW (2014) Role of the microbiota in immunity and inflammation. Cell 157(1):
121–141
Brinegar K, DeDeo S, Lazer D (2021) Ethnicity and representation in academic machine learning
discourse. arXiv:2104.05560
Bzdok D, Ioannidis JP (2019) Exploration, inference, and prediction in neuroscience and biomed-
icine. Trends Neurosci 42(4):251–262. https://doi.org/10.1016/j.tins.2019.02.003
Cabili MN, Dunagin MC, McClanahan PD, Biaesch A, Padovan-Merhar O, Regev A, Raj A (2020)
Localization and abundance analysis of human lncRNAs at single-cell and single-molecule
resolution. Genes Dev 34(7–8):440–451
Chen Y, Wang Q, Jiang H et al (2022) Machine learning models for enhanced bioethanol
production through fermentation process optimization. Bioenergy Res:1–12
Esteva A, Robicquet A, Ramsundar B, Kuleshov V, DePristo M, Chou K, Dean J (2019) A guide to
deep learning in healthcare. Nat Med 25(1):24–29
Falkowski PG, Fenchel T, Delong EF (2008) The microbial engines that drive Earth’s biogeo-
chemical cycles. Science 320(5879):1034–1039. https://doi.org/10.1126/science.1153213
Forslund K, Hildebrand F, Nielsen T et al (2015) Disentangling type 2 diabetes and metformin
treatment signatures in the human gut microbiota. Nature 528(7581):262–266
Franzosa EA, Hsu T, Sirota-Madi A, Shafquat A, Abu-Ali G, Morgan XC, Huttenhower C (2018)
Sequencing and beyond: integrating molecular ‘omics’ for microbial community profiling. Nat
Rev Microbiol 16(11):591–606
Guttenberg N, Warfield K, Pickett BE, Ledogar RJ (2020) Brain size and network properties of
human population genetic networks. Nat Commun 11(1):1–12. https://doi.org/10.1038/s41467-
020-19135-w
278 M. Naveed et al.

Haberbeck LU, Wang X, Michiels C, Devlieghere F, Uyttendaele M, Geeraerd AH (2017) Cross-

protection between controlled acid-adaptation and thermal inactivation for 48 Escherichia coli
strains. Int J Food Microbiol 241:206–214
Haeuser E, Dawson N (2020) Microbiome data should be regulated as personal data. Nat Med
26(12):1806–1808
Haq SM, Yaqoob U, Majeed M, Amjad MS, Hassan M, Ahmad R, Morales-de la Nuez A (2022)
Quantitative ethnoveterinary study on plant resource utilization by indigenous communities in
high-altitude regions. Front Vet Sci 9:94404
Hassan M, Haq SM, Ahmad R, Majeed M, Sahito HA, Shirani M, Mubeen I, Aziz MA, Pieroni A,
Bussmann RW, Alataway A, Dewidar AZ, Al-Yafrsi M, Elansary HO, Yessoufou K (2022a)
Traditional use of wild and domestic fauna among different ethnic groups in the Western
Himalayas? Cross cultural analysis. Animals 12:17
Hassan M, Haq SM, Majeed M, Umair M, Sahito HA, Shirani M, Waheed M, Aziz R, Ahmad R,
Bussmann RW, Alataway A, Dewidar AZ, El-Abedin TKZ, Al-Yafrsi M, Elansary HO,
Yessoufou K (2022b) Traditional food and medicine: ethno-traditional usage of fish fauna
across the valley of Kashmir: a Western Himalayan region. Diversity 14:6
Holzinger A, Langs G, Denk H, Zatloukal K, Müller H (2017) Causability and explainability of
artificial intelligence in medicine. Wiley Interdiscip Rev Data Mining Knowl Discov 7(6):e1212
Hu Y, Yang X, Qin J, Lu N, Cheng G, Wu N, Lin L (2020) Metagenome-wide analysis of antibiotic
resistance genes in a large cohort of human gut microbiota. Nat Commun 11(1):1–12
Jamil MD, Waheed M, Akhtar S, Bangash N, Chaudhari SK, Majeed M, Hussain M, Ali K, Jones
DA (2022) Invasive plants diversity, ecological status, and distribution pattern in relation to
edaphic factors in different habitat types of district Mandi Bahauddin, Punjab, Pakistan.
Sustainability (Switzerland) 14:20
Jeffery IB, O'Toole PW, Öhman L, Claesson MJ, Deane J, Quigley EM, Simrén M (2012) An
irritable bowel syndrome subtype defined by species-specific alterations in faecal microbiota.
Gut 61(7):997–1006
Jeong H, Tombor B, Albert R, Oltvai ZN, Barabási AL (2000) The large-scale organization of
metabolic networks. Nature 407(6804):651–654. https://doi.org/10.1038/35036627
Kamel Boulos MN, Resch B, Crowley DN, Breslin JG, Sohn G, Burtner R, Lu Z (2011)
Crowdsourcing, citizen sensing and sensor web technologies for public and environmental
health surveillance and crisis management: trends, OGC standards and application examples.
Int J Health Geogr 10(1):67. https://doi.org/10.1186/1476-072X-10-67
Karstens L, Asquith M, Davin S, Fair DA, Gregory WT, Wolfe AJ, McWeeney SK (2019)
Controlling for contaminants in low-biomass 16S rRNA gene sequencing experiments.
mSystems 4(4):e00290–e00219
Khan AM, Li Q, Saqib Z, Khan N, Habib T, Khalid N, Majeed M, Tariq A (2022) MaxEnt
modelling and impact of climate change on habitat suitability variations of economically
important Chilgoza pine (Pinus gerardiana wall.) in South Asia. Forests 13:5
Khoja AA, Haq SM, Majeed M, Hassan M, Waheed M, Yaqoob U, Bussmann RW, Alataway A,
Dewidar AZ, Al-Yafrsi M, Elansary HO, Yessoufou K, Zaman W (2022) Diversity, ecological
and traditional knowledge of pteridophytes in the Western Himalayas. Diversity 14:8
Knight R, Vrbanac A, Taylor BC, Aksenov A, Callewaert C, Debelius J, Flores R (2018) Best
practices for analysing microbiomes. Nat Rev Microbiol 16(7):410–422
Lam TB, Hultcrantz M, Wallis C et al (2022) Robot-assisted radical prostatectomy versus open
radical prostatectomy: a systematic review and meta-analysis. Eur Urol 17:2617–2631. https://
doi.org/10.1016/j.eururo.2022.01.057
Laxminarayan R, Duse A, Wattal C, Zaidi AK, Wertheim HF, Sumpradit N, Cars O (2013)
Antibiotic resistance—the need for global solutions. Lancet Infect Dis 13(12):1057–1098.
https://doi.org/10.1016/S1473-3099(13)70318-9
Lucas TC (2020) A translucent box: interpretable machine learning in ecology. Ecol Monogr 90(4):
e01422
17 Use Cases and Future Aspects of Intelligent Techniques in Microbial. . . 279

Lutgring JD, Machado MJ, Benitez AJ (2020) Evaluation of the VITEK® 2 automated suscepti-
bility testing system against carbapenemase-producing enterobacteriaceae with a modified
carbapenem inactivation method. J Clin Microbiol 58(5):e02088–e02019. https://doi.org/10.
1128/JCM.02088-19
Majeed M, Bhatti KH, Amjad MS, Abbasi AM, Rashid A, Nawaz F, Ahmad KS (2020a) Ethno-
veterinary practices of Poaceae taxa in Punjab, Pakistan
Majeed M, Bhatti KH, Amjad MS, Abbasi M, Id RWB, Nawaz F, Rashid A, Mehmood A, Id MM,
Khan WM, Id SA (2020b) Ethno-veterinary uses of Poaceae in Punjab, Pakistan. PLoS One 15:
e0241705. https://doi.org/10.1371/journal.pone.0241705
Majeed M, Bhatti KH, Amjad MS (2021a) Impact of climatic variations on the flowering phenology
of plant species in Jhelum district, Punjab, Pakistan. Appl Ecol Environ Res 19:5
Majeed M, Bhatti KH, Pieroni A, Sõukand R, Bussmann RW, Khan AM, Chaudhari SK, Aziz MA,
Amjad MS (2021b) Gathered wild food plants among diverse religious groups in Jhelum
District, Punjab, Pakistan. Foods 10:3
Majeed M, Tariq A, Anwar MM, Khan AM, Arshad F, Mumtaz F, Farhan M, Zhang L, Zafar A,
Aziz M, Abbasi S, Rahman G, Hussain S, Waheed M, Fatima K, Shaukat S (2021c) Monitoring
of land use? And cover change and potential causal factors of climate change in Jhelum District,
Punjab, Pakistan, through GIS and multi-temporal satellite data. Land 10:10
Majeed M, Lu L, Haq SM, Waheed M, Sahito HA, Fatima S, Aziz R, Bussmann RW, Tariq A,
Ullah I, Aslam M (2022a) Spatiotemporal distribution patterns of climbers along an abiotic
gradient in Jhelum District, Punjab, Pakistan. Forests 13:8
Majeed M, Tariq A, Haq SM, Waheed M, Anwar MM, Li Q, Aslam M, Abbasi S, Mousa BG, Jamil
A (2022b) A detailed ecological exploration of the distribution patterns of wild Poaceae from
the Jhelum District (Punjab), Pakistan. Sustainability (Switzerland) 14:7
Majeed M, Lu L, Anwar MM, Tariq A, Qin S, El-Hefnawy ME, El-Sharnouby M, Li Q, Alasmari A
(2023) Prediction of flash flood susceptibility using integrating analytic hierarchy process
(AHP) and frequency ratio (FR) algorithms. Front Environ Sci 10:1037547. https://doi.org/10.
3389/fenvs.2022.1037547
Matchado MS et al (2021) Network analysis methods for studying microbial communities: a mini
review. Comput Struct Biotechnol J 19:2687–2698. https://doi.org/10.1016/j.csbj.2021.05.001
Nebert DW, Zhang G, Vesell ES, Dixon K (2020) The human cytochrome P450 (CYP) allele
nomenclature database: a one-stop site for nomenclature, functional-allelic-variant reference,
and genotype–phenotype relations. Hum Genomics 14(1)
Pasolli E, Asnicar F, Manara S et al (2019) Extensive unexplored human microbiome diversity
revealed by over 150,000 genomes from metagenomes spanning age, geography, and lifestyle.
Cell 176(3):649–662
Qu K, Guo F, Liu X, Lin Y, Zou Q (2019) Application of machine learning in microbiology. Front
Microbiol 10:827
Ribeiro MT, Singh S, Guestrin C (2016) “Why should I trust you?” Explaining the predictions of
any classifier. In: Proceedings of the 22nd ACM SIGKDD international conference on knowl-
edge discovery and data mining, pp 1135–1144
Sahayasheela VJ, Lankadasari MB, Dan VM, Dastager SG, Pandian GN, Sugiyama H (2022)
Artificial intelligence in microbial natural product drug discovery: current and emerging role.
Nat Prod Rep 39:2215
Sharma A, Gupta A, Patel D et al (2022) Machine learning-driven metabolic engineering for
enhanced secondary metabolite production in streptomyces. 3 Biotech 12:449
Sinha D, Maurya AK, Abdi G, Majeed M, Agarwal R, Mukherjee R, Ganguly S, Aziz R, Bhatia M,
Majgaonkar A, Seal S, Das M, Banerjee S, Chowdhury S, Adeyemi SB, Chen JT (2023)
Integrated genomic selection for accelerating breeding programs of climate-smart cereals.
Genes 14:7
Smith J, Brown L, Garcia M et al (2022) Integrating artificial intelligence for enhanced bioreme-
diation of industrial contaminated sites. Environ Sci Eng 5:100016
280 M. Naveed et al.

Tang H, Zhao X, Dube L, Boudreau RA, Fang R, Xu L (2021) An integrated smartphone-based

platform for rapid antimicrobial susceptibility testing. Adv Sci 8(18):2101407. https://doi.org/
10.1002/advs.202101407
Tassadduq SS, Akhtar S, Waheed M, Bangash N, Nayab DE, Majeed M, Abbasi S, Muhammad M,
Alataway A, Dewidar AZ, Elansary HO, Yessoufou K (2022) Ecological distribution patterns of
wild grasses and abiotic factors. Sustainability (Switzerland) 14:18
Thompson LR, Sanders JG, McDonald D, Amir A, Ladau J, Locey KJ, Knight R (2017) A
communal catalogue reveals Earth’s multiscale microbial diversity. Nature 551(7681):457–463
Tirtawijaya G, Meinita MDN, Marhaeni B, Haque MN, Moon IS, Hong Y-K (2018) Neurotrophic
activity of the Carrageenophyte Kappaphycus alvarezii cultivated at different depths and for
different growth periods in various areas of Indonesia. Evid Based Complement Alternat Med
2018:1098076
Ullah I, Aslam B, Shah SHIA, Tariq A, Qin S, Majeed M, Havenith HB (2022) An integrated
approach of machine learning, remote sensing, and GIS data for the landslide susceptibility
mapping. Land 11:8
Waheed M, Arshad F, Majeed M, Fatima S, Mukhtar N, Aziz R, Mangrio WM, Almohamad H,
Dughairi AA, Al-Mutiry M, Abdo HG (2022) Community structure and distribution pattern of
woody vegetation in response to soil properties in semi-arid Lowland District Kasur Punjab,
Pakistan. Land 11:12
Waheed M, Arshad F, Majeed M, Haq SM, Aziz R, Bussmann RW, Ali K, Subhan F, Jones DA,
Zaitouny A (2023) Potential distribution of a noxious weed (Solanum viarum Du-nal), current
status, and future invasion risk based on MaxEnt modeling. In: Geology, ecology, and land-
scapes. Taylor & Francis, p 1. https://doi.org/10.1080/24749508.2023.2179752
Yao M, Liu Z, Hou L et al (2021) Machine learning models for predicting nutrient concentrations
and cycling rates in polluted river ecosystems. Glob Biogeochem Cycles 35(9):
e2021GB007146
Zhang B, Sun L, Zheng W et al (2021) Machine learning reveals the environmental and manage-
ment impacts on soil microbial communities in Chinese tea plantations. Environ Microbiol
23(7):3435–3449
Zhou Y, Gao H, Mihindukulasuriya KA, La Rosa PS, Wylie KM, Vishnivetskaya T, Jansson JK
(2018) Biogeography of the ecosystems of the healthy human body. Genome Biol 19(1):1–14.
https://doi.org/10.1186/s13059-018-1556-2
Zmora N, Zilberman-Schapira G, Suez J et al (2018) Personalized gut mucosal colonization
resistance to empiric probiotics is associated with unique host and microbiome features. Cell
174(6):1388–1405
Chapter 18
Early Crop Disease Identification Using
Multi-fork Tree Networks and Microbial
Data Intelligence

S. S. Ittannavar, B. P. Khot, Vibhor Kumar Vishnoi,

Swati Shailesh Chandurkar, and Harshal Mahajan

Abstract In the early stages of crop disease, timely acquisition of information about
crop diseases, determination of the causes and severity of infection, and targeted
treatment are essential for preventing a decline in crop yield caused by disease
spread. To address the issue of low accuracy in traditional deep learning networks
for early crop disease identification, we propose an improved attention mechanism-
based multi-fork tree network method. This method combines the attention mecha-
nism with a residual network to recalibrate disease feature maps, resulting in
SMLP_Res (Squeeze-Multi-layer Perceptron ResNet). Additionally, we extend the
high-feature extraction-capable SMLP_ResNet (Squeeze-Multi-Layer Perceptron
ResNet) network with a multi-fork tree structure, simplifying the task of early crop
disease identification and effectively extracting early disease features. In our exper-
iments, we use two datasets, Plant Village and AI Challenger 2018, to train and
validate three network models: 18-layer ResNet, SE_ResNet, and SMLP_ResNet, as
well as their equivalent multi-fork tree structure models, to assess the impact of
SMLP_Res and the multi-fork tree structure on crop disease identification models.
The experimental analysis shows that the three network models, 18-layer ResNet,
SE_ResNet, and SMLP_ResNet, all achieve an accuracy rate of over 99% in disease
identification on the Plant Village dataset, where disease features are more

S. S. Ittannavar (✉) · B. P. Khot

ECE Department, Hirasugar Institute of Technology, Belgaum, India
V. K. Vishnoi
College of Computing Sciences and Information Technology, Teerthanker Mahaveer
University, Moradabad, India
e-mail: [email protected]
S. S. Chandurkar
Pimpri Chinchwad College of Engineering, Pune, India
e-mail: [email protected]
H. Mahajan
Indira College of Engineering and Management, Pune, India
e-mail: [email protected]

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 281
A. Khamparia et al. (eds.), Microbial Data Intelligence and Computational
Techniques for Sustainable Computing, Microorganisms for Sustainability 47,
https://doi.org/10.1007/978-981-99-9621-6_18
282 S. S. Ittannavar et al.

pronounced. However, their accuracy rates on the early disease dataset AI Chal-
lenger 2018 do not exceed 87%. SMLP_ResNet, due to the inclusion of the
SMLP_Res module, provides more comprehensive feature extraction for crop dis-
ease data, resulting in better detection performance. Among the three early disease
identification models with multi-fork tree structures, all models show significant
improvements in accuracy on the AI Challenger 2018 dataset. The multi-fork tree
SMLP_ResNet outperforms the other two models, achieving the best performance
with a cherry early disease identification accuracy rate of 99.13%. The proposed
multi-fork tree SMLP_ResNet crop early disease identification model simplifies the
recognition task, suppresses noise transmission, and achieves high accuracy.

Keywords Crop disease identification · Early detection · Multi-fork tree networks ·

Microbial data intelligence · Deep learning · Feature extraction · Disease
classification

18.1 Introduction

Diseases represent a significant constraint on the quality and sustainable growth of

crop output. During the initial phases of disease infestation, it is crucial to promptly
acquire illness-related data and accurately diagnose the underlying factors contrib-
uting to the infection. Through precise evaluation of leaf disease severity, imple-
mentation of suitable chemical interventions, and reduction of pesticide usage, it is
possible to mitigate environmental pollution and successfully prevent and manage
illnesses, hence averting a decrease in crop productivity resulting from disease
propagation. Crop diseases can arise due to a multitude of reasons, and they have
the potential to exhibit a range of symptoms under varying environmental conditions
and at different developmental phases of the crop. The conventional method of
disease identification in agriculture involves the expertise of agricultural profes-
sionals who possess substantial practical knowledge. This approach is characterised
by its time-consuming nature and a tendency to yield erroneous results.
The advent of information technology has facilitated the utilisation of image
processing and machine learning methodologies in the field of crop disease diag-
nostics. These techniques have proven to be highly valuable in enabling rapid,
precise, and non-invasive identification of diseases affecting crops. In the realm of
conventional disease identification, scholars have employed infrared imaging tech-
nology. The utilisation of thermal infrared imaging by the author (Karanth et al.
2023) facilitated the acquisition of temperature data pertaining to both healthy and
diseased areas of rapeseed plants, ultimately leading to the successful early diagnosis
of rapeseed club root disease. The author (Gavidia et al. 2023) utilised hyper spectral
imaging to identify and categorise rice blast disease. This was achieved by
employing Principal Component Analysis (PCA) and Competitive Adaptive
Reweighted Algorithm (CARS) to choose feature variables for disease grading
clustering algorithms. These algorithms are commonly employed to group data
with comparable characteristics. However, they encounter difficulties such as
18 Early Crop Disease Identification Using Multi-fork Tree Networks. . . 283

sensitivity to sample centroids, substantial fluctuations in accuracy (Kutyauripo et al.

2023), and intricate feature distance descriptions. Consequently, scholars have made
enhancements to clustering algorithms in order to get disease identification. The
SAK-means approach, as proposed by the author (Holzinger et al. 2023), utilises
Hadoop parallel processing to autonomously ascertain the appropriate number of
sample categories and identify photos related to citrus red spider mite disease.
Nevertheless, this methodology is limited in its ability to detect many diseases due
to the dataset’s exclusive inclusion of single-crop, single-disease information. The
researchers utilised a dataset consisting of four distinct diseases, namely angular leaf
spot, anthracnose, bacterial wilt, and powdery mildew, along with a set of healthy
leaves (Harner et al. 2023; Jagadeesan et al. 2019; Maier-Hein et al. 2022). They
successfully conducted multi-class disease recognition by employing a Support
Vector Machine (SVM) with a radial basis function kernel. Support Vector
Machines (SVMs) are limited to addressing binary classification tasks, hence intro-
ducing challenges in the training procedure.
All the above-mentioned disease recognition methods are limited to single crops
(Ren et al. 2022; Rowan and Galanakis 2020). When the data environment becomes
complex, and the number of diseases to be identified increases, disease recognition
accuracy still needs improvement. Convolutional neural networks (CNNs) have
made significant strides in the field of computer vision, including image classifica-
tion, object detection, and super-resolution reconstruction, gradually integrating
deep learning techniques into crop disease recognition. Using CNNs has yielded
favourable results in plant leaf (Rowan and Galanakis 2020) and disease recognition
(Nychas et al. 2016). The introduction of the AlexNet network (Saeed et al. 2022;
Varga and Csukas 2022) greatly improved the accuracy of crop disease recognition
compared to traditional algorithms. Building upon this, methods such as VGG
(Rowan et al. 2022), ResNet (Pakseresht et al. 2023), and GoogLeNet (Mrabet
2023) have improved network feature extraction capabilities by altering network
depth and structure. Researchers have applied DCNN (Deep Convolutional Neural
Networks) in banana disease (Joseph 2023), corn disease (Fernández-Ríos et al.
2022), and citrus disease (De Roever 1998) recognition, achieving good results.
However, these deep convolutional network models achieve disease recognition
only when visual features are prominent and are not suitable for early disease
recognition in multiple crops.
To construct a highly accurate early crop disease recognition model, this study
conducts experiments using the Plant Village dataset (Marvin et al. 2009), which
contains crop disease data with prominent features, and the AI Challenger dataset,
which exhibits significant differences in hierarchical features. We introduce attention
mechanisms into the crop disease recognition model to suppress background noise
information and select effective disease features. We also adopt tree structures as
classification algorithms in the data structure domain (Das et al. 2022; Tirkey et al.
2023; Nigam et al. 2023; Abbasi et al. 2023) and integrate them with an improved
attention mechanism to create a disease recognition model. This aims to enhance the
ability to identify early crop diseases, achieve early disease detection and treatment,
and reduce pesticide usage.
284 S. S. Ittannavar et al.

18.2 Crop Disease Data and Disease Recognition Methods

18.2.1 Experimental Data

The quality of disease diagnosis is heavily influenced by the characteristics of the

sample data. The bigger the variety of disease cases, the smaller the distinctions
across samples, and the more difficult disease detection becomes. We trained multi-
crop disease detection neural network models on Plant Village and AI Challenger
2018, two datasets with varying degrees of disease recognition difficulty, and
evaluated their accuracy and resilience (Islam et al. 2023; Zhou et al. 2021; Reumaux
et al. 2023).
Apple, blueberry, cherry, corn, grape, peach, potato, raspberry, soya bean, pump-
kin, strawberry, and tomato are just a few of the crops whose illnesses are
represented in the 54,309 crop photos available in the Plant Village dataset. Moulds,
fungus, and bacteria all have a role in causing these illnesses. In addition, there are
14 photos of healthy crop leaves included in the collection. There are 38 different
types of data in this set (Gao et al. 2021; Kaveney et al. 2023).
There are almost 30,000 photos of agricultural illnesses in the AI Challenger 2018
dataset. These diseases affect a total of ten different crops: apples, cherries, grapes,
citrus fruits, peaches, strawberries, tomatoes, chili peppers, corn, and potatoes. There
are a total of 59 classes in the dataset, which are based on the kind of crop, the type of
disease, and the severity of the disease (Xiong et al. 2020; Modi et al. 2023; Dutta
et al. 2022).
The parameters of the two datasets are listed in Table 18.1.
Figure 18.1 displays a comparison of disease images from Plant Village and AI
Challenger 2018. A noticeable similarity can be observed in the disease severity of
late-stage diseases between AI Challenger 2018 and Plant Village. In the case of the
ten crops in AI Challenger 2018, each crop exhibits different disease conditions, and
diseases are classified into early and late stages based on their characteristics. Late-
stage diseases are easily recognisable due to their distinct features, but the similarity
between the features of healthy leaves and early-stage disease leaves is relatively
high. Additionally, there is a wide variety of diseases, posing a significant challenge
to disease recognition. Therefore, selecting datasets with distinct disease feature
representations validates the effectiveness of the early disease recognition model in
recognising diseases efficiently (Degu et al. 2023; Mensah et al. 2023; Noshiri et al.
2023; Kendler et al. 2022).

Table 18.1 Crop disease Dataset Plant Village AI Challenger 2018

experimental datasets
Number of samples 54,306 35,861
Number of classes 38 59
Number of cores 14 10
Number of diseases 24 27
18 Early Crop Disease Identification Using Multi-fork Tree Networks. . . 285

Tomato Cherry
yellow leaf powdery Grape black Potato early Strawberry
curl virus mildew Apple scab corn rust rot
ot blight Peach scab leaf blight

healthy

Early
stage of
disease

AI
Challenger

Late stage
of disease

Plant
Village

Fig. 18.1 Comparison of images in the experimental datasets

18.2.2 Disease Recognition Methods

When constructing network models, it’s common to opt for deep models to enhance
learning capabilities. However, as networks become deeper, their structure becomes
more complex and redundant, leading to issues like overfitting and gradient
vanishing (Zhao et al. 2023). In the process of feature extraction, early layers of
the network might weaken feature information, resulting in feature loss and network
degradation. Residual Networks (ResNets) introduced the concept of residual struc-
tures, combining shortcut connections and identity mappings to fuse features from
previous layers, thus extracting richer feature information. The Squeeze and Exci-
tation Networks (SENet) model enhances model performance by focusing on rela-
tionships between feature channels, granting Convolutional Neural Networks
(CNNs) the ability to recognise critical features. The core idea of the SE attention
mechanism lies in the compression (Squeeze) and excitation (Excitation) phases,
where it automatically determines the importance of different channels to enhance
feature expression. Combining the characteristics of these two models, an improved
attention mechanism called SMLP_Res is proposed for early crop disease recogni-
tion. This module enhances the SE module using multi-layer perceptron (MLP) and
residual structures to differentiate the importance of feature map channels, reduce
feature loss during transmission, and incorporate the concept of multiple trees by
breaking down complex networks into simpler ones to improve early disease
recognition accuracy (Agarwal et al. 2020).
286 S. S. Ittannavar et al.

18.2.2.1 Improved Attention Module (SMLP)

The SE attention mechanism extracts detailed features from the entire feature map
through two stages: compression and excitation. Because early disease recognition
places a high demand on the network’s ability to extract details, the SMLP module
replaces the excitation part of SE with MLP. MLP leverages the mathematical
property of approximating arbitrary function mappings, learning the importance of
each channel in the feature map, assigning corresponding weights, and enhancing
the expression of significant features while weakening noise information represen-
tation, thus separating mixed features (SMLP). The structure of the SMLP module is
depicted in Fig. 18.2.
GAP (Global Average Pooling) in SMLP encodes global information while
preserving the feature information of the input feature map U. Fsq(uc) compresses
the feature map U into a 1 × 1 × C format and encodes each channel uc to integrate all
feature information. The expression is as follows:

1 j X
Grq ðdÞ = v ði, jÞ: ð18:1Þ
j × Xi i=1 j=1 c

The encoded feature map is then input into the MLP to learn nonlinear relation-
ships between channels. The structure of MLP is illustrated in Fig. 18.3, where the
1 × 1 × D feature map is mapped to yl-1 and serves as the input for layer l. This layer
has n nodes, corresponding weights Xl, bias vectors cl, and al as the output feature
map for layer l. σ represents the activation function, and the expression for the l-th
layer output in MLP is as follows:

αi = σ X i yi - 1 þ ci ð18:2Þ

Squeeze

vd

GAP

vd
Excitation

MLP

Scale Sigmoid

Fig. 18.2 SMLP module

18 Early Crop Disease Identification Using Multi-fork Tree Networks. . . 287

Yl
Yl-1

Yl+1
Yl T1
V1
T2
V2
… … ….
…. ….
Td-1
Vd-1
Td
Vd

Sigmoid

Linear mapping

Fig. 18.3 Schematic diagram of MLP structure

αi = X i yi - 1 þ ci : ð18:3Þ

To represent the nonlinear dependencies between different channels in the feature

map, the Sigmoid function is employed for nonlinear mapping, preserving the
diversity of channel relationships, as shown in Eq. (18.4). This function maps the
nonlinear dependencies between channels to a weight vector s in a multi-
dimensional space, where t,t 2 (0,1). GScale applies s to the input feature map V,
constructing a feature map with weight information.

1
Sigmoid αi = ð18:4Þ
1 þ e - ð αi Þ
Gscale ðt, xÞ = t  x: ð18:5Þ

SMLP adjusts feature information at the channel level, achieving feature

enhancement and recalibration of feature maps.

18.2.2.2 Improved Residual Network Structure (SMLP_ResNet)

Deep networks can extract richer feature information, but as the network depth
increases, issues like gradient vanishing may occur, leading to a decrease in the
recognition accuracy of network models. The residual structure, which incorporates
288 S. S. Ittannavar et al.

Class1

SMLP_Res Class2
299 299 × 3
FC Softmax

CER …..

Conv BN ReLu Class1d

(a) SMLP_ResNet model

Conv 3 × 3
Din = = Dout

BN + ReLu CBR

SMLP Res2
Conv 3 × 3
Conv 1× 1

SMLP

BN

ReLu

(b) SMLP_Res module

Fig. 18.4 SMLP_ResNet model structure diagram. (a) SMLP_ResNet model. (b) SMLP_Res
module

shortcut connections and identity mappings, continuously merges feature informa-

tion from previous layers, effectively alleviating the network degradation caused by
depth and enhancing the feature extraction capability of the network model.
The combination of the residual module and SMLP is defined as the SMLP_Res
residual module. Based on this module, Batch Normalization (Manavalan 2020;
Rani and Singh 2022; Bondad et al. 2023) and ReLU (Rectified Linear Unit)
activation functions, among others, are integrated to construct the SMLP_ResNet
classification network model, as shown in Fig. 18.4a.
18 Early Crop Disease Identification Using Multi-fork Tree Networks. . . 289

Table 18.2 SMLP_ResNet disease model parameter table

Parameter SMLP_ResNet18 SMLP_ResNet50 SMLP_ResNet101
n 2 3 3
m 2 4 4
l 2 6 23
w 2 3 3

In Fig. 18.4a, input images, after data augmentation and resizing to a uniform
size, are fed into the SMLP_ResNet classification network model. SMLP_ResNet
consists of five modules: CBR, SMLP_Res, Fully Connected (FC), pooling layers,
and Softmax. CBR and SMLP_Res perform different levels of disease feature
extraction. During feature extraction, two pooling methods, maximum pooling and
average pooling, are used. Maximum pooling preserves texture features of feature
maps, while average pooling retains the overall feature information of the images.
FC is used to learn feature information from different categories, and the Softmax
function accomplishes the mapping to achieve the final category output.
Figure 18.4b provides the SMLP_Res module, which has two structures:
SMLP_Res1 and SMLP_Res2. SMLP_Res1 represents cases where the input and
output feature map channels are the same, i.e. din = dout, and it directly fuses the
input and output feature maps at the channel level. SMLP_Res2 represents cases
where din ≠ dout. It adjusts the number of channels in the feature map using 1 ×
1 convolutions to make din = dout before merging the feature maps. The parameters
n, m, l, and w are used to specify the quantity of SMLP_Res1 and SMLP_Res2
structures when building the model. Different network depths can be constructed by
setting these four parameters. For example, to build an 18-layer SMLP_ResNet
network model, you can set n = 2, m = 2, l = 2, and w = 2. The specific parameter
settings are listed in Table 18.2.
Table 18.2 lists three commonly used network model depths: 18 layers, 50 layers,
and 101 layers, along with their parameter settings. When applying convolutional
kernels to feature maps, taking a single channel as an example, Eq. (18.6) represents
the convolution operation on the feature map y of channel yc.

f -1 f -1
bi,j = m=0 n=0
wm,n ydiþm,jþn þ wc : ð18:6Þ

Here, bi,j represents the pixel at the i-th row and j-th column of the feature map
after convolution, f is the size of the convolutional kernel, wm,n represents the weight
parameters of the convolutional kernel, and ydiþm,jþn represents the pixel at the (i + m)-
th row and ( j + n)-th column of the input feature channel yc. wc is the bias for fine-
tuning the convolution result. After passing through the Convolution (Conv) layer,
the feature map undergoes normalisation in the Batch Normalization (BN) layer,
mitigating issues like slow convergence and low generalisation caused by sample
distribution. For a batch of sample data β = {y1, ... ,ym},
290 S. S. Ittannavar et al.

1 m
μβ ← i=1
yi ð18:7Þ
m
1 m 2
σ 2β ← yi - μ β ð18:8Þ
m
yi - μ β
yi ← ð18:9Þ
σ 2β þ ε

xi ← γyi þ β  BN γ,β ðyi Þ: ð18:10Þ

Equations (18.7) and (18.8) calculate the mean and variance of the input batch
data, denoted as μβ and σ 2β, respectively. Equation (18.9) is used for data
normalisation, where ε is a correction parameter to prevent division by zero.
However, since xi is constrained below the normal distribution, it affects the
expression of data information. To address this issue, scaling factor γ and shift factor
β are introduced to enhance network expressiveness, as shown in Eq. (18.10). The
feature map processed by the BN layer is activated by the ReLU function, as shown
in Eq. (18.11).

ReLU = maxð0, yÞ, y 2 S: ð18:11Þ

The feature map processed by CBR and BN is then input into the SMLP module
to calibrate the feature map. The choice of SMLP_Res1 or SMLP_Res2 determines
how the feature map x is fused at the channel level. After activation by the ReLU
function, the output feature map x^ combines information from multiple sources,
achieving feature extraction in the SMLP_ResNet network.

18.2.2.3 Disease Recognition Model Based on Multi-branch

SMLP_ResNet

Most crop disease recognition models use separate network models. When the
number of crops and disease types increases, the classification count grows expo-
nentially, increasing the network’s burden and decreasing disease recognition accu-
racy. Crop disease recognition tasks can be decomposed into two classification tasks:
crop recognition and disease recognition. Combining a multi-branch tree model, the
crop recognition model serves as the root node for crop classification, while the
disease recognition model serves as child nodes for recognising diseases in the same
crop. The multi-branch tree model employs multiple classification network models
for early disease recognition, improving disease recognition accuracy.
Classification models implement crop and disease recognition tasks, and a loss
function is used in the classification task to measure the distance between predicted
results and true categories. By adjusting model parameters, the loss function value is
minimised. CNN models output results in vector structure rather than probability
distribution. Therefore, the Softmax function is used to convert the vector into a
18 Early Crop Disease Identification Using Multi-fork Tree Networks. . . 291

probability distribution, and the loss is calculated using the cross-entropy loss
function (Eq. 18.12).

loss = - i
xi ln bi , ð18:12Þ

where bi is the i-th prediction value output by Softmax, and xi represents the
corresponding true classification. The loss value is backpropagated to adjust the
network model parameters. Based on the SMLP_ResNet classification model, the
structure of the early disease recognition model with multi-branch SMLP_ResNet.
The SMLP_ResNet architecture is designed as a hierarchical network consisting
of multiple layers, with each layer representing a tree structure. The first network is
tasked with the identification of different crop varieties; however the subsequent
network is specifically built to acquire knowledge on the detection of diverse
diseases that impact a singular crop. The crop disease data undergoes a hierarchical
processing approach, wherein it is divided into two unique categories: crop type data
and crop disease data. Following this, the data pertaining to the disease is divided
into two sets: a training set and a test set. This division is carried out using a random
partitioning technique, with a ratio of 4:1. The training dataset was employed to train
the multi-tree SMLP_ResNet model. The test dataset was utilised to assess the
efficacy of the trained multi-tree SMLP_ResNet model in detecting the influence
of the model on agricultural crops. The first phase in the procedure entails feeding
augmented images of tomato diseases into the multi-tree SMLP_ResNet model to
facilitate the early detection of illnesses. The first layer of the network is responsible
for recognising the input crop as a tomato and then relaying this information to the
subsequent layer of the network. The secondary network assumes the responsibility
of detecting many health issues, including both the initial and advanced phases of
powdery mildew disease. In order to assess the precision of the multi-tree
SMLP_ResNet model’s predictions, the cross-entropy function is utilised to quantify
the disparity between the true label assigned to a picture and the label projected by
the model. The network parameters are subsequently modified utilising the gradient
descent technique. When the observation of losing value occurs, the model is stored
when reaching a predetermined threshold. To boost the sickness detection network’s
potential for generalisation, a series of image processing techniques are applied,
including augmentation and standardisation. These techniques result in a consistent
image size of 224 × 224 × 3. The augmented dataset is employed as input to train the
multi-tree SMLP_ResNet network model for the aim of model development. The
SMLP_ResNet network model is improved through the integration of several trees,
hence enhancing the network’s capacity to extract features. This functionality allows
individual sub-networks to effectively manage distinct learning tasks, leading to
enhanced accuracy in total network recognition. Furthermore, the proposed model
improves the precision of early illness categorisation and minimises the influence of
feature similarity and other variables on disease identification accuracy through the
reduction of data kinds and quantity.
292 S. S. Ittannavar et al.

18.3 Experiment

The crop disease recognition model in this experiment is trained using an NVIDIA
Tesla P100 model GPU. The video memory capacity is 16 GB. The construction of a
training platform for deep learning algorithms is founded upon the utilisation of the
Ubuntu 16.04 64-bit operating system and the PyTorch frameworks. The current
version of Python is 3.7.6. The PyTorch version utilised is 1.3.0, while the CUDA
API version employed is 10.0. Additionally, the CuDNN version employed is 7.5.1.

18.3.1 Experimental Dataset and Parameter Settings

The SMLP_ResNet, ResNet, and SE_ResNet models, each consisting of 18 layers,

are trained using Plant Village and AI Challenger 2018 datasets. The purpose of this
training is to evaluate the influence of the SMLP_Res module on the capacity to
recognise diseases. The AI Challenger 2018 dataset is utilised to train two network
models: the multi-tree SMLP_ResNet and the multi-tree ResNet with identical
structures. Additionally, the SE_ResNet network is introduced. The performance
of the multi-tree SMLP_ResNet network model in disease recognition is evaluated
by comparing it with three disease identification models consisting of 18 layers,
including the multi-tree disease identification model. In order to enhance the training
efficiency of the model, the utilisation of the gradient descent technique, specifically
Stochastic Gradient Descent (SGD), is employed to expedite the convergence of the
loss function and determine the optimal network weight. Through a series of iterative
experiments, the initial learning rate is established at 0.05, while the momentum
factor is set to 0.9. Considering the limited video memory capacity of the graphics
card, which amounts to 16 GB, network architectures consisting of 18, 50, and
101 layers are utilised. The batch size is configured as 256, 128, and 64, correspond-
ingly. Each training sample is iterated once, with a total of 101 iteration rounds, also
known as epochs. The neural network undergoes training through the implementa-
tion of adaptive decrease of the learning rate in order to achieve the most optimal
model. The evaluation of the crop disease network model involves assessing its
average accuracy and precision, utilising evaluation metrics such as rate, recall rate,
and weighted average score F1.

18.3.2 Disease Identification Experiment Based

on SMLP_ResNet

The training results of the SMLP_ResNet network were compared with the ResNet
and SE_ResNet network models to test the performance of the SMLP_ResNet
network model. Three different network models were trained on the Plant Village
18 Early Crop Disease Identification Using Multi-fork Tree Networks. . . 293

Table 18.3 Comparison of disease identification methods

Evaluation Plant Village dataset AI Challenger 2018 dataset

index ResNet SE_ResNet SMLP_ResNet ResNet SE_ResNet SMLP_ResNet
Accuracy 99.05 99.19 99.32 83.83 85.53 86.93
Precision 98.57 99 99.1 79.8 82.64 84.15
Recall 98.52 98.7 98.78 78.16 81.77 83.42
F1 score 98.53 98.87 98.92 78.52 81.92 83.6

and AI Challenger 2018 datasets, and the experimental results are listed in
Table 18.3.
The analysis of the data shown in Table 18.3 reveals that the SMLP_ResNet
model had superior performance compared to the other models across both datasets,
under identical experimental settings. The SMLP_ResNet model exhibited an
increase in accuracy ranging from 0.13 to 0.27% when compared to the ResNet
and SE_ResNet models in the Plant Village dataset. The SMLP_ResNet model
exhibited a notable level of precision, with a rate of 99.32%. Furthermore, it
demonstrated a higher level of accuracy with a precision rate of 99.10%, recall
rate of 98.78%, and an F1 score of 98.91% in comparison to the other two models.
Significantly, a marginal disparity of 1% in accuracy was observed between the
aforementioned models. The SMLP_ResNet model exhibited significant improve-
ments in precision, recall, and F1 score when compared to the ResNet model, leading
to an estimated increase of 5%. The experimental results suggest that the SE_ResNet
network architecture, which integrates the SE module, exhibits enhanced perfor-
mance in comparison to the ResNet model. In addition, the SMLP_ResNet model,
which combines the SMLP module with residual structures, demonstrated higher
recognition results. This observation implies that the SMLP_Res architecture
enhances the capacity for feature extraction to a certain extent.
ResNet, SE_ResNet, and SMLP_ResNet achieved accuracy rates exceeding 99%
in the domain of crop disease recognition, as observed in the utilisation of the Plant
Village dataset. Nevertheless, it is important to acknowledge that under the AI
Challenger 2018 dataset, the accuracy rates of the three network models did not
exceed 87%. This implies that disease recognition has a comparatively diminished
level of accuracy. The observed discrepancy can potentially be ascribed to the
composition of the Plant Village dataset, which has a total of 38 distinct types of
crop diseases. Every individual sample in this dataset displays distinct illness
characteristics that are easily distinguishable through visual observation. The dataset
utilised in the AI Challenger 2018 competition comprises a total of 59 unique
classifications, hence facilitating a comparative analysis. Moreover, the delineations
between diseases in their initial stages and leaves in a state of good health exhibit a
considerably diminished level of prominence. Therefore, it is evident that the
accuracy of all three models in disease recognition on the AI Challenger 2018
early disease dataset is relatively lower.
294 S. S. Ittannavar et al.

18.3.3 Early Disease Recognition Experiment Based

on Multi-branch SMLP_ResNet

Accurate crop disease recognition in the Plant Village dataset was obtained by
employing the 18-layer ResNet, SE_ResNet, and SMLP_ResNet network models.
Nevertheless, the level of accuracy in disease recognition within the early disease
sample, specifically AI Challenger 2018, was suboptimal. The present study
involved the construction of a crop early disease recognition network model utilising
a multi-branch structure. The experimental outcomes were analysed to investigate
the influence of the multi-branch structure on the accuracy of the model.
In order to examine the impact of the multi-branch structure model on early crop
disease detection, this study employed three network models, namely ResNet,
SE_ResNet, and SMLP_ResNet, as the foundational frameworks for constructing
similar structured multi-branch models. The three crop disease recognition models
underwent training using the AI Challenger 2018 dataset, following identical set-
tings. The analysis focused on examining the impact of multi-branch structure
network models on model performance by evaluating the findings of early disease
recognition in crops. The figure presented in Fig. 18.5 demonstrates the level of
precision achieved in the early detection of diseases by the three multi-branch
models. The ResNet network model’s multi-branch topology showed reduced accu-
racy in disease recognition across all crops. In contrast, the SMLP_ResNet network
model with several branches consistently demonstrated superior accuracy compared
to the other two networks. The multi-branch SMLP_ResNet network model showed
notable enhancement in accurately identifying crop diseases in apple, cherry, pepper,
and pear crops. The model achieved a maximum accuracy of 99.13%. Upon
analysing the outcomes pertaining to the identification of diseases in grape and
strawberry plants, it was observed that the ResNet network model exhibited the
least accuracy in recognition, amounting to 76.4%. This discrepancy was

Fig. 18.5 Crop disease recognition accuracy

18 Early Crop Disease Identification Using Multi-fork Tree Networks. . . 295

Fig. 18.6 Accuracy of different models in the AI Challenger 2018 dataset

particularly evident when compared to the SE_ResNet network model. In relation to

the classification of citrus diseases, it was observed that all three network models
showed relatively low identification accuracy, with none above 78%. Among these
models, the multi-branch SMLP_ResNet network model achieved the highest rec-
ognition accuracy, reaching 77.66%.
Analysing the experimental data reveals that the ResNet network model has
relatively weak feature extraction capabilities, resulting in a significant difference
in disease recognition accuracy compared to the other two network models. In
contrast, the SE_ResNet network model, which incorporates an attention mecha-
nism, extracts crop disease data features more thoroughly, leading to a noticeable
improvement in disease recognition accuracy. The improved attention mechanism in
the multi-branch SMLP_ResNet network model effectively suppresses noise trans-
mission and efficiently extracts early disease feature information, resulting in higher
accuracy. Figure 18.6 provides a comparison of the average accuracy of the three
18-layer disease recognition models and the multi-branch disease recognition
models.
All multi-branch illness recognition models have an accuracy of 87% or above,
with the highest average accuracy achieved by the multi-branch SMLP_ResNet
disease recognition model (average accuracy: 90.1%). The AI Challenger 2018
dataset shows that the multi-branch structure network models significantly
outperform the three 18-layer disease recognition models in terms of early crop
disease recognition accuracy. Experimental results show that multi-branch ResNet,
SE_ResNet, and SMLP_ResNet disease recognition models on the AI Challenger
2018 dataset significantly improve accuracy and meet expectations for multi-crop
early disease recognition compared to the three 18-layer disease recognition models.
In terms of accuracy in early disease recognition across many crops, the experimen-
tal results show that the multi-branch SMLP_ResNet network model performs better
than the multi-branch ResNet network model and the multi-branch SE_ResNet
network model.
296 S. S. Ittannavar et al.

18.4 Conclusion

The identification of crop diseases has a substantial influence on agricultural pro-

ductivity. The prompt and targeted response to address crop health issues is facili-
tated by the early detection of crop diseases. The issue pertaining to the early
identification of diseases in many crops presents an even greater level of complexity.
Datasets such as Plant Village and AI Challenger 2018 are frequently employed in
the training of disease recognition models due to their incorporation of diverse crops
and their corresponding prevalent diseases. In contrast to the Plant Village dataset,
the AI Challenger 2018 dataset encompasses a greater number of categories and a
substantial quantity of early illness samples. Consequently, this phenomenon
imposes greater requirements on the network’s ability to extract features, thereby
augmenting the intricacy of crop disease recognition assignments. An enhanced
attention mechanism multi-branch crop early disease recognition model was devel-
oped by using the SE attention mechanism and a multi-branch structure. Initially, the
SMLP_Res module was established with the aim of augmenting the disease model’s
feature extraction capabilities. The application of the attention mechanism to multi-
channel disease data involved the utilisation of MLP analysis to quantify various
disease picture channel properties. This approach aimed to enhance the expression of
significant features while simultaneously inhibiting the transmission of noise infor-
mation. Furthermore, the integration of Residual (Res) structures was employed to
establish connections across various layers, hence enhancing the transfer of cross-
layer feature information and mitigating the loss of features during the transmission
process. In order to address the complexity of crop disease recognition, a two-level
multi-branch recognition model was developed. This model effectively simplifies
the problem by dividing it into separate tasks of crop recognition and disease
recognition. The utilisation of this approach resulted in a decrease in the learning
burden of individual models, leading to an enhancement in the accuracy of early
disease recognition through multi-crop techniques. Based on the empirical findings
obtained from the 18-layer models and the multi-branch models, it can be deduced
that the SMLP_Res module demonstrates a notable capability in extracting illness
features. The disease recognition accuracy achieved on the AI Challenger 2018
dataset attained a peak value of 86.93%, exhibiting a 3% enhancement in compar-
ison to the alternative two models. On the other hand, the disease recognition model
within the multi-branch model had the highest level of accuracy, reaching 90.07%.
The study demonstrated enhancements in the management of many crop diseases,
providing evidence of the efficacy and resilience of SMLP_ResNet. The implemen-
tation of a multi-branch model in constructing a disease recognition model has been
found to significantly enhance the accuracy of crop disease recognition.
18 Early Crop Disease Identification Using Multi-fork Tree Networks. . . 297

References

Abbasi R, Martinez P, Ahmad R (2023) Crop diagnostic system: a robust disease detection and
management system for leafy green crops grown in an aquaponics facility. Artif Intell Agric 10:
1–12. https://doi.org/10.1016/j.aiia.2023.09.001
Agarwal M, Gupta SK, Biswas KK (2020) Development of efficient CNN model for tomato crop
disease identification. Sustain Comput Inform Syst 28:100407. https://doi.org/10.1016/j.
suscom.2020.100407
Bondad J, Harrison MT, Whish J, Sprague S, Barry K (2023) Integrated crop-disease models: new
frontiers in systems thinking. Farm Syst 1(1):100004. https://doi.org/10.1016/j.farsys.2023.
100004
Das KP, Sharma D, Satapathy BK (2022) Electrospun fibrous constructs towards clean and
sustainable agricultural prospects: SWOT analysis and TOWS based strategy assessment. J
Clean Prod 368:133137. https://doi.org/10.1016/j.jclepro.2022.133137
De Roever C (1998) Microbiological safety evaluations and recommendations on fresh produce.
Food Control 9(6):321–347. https://doi.org/10.1016/S0956-7135(98)00022-X
Degu T, Alemu T, Desalegn A, Amsalu B, Assefa A (2023) Association of cropping practices,
cropping areas, and foliar diseases of common bean (Phaseolus vulgaris L.) in Ethiopia. J Agric
Food Res 14:100765. https://doi.org/10.1016/j.jafr.2023.100765
Dutta K, Talukdar D, Bora SS (2022) Segmentation of unhealthy leaves in cruciferous crops for
early disease detection using vegetative indices and Otsu thresholding of aerial images. Mea-
surement 189:110478. https://doi.org/10.1016/j.measurement.2021.110478
Fernández-Ríos A, Laso J, Hoehn D, Amo-Setién FJ, Abajas-Bustillo R, Ortego C, Fullana-i-
Palmer P, Bala A, Batlle-Bayer L, Balcells M, Puig R, Aldaco R, Margallo M (2022) A critical
review of superfoods from a holistic nutritional and environmental approach. J Clean Prod 379
(Part 1):134491. https://doi.org/10.1016/j.jclepro.2022.134491
Gao R, Wang R, Feng L, Li Q, Wu H (2021) Dual-branch, efficient, channel attention-based crop
disease identification. Comput Electron Agric 190:106410. https://doi.org/10.1016/j.compag.
2021.106410
Gavidia JC, Chinelatto GF, Basso M, da Ponte Souza JP, Soltanmohammadi R, Vidal AC,
Goldstein RH, Mohammadizadeh S (2023) Utilizing integrated artificial intelligence for char-
acterizing mineralogy and facies in a pre-salt carbonate reservoir, Santos Basin, Brazil, using
cores, wireline logs, and multi-mineral petrophysical evaluation. Geoenergy Sci Eng 231:
212303. https://doi.org/10.1016/j.geoen.2023.212303
Harner I, Anast J, Brehm-Stecher B (2023) Food safety applications of genomic technologies. In:
Reference module in food science. Elsevier. https://doi.org/10.1016/B978-0-12-822521-9.
00202-1
Holzinger A, Keiblinger K, Holub P, Zatloukal K, Müller H (2023) AI for life: trends in artificial
intelligence for biotechnology. New Biotechnol 74:16–24. https://doi.org/10.1016/j.nbt.2023.
02.001
Islam MM, Adil MA, Talukder MA, Ahamed MK, Uddin MA, Hasan MK, Sharmin S, Rahman
MM, Debnath SK (2023) DeepCrop: Deep learning-based crop disease prediction with web
application. J Agric Food Res 14:100764. https://doi.org/10.1016/j.jafr.2023.100764
Jagadeesan B, Gerner-Smidt P, Allard MW, Leuillet S, Winkler A, Xiao Y, Chaffron S, Van Der
Vossen J, Tang S, Katase M, McClure P, Kimura B, Chai LC, Chapman J, Grant K (2019) The
use of next generation sequencing for improving food safety: translation into practice. Food
Microbiol 79:96–115. https://doi.org/10.1016/j.fm.2018.11.005
Joseph A (2023) Chapter 6: Salinity tolerance of inhabitants in thalassic and athalassic saline and
hypersaline waters and salt flats. In: Joseph A (ed) Water worlds in the solar system. Elsevier, pp
255–310. https://doi.org/10.1016/B978-0-323-95717-5.00017-7
Karanth S, Benefo EO, Patra D, Pradhan AK (2023) Importance of artificial intelligence in
evaluating climate change and food safety risk. J Agric Food Res 11:100485. https://doi.org/
10.1016/j.jafr.2022.100485
298 S. S. Ittannavar et al.

Kaveney B, Barrett-Lennard E, Minh KC, Duy MD, Thi KPN, Kristiansen P, Orgill S, Stewart-
Koster B, Condon J (2023) Inland dry season saline intrusion in the Vietnamese Mekong River
Delta is driving the identification and implementation of alternative crops to rice. Agric Syst
207:103632. https://doi.org/10.1016/j.agsy.2023.103632
Kendler S, Aharoni R, Young S, Sela H, Kis-Papo T, Fahima T, Fishbain B (2022) Detection of
crop diseases using enhanced variability imagery data and convolutional neural networks.
Comput Electron Agric 193:106732. https://doi.org/10.1016/j.compag.2022.106732
Kutyauripo I, Rushambwa M, Chiwazi L (2023) Artificial intelligence applications in the agrifood
sectors. J Agric Food Res 11:100502. https://doi.org/10.1016/j.jafr.2023.100502
Maier-Hein L, Eisenmann M, Sarikaya D, März K, Collins T, Malpani A, Fallert J, Feussner H,
Giannarou S, Mascagni P, Nakawala H, Park A, Pugh C, Stoyanov D, Vedula SS, Cleary K,
Fichtinger G, Forestier G, Gibaud B, Grantcharov T, Hashizume M, Heckmann-Nötzel D,
Kenngott HG, Kikinis R, Mündermann L, Navab N, Onogur S, Roß T, Sznitman R, Taylor
RH, Tizabi MD, Wagner M, Hager GD, Neumuth T, Padoy N, Collins J, Gockel I, Goedeke J,
Hashimoto DA, Joyeux L, Lam K, Leff DR, Madani A, Marcus HJ, Meireles O, Seitel A,
Teber D, Ückert F, Müller-Stich BP, Jannin P, Speidel S (2022) Surgical data science—from
concepts toward clinical translation. Med Image Anal 76:102306. https://doi.org/10.1016/j.
media.2021.102306
Manavalan R (2020) Automatic identification of diseases in grains crops through computational
approaches: a review. Comput Electron Agric 178:105802. https://doi.org/10.1016/j.compag.
2020.105802
Marvin HJP, Kleter GA, Frewer LJ, Cope S, Wentholt MTA, Rowe G (2009) A working procedure
for identifying emerging food safety issues at an early stage: implications for European and
international risk management practices. Food Control 20(4):345–356. https://doi.org/10.1016/
j.foodcont.2008.07.024
Mensah PK, Akoto-Adjepong V, Adu K, Ayidzoe MA, Bediako EA, Nyarko-Boateng O,
Boateng S, Donkor EF, Bawah FU, Awarayi NS, Nimbe P, Nti IK, Abdulai M, Adjei RR,
Opoku M, Abdulai S, Amu-Mensah F (2023) CCMT: dataset for crop pest and disease
detection. Data Brief 49:109306. https://doi.org/10.1016/j.dib.2023.109306
Modi RU, Kancheti M, Subeesh A, Raj C, Singh AK, Chandel NS, Dhimate AS, Singh MK, Singh
S (2023) An automated weed identification framework for sugarcane crop: a deep learning
approach. Crop Prot 173:106360. https://doi.org/10.1016/j.cropro.2023.106360
Mrabet R (2023) Chapter 2: Sustainable agriculture for food and nutritional security. In: Farooq M,
Gogoi N, Pisante M (eds) Sustainable agriculture and the environment. Academic Press, pp
25–90. https://doi.org/10.1016/B978-0-323-90500-8.00013-0
Nigam S, Jain R, Marwaha S, Arora A, Haque MA, Dheeraj A, Singh VK (2023) Deep transfer
learning model for disease identification in wheat crop. Ecol Informat 75:102068. https://doi.
org/10.1016/j.ecoinf.2023.102068
Noshiri N, Beck MA, Bidinosti CP, Henry CJ (2023) A comprehensive review of 3D convolutional
neural network-based classification techniques of diseased and defective crops using non-UAV-
based hyperspectral images. Smart Agric Technol 5:100316. https://doi.org/10.1016/j.atech.
2023.100316
Nychas G-JE, Panagou EZ, Mohareb F (2016) Novel approaches for food safety management and
communication. Curr Opin Food Sci 12:13–20. https://doi.org/10.1016/j.cofs.2016.06.005
Pakseresht A, Yavari A, Kaliji SA, Hakelius K (2023) The intersection of blockchain technology
and circular economy in the agri-food sector. Sustain Prod Consump 35:260–274. https://doi.
org/10.1016/j.spc.2022.11.002
Rani AP, Singh NS (2022) Protecting the environment from pollution through early detection of
infections on crops using the deep belief network in paddy. Total Environ Res Themes 3:
100020. https://doi.org/10.1016/j.totert.2022.100020
Ren Q-S, Fang K, Yang X-T, Han J-W (2022) Ensuring the quality of meat in cold chain logistics: a
comprehensive review. Trends Food Sci Technol 119:133–151. https://doi.org/10.1016/j.tifs.
2021.12.006
18 Early Crop Disease Identification Using Multi-fork Tree Networks. . . 299

Reumaux R, Chopin P, Bergkvist G, Watson CA, Öborn I (2023) Land Parcel Identification System
(LPIS) data allows identification of crop sequence patterns and diversity in organic and
conventional farming systems. Eur J Agron 149:126916. https://doi.org/10.1016/j.eja.2023.
126916
Rowan NJ, Galanakis CM (2020) Unlocking challenges and opportunities presented by COVID-19
pandemic for cross-cutting disruption in agri-food and green deal innovations: Quo Vadis? Sci
Total Environ 748:141362. https://doi.org/10.1016/j.scitotenv.2020.141362
Rowan NJ, Murray N, Qiao Y, O'Neill E, Clifford E, Barceló D, Power DM (2022) Digital
transformation of peatland eco-innovations (‘Paludiculture’): enabling a paradigm shift towards
the real-time sustainable production of ‘green-friendly’ products and services. Sci Total Environ
838(Part 3):156328. https://doi.org/10.1016/j.scitotenv.2022.156328
Saeed R, Feng H, Wang X, Zhang X, Fu Z (2022) Fish quality evaluation by sensor and machine
learning: a mechanistic review. Food Control 137:108902. https://doi.org/10.1016/j.foodcont.
2022.108902
Tirkey D, Singh KK, Tripathi S (2023) Performance analysis of AI-based solutions for crop disease
identification, detection, and classification. Smart Agric Technol 5:100238. https://doi.org/10.
1016/j.atech.2023.100238
Varga M, Csukas B (2022) Foundations of programmable process structures for the unified
modeling and simulation of agricultural and aquacultural systems. In: Information processing
in agriculture. Elsevier. https://doi.org/10.1016/j.inpa.2022.10.001
Xiong Y, Liang L, Wang L, She J, Wu M (2020) Identification of cash crop diseases using
automatic image segmentation algorithm and deep learning with expanded dataset. Comput
Electron Agric 177:105712. https://doi.org/10.1016/j.compag.2020.105712
Zhao H, Huang X, Yang Z, Li F, Ge X (2023) Synergistic optimization of crops by combining early
maturation with other agronomic traits. Trends Plant Sci 28(10):1178–1191. https://doi.org/10.
1016/j.tplants.2023.04.011
Zhou J, Li J, Wang C, Wu H, Zhao C, Teng G (2021) Crop disease identification and interpretation
method based on multimodal deep learning. Comput Electron Agric 189:106408. https://doi.
org/10.1016/j.compag.2021.106408
Chapter 19
Guarding Maize: Vigilance Against
Pathogens Early Identification, Detection,
and Prevention

Khalil Ahmed, Mithilesh Kumar Dubey, and Sudha Dubey

Abstract Maize, also known as corn, is a crucial crop worldwide. It is noticeable

that due to diseases maize production is affected. Maize diseases can have a
significant impact on yield. Common diseases include those caused by viruses,
bacteria, fungi, and other pathogens. Proper artificial intelligence-based crop man-
agement and disease control are essential for maximizing maize production. Corn
leaf stage methods are used to assess the growth and development of maize plants.
These stages are often described using a system called the V-stage or “the Vegetative
stage.” Corn is a widely cultivated cereal crop that has a rich history dating back
7000 years, originating from wild grasses in Central America and having significant
cultural, economic, and nutritional importance worldwide. Around the globe, India
is the fifth largest maize (corn) producer according to the USDA 2022–2023 report,
which produces 3,20,00,000 million metric tons of corn annually. The United States
holds the first position with 35,38,36,000 million metric tons of production.
According to the USDA report maize production decreased by 1216.87 million
tons last year. The prophecy indicated that in 2023 maize production also decreased
by 1216.87 million tons, which is 4.52% of the globe. This continuous decrease in
maize crops is due to diseases caused by bacteria, fungi, and viruses. Corn seed
undergoes various stages from germination to the dotage stage, absorbing water and
other essential nutrients for growth. Due to climatic conditions corn plants suffer
from such diseases, identification of disease and detection at the prior stage are very
important so that quality and yield of maize crops are not affected. For this, we
introduce the anatomy of maize plants, the key features of a crop, leaf stage and

K. Ahmed · M. K. Dubey (✉)

School of Computer Application, Lovely Professional University, Phagwara, Punjab, India
e-mail: [email protected]
S. Dubey
Department of Sociology, Lovely Professional University, Phagwara, Punjab, India

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 301
A. Khamparia et al. (eds.), Microbial Data Intelligence and Computational
Techniques for Sustainable Computing, Microorganisms for Sustainability 47,
https://doi.org/10.1007/978-981-99-9621-6_19
302 K. Ahmed et al.

methods, the maturity cycle of maize plants, and a brief overview of the diseases
caused by fungi, bacteria, and viruses in maize.

Keywords Maize disease · Leaf stage · Early identification · Zea mays

19.1 Introduction

Maize, scientifically known as Zea mays, is a widely cultivated cereal crop that has
significant cultural, economic, and nutritional importance around the world
(Hallauer and Carena 2009). The history of the maize crop spans thousands of
years. Originating 7000 years ago from wild grasses in Central America (McHale
et al. 2012a), the areas in which ancient civilizations such as the Maya, Inca, and
Aztecs engaged in the farming of maize, it has a long history of cultivation by
indigenous people in America. It was a vital crop for the ancient civilization of
America. It is an annual plant with a distinctive morphology, including tall, jointed
stems, long leaves arranged alternatively, distinctive cobs with rows of kernels, and a
tassels-like inflorescence known as the male flower. The female flower is enclosed in
a husk, later known as corn or maize kernels (Zhang et al. 2012) as shown in
Fig. 19.1.
Tessel (Male florescence)

Tassel
Central spike
Lateral braanch Node

Peduncle Flag leaf

(Female Floresence)

Leaf blade
Stilt roots
Silk Ligule
Husks

s
Mid-rib Leaf sheath

Inter Node

Node Stilt roots

Soil surface

Corn Husk
Corn Cub

Root system

Fig. 19.1 Anatomy and morphology of maize plant

19 Guarding Maize: Vigilance Against Pathogens Early Identification,. . . 303

Maize is grown in extensive climates and geographical regions, one of the most
often used cultivated crops around the globe. The major maize-producing countries
include the United States, China, Brazil, and India (Ranum et al. 2014). The crop’s
adaptability and high-yield potential have contributed to its widespread cultivation,
providing essential carbohydrates, dietary fibers (B-complex vitamins, especially
niacin and thiamine), and some protein. Despite human use, it is as a raw product,
and ethanol production, and in the manufacturing of industrial products. Maize is the
stable food for billions of people worldwide. This diversity has prompted the
creation of numerous sorts of maize, including tremendous genetic diversity helps
to the creation of different varieties of maize plus dent, flint, popcorn, and sweet
corn, each with specific attributes and uses. Genetic diversity also contributes to the
plant’s adaptability and resilience to different environmental conditions (McHale
et al. 2012b). Successful maize cultivation involves careful consideration of factors
such as soil types, temperature, availability of water, and disease management.
Proper planting, fertilization, irrigation, and disease control are critical for achieving
optimal yields.

19.1.1 Key Features of Maize

Versatility Maize is versatile and can be used in various forms, such as fresh, dried,
or processed (FAOSTAT 2023). It can be consumed as whole kernels, ground into
flour, or used to produce various food products like corn syrup, corn oil, and
popcorn.
Cultural Significance Maize has a deep cultural and historical significance for
many indigenous communities in the Americas (Ranum et al. 2014). It played a
central role in the diet, rituals, and economies of these civilizations.
Botanical Structure A maize plant consists of different parts. The ear is the female
inflorescence where the kernels develop, while the tassel is the male inflorescence
that produces pollen as in Fig. 19.1. Maize plants have long, narrow leaves and a
fibrous root system (Doebley et al. 2006). The fibrous root system maximizes the
plant’s ability to access soil nutrients, contributing to its overall growth and produc-
tivity. These leaves are usually around 30–100 cm in length, depending on factors
such as genetics, environmental conditions, and the level of growth.
Cultivation and Climatic Conditions Maize is grown in a wide range of climates,
from tropical to temperate regions. It requires well-drained soil and sufficient
sunlight. It is considered a C4 plant (Sadras and Calderini 2009), which means it
is efficient at photosynthesis, making it well suited for environments with higher
temperatures.
Economic Importance Maize is a staple crop in many countries, serving as a major
food source for living. It also has industrial applications in sectors such as food
processing, animal feed, and biofuel production.
304 K. Ahmed et al.

19.1.2 A Closer Look at World Maize Harvesting

The world corn production in the years 2022–2023 will be 1161.86 million metric
tons around the globe according to data from the USDA. According to USDA data,
in December 2022, the maize crop was less by 6.52 million metric tons than the
previous month. In the last year maize production decreased by 1216.87 million tons
(Dotasara and Choudhary 2023). The forecast for the end of 2023 anticipates a
global maize production decrease of 55.00 million tons, equivalent to a 4.52%
reduction. Statistical data of maize production is shown in Fig. 19.2.
India is the fifth largest impresario of maize in the world. Maize is grown in
various states across the country, contributing significantly to both food and fodder
requirements. Maize is cultivated in the Kharif season in India, which starts with the
onset of the monsoon (June/July) and extends until October, and during winter (rabi)
in October or November. The crops are then harvested in spring. In India, Kharif
maize accounts for approximately 83% of total maize cultivation, whereas rabi
maize makes up the remaining 17%. More than 70% of Kharif maize is grown
without irrigation, relying solely on rainwater, which exposes it to a range of pests
and environmental challenges as shown in Fig. 19.3. The exact timing of maize
harvesting can vary based on the region and local climate conditions. The production
of maize in India has been increasing steadily over the years due to its diverse uses.
Almost 70% of the rural population in India is directly dependent on agriculture,
which contributes 50% of in-country employment and 18% of the country’s GDP
(Monterrubio-Solís et al. 2023). Around the globe four billion people are indirectly
dependent on maize for food as per (FAO) 2021 report. The maize crop is facing
significant challenges in terms of reduced production due to the outbreak of various
diseases worldwide. These diseases have the potential to cause extensive damage to
maize crops, leading to yield losses and economic setbacks. Maize is a high-yield
crop, and farmers must protect the plant while growing and avoid production
damage.

Fig. 19.2 Maize growth and yield data 2022–2023 (USDA)

19 Guarding Maize: Vigilance Against Pathogens Early Identification,. . . 305

Fig. 19.3 A map of maize production worldwide in the years 2022–2023 (USDA)

Fig. 19.4 Methods of maize leaf staging

19.1.3 Determining Corn Leaf Stages

Corn leaf stages refer to the various growth stages that a corn plant goes through
during its development (Forde and Clarkson 1999). Farmers and agronomists fre-
quently utilize these stages to monitor and control crop growth and development.
The most commonly used system to describe corn leaf stages is the one introduced
by the USDA, which uses a numerical scale from VE (emergence) to R6 (physio-
logical maturity) as shown in Fig. 19.4 (Smith and Zeeman 2006).

19.1.3.1 Leaf Collar Method

This approach involves determining the developmental stage of corn leaves by

tallying the count of the plant leaves that exhibit noticeable leaf collars. The process
commences with the smallest genuine leaf characterized by its short length and
rounded tip and concludes with the highest leaf that possesses a discernible leaf
306 K. Ahmed et al.

5th Leaf collar

Emerging Leaf

3rd Leaf collar

4th Leaf collar

2nd Leaf collar

First Leaf collar

with round tip

Fig. 19.5 Leaf collar method v5

collar. The leaf collar, resembling a light-colored band, is situated toward the base of
an exposed leaf blade point of contact between the leaf blade and the plant’s stem
Leaves positioned within the cluster, which have not yet fully unfurled and lack an
observable leaf collar, are not taken into account using this leaf staging technique
(Kumar et al. 2022). This is typically denoted as “V” stages, for example, V2,
signifying two leaves with visible leaf collars. In the United States, the leaf collar
method is widely favored and extensively employed as shown in Fig. 19.5.

19.1.3.2 Droopy Leaf Method

This form of leaf staging, like the leaf collar approach, starts with a short initial leaf.
Leaf counting then changes, concluding with the leaf that is at least 40–50% exposed
from the whorl, rather than the highest leaf with a visible collar. The tip of this
“indicator” leaf generally “droops” or hangs down in knee-high corn or older, thus I
refer to this as the “droopy” leaf approach (Gao et al. 2023).

19.1.3.3 Leaf Tip Method

Tally, the points of the leaflet edges, starts from the plant’s base and moves upward.
Be sure to include any recently sprouted leaves that emerge among the group. Be
cautious when applying the leaf tip technique, as various hybrid plants exhibit
unique ways of growing (ur Rehman et al. 2021). For example, certain hybrids are
deliberately cultivated to have shorter spaces between leaves, leading to a higher
count of the leaflet edges prior to their curvature and transformation into fully mature
leaves.
19 Guarding Maize: Vigilance Against Pathogens Early Identification,. . . 307

19.1.3.4 Corn Height Method

In this method the vertical distance from the base level to the highest point of the
downward-pointing tip on the uppermost leaf of the arch is calculated. It’s crucial to
understand that we are not measuring the highest point of the entire plant’s structure
(Naseem et al. 2022a). Please remember that utilizing this method to assess the
developmental phase of corn may not be highly accurate, as the crop’s height can
vary due to factors in the environment and how it’s managed.

19.1.4 The Growth Cycle of the Maize Plant

The maize plant undergoes several distinct stages during its life cycle as different
varieties and environmental conditions can slightly alter the timing of these stages.
On average, it generally requires 100–120 days for the majority of maize cultivars to
mature from planting to harvest. However, the environment and other circumstances,
such as the product’s intended purpose, have a significant impact on the precise time
of harvest.
Seed Germination The life cycle begins when a maize seed germinates as shown in
Fig. 19.6. The seed absorbs water from the soil moisture, the seed becomes soft, and
the diameter also increases. Due to the softness, the germination period of maize
seed takes 4–7 days, and the radicle emerges first followed by the shoot (Maitra et al.
2020).

Fig. 19.6 Seed germination and radicle root emerging process

308 K. Ahmed et al.

Vegetative Growth The young seedling develops leaves, and the stem elongates.
The plant focuses on establishing a strong root system and growing above-ground
structures to capture sunlight for photosynthesis (Naseem et al. 2022b).
Tillering As the plant grows, it produces side shoots known as tillers. Each tiller
has its own set of leaves and can potentially form ears of corn.
Stem Elongation The stem continues to grow, and the plant gains height. More
leaves are produced at the top of the plant.
Pre-flowering During this phase, the plant transitions from vegetative growth to
reproductive growth. The tassel, which contains the male flowers (anthers), emerges
from the top of the plant.
Silking Female reproductive structures, called silk, emerge from the ear of corn.
Each silk corresponds to a potential kernel. Pollination occurs when pollen from the
tassel lands on the silks.
Pollination The wind carries pollen from the tassel to the silks. Each pollen grain
that lands on a silk forms a pollen tube that travels down the silk to fertilize an
individual ovule, leading to kernel development.
Kernel Development Fertilized ovules develop into kernels. Each kernel
undergoes stages of growth, starting with a milky fluid and progressing to a
dough-like consistency.
Grain Filling Kernels continue to grow and accumulate starch, sugars, and other
nutrients (World Agricultural Production 2023). The plant’s focus shifts to providing
nutrients to the developing kernels.
Maturity The maize plant reaches maturity when the kernels have fully developed
and dried. The leaves may start to turn brown, and the plant’s energy is concentrated
in the mature kernels.
Harvest Once the maize reaches the desired moisture content and maturity level,
it’s ready for harvest as shown in Fig. 19.7.
Senescence The plant’s remaining vegetative components begin to wither and die
after harvest. This completes the life cycle of the maize plant. Please keep in mind
that the timing of these stages can vary based on factors such as temperature, day
length, and local growing conditions. Additionally, different varieties of maize may
exhibit slight variations in their growth patterns.

19.2 Overview of Maize Crop Diseases

The most common disease in maize crops can be caused by fungi, bacteria, viruses,
and mollicutes, which can result in the most prevalent diseases in maize (Santos et al.
2022). Microorganisms can spread diseases such as rust, smut, and blight in maize.
19 Guarding Maize: Vigilance Against Pathogens Early Identification,. . . 309

Fig. 19.7 The maize corn life cycle

Fig. 19.8 Maize disease categorization

Bacteria are unicellular microorganisms that can cause diseases such as bacterial wilt
and leaf spot. Viruses are infectious agents that cause diseases, e.g., maize dwarf
mosaic virus and maize streak virus. Mollicutes are a type of bacteria that lack a cell
wall and can cause diseases such as corn stunt disease (Zhao et al. 2020). These
pathogens can negatively impact maize production and result in yield losses.
Broadly maize diseases are classified into three categories as shown in Fig. 19.8.
310 K. Ahmed et al.

19.2.1 Fungi Diseases

Fungal diseases pose a significant threat to maize plants and can have a considerable
impact on their growth, yield, and overall health. Maize, also known as corn, is a
staple crop that is widely cultivated for various purposes such as food, animal feed,
and industrial products. However, its susceptibility to fungal infections makes it
vulnerable to a range of diseases that can lead to reduced crop productivity and
economic losses for farmers. Fungus disease infects the plant from root to tassel
(FAOSTAT 2023). These fungi often thrive in humid and moist conditions, which
create an ideal environment for their growth and spread. Some common fungal
diseases that affect maize plants include the following.

19.2.1.1 Brown Spot

Brown spot disease, also known as brown spot leaf blight. Caused by the fungus
Bipolaris maydis, known as Helminthosporium maydis (Subedi et al. 2023). The
most distinctive symptom of this disease is the appearance of brown, oval-to-
irregularly shaped spots on the leaves of maize plants. A golden halo frequently
surrounds these dots. Over time, the brown spots can coalesce, leading to the
formation of larger lesions. The fungus thrives in high-humidity environments,
most active at 20–30 °C (68–86 °F) (Afzaal et al. 2022). Frequent rainfall or
irrigation leads to moisture on the leaves providing a suitable environment for the
fungus to grow and reproduce. The fungus can survive on crop residues and in the
soil as a source of infection for new crops in subsequent growing seasons. It can also
be transmitted through contaminated seeds; severe infections can cause the leaves to
wither and die, resulting in blighting. To manage brown spot disease in maize,
several strategies can be employed: resistant varieties, planting maize varieties that
are resistant to the disease, can be an effective way to prevent infection. Crop
rotation is another method to avoid planting maize in the same field repetitively to
reduce the buildup of fungal spores in the soil. In severe cases, fungicides can be
used to control the disease. However, this should be done under the guidance of
agricultural experts. Proper irrigation—avoid overhead irrigation—is another strat-
egy as it can create a humid environment that favors the fungus. Drip irrigation or
furrow irrigation is preferable. Crop residue management is to remove and destroy
crop residues after harvest to reduce the source of fungal spores for the next season
as shown in Fig. 19.9.

19.2.1.2 Downy Mildew

Downy mildew is a fungal disease that affects various plants, including maize (corn).
It is a common fungal disease caused by various species of the oomycete pathogen,
typically caused by the pathogen Peronosclerospora spp. (Kominko et al. 2021). A
19 Guarding Maize: Vigilance Against Pathogens Early Identification,. . . 311

Fig. 19.9 Helminthosporium maydis infect corn leaf

general overview of the various types of downy mildews that can affect corn plants.
This group of fungi causes downy mildew in corn. Among them, Peronosclerospora
maydis is the most notable species. It primarily affects maize and can cause
yellowing and wilting of leaves, along with reduced yield. Species within this
group, such as Sclerophthora macrospora, can cause downy mildew in corn.
Symptoms include stunting, yellowing of leaves, and the formation of a white to
grayish fungal growth on the undersides of leaves. Some members of this genus, like
Pseudoperonospora cubensis, can infect corn plants. Symptoms include yellowing
of leaves, white or grayish fungal growth on leaf undersides, and reduced plant
vigor. Basiomycetes spp. are fungi which cause downy mildew in corn. They can
lead to leaf discoloration, stunting, and poor crop development (Maitra et al. 2020).
These various types of downy mildew in corn are typically managed through cultural
and crop rotation practices, the use of resistant corn varieties, and the application of
fungicides when necessary. These pathogens thrive in humid and moist conditions
and can severely impact maize crops. It affects a wide range of plants, including
maize. The symptoms of downy mildew in maize plants can vary, but they often
include yellow to pale green lesions on the upper side of leaves, the characteristic
downy growth on the undersides of leaves, stunted growth and reduced yield, and
premature death of infected plants (Li et al. 2020). Downy mildew pathogens
overwinter as resting structures in infected plant debris or soil. The pathogen then
grows inside the plant, causing the symptoms mentioned above.
It’s essential for farmers and growers to monitor their corn crops regularly to
detect early signs of downy mildew and take appropriate measures to prevent its
spread. Downy mildews can be a significant concern for maize growers, leading to
reduced crop yields and economic losses as shown in Fig. 19.10. Understanding the
causes, symptoms, and management strategies for this disease is crucial for effective
disease control in maize cultivation.

Tar Spot

Tar spot is a fungal disease that affects maize (corn) plants (Xu et al. 2023). It can be
determined by a formation of tiny, brought-up, black specks on maize plant foliage,
312 K. Ahmed et al.

Fig. 19.10 Downy mildew-infected corn leaf

Fig. 19.11 Tar spot-infected corn leaf

husks, and, in rare cases, stems. These black spots often cluster together and can vary
in size. The name “tar spot” comes from the resemblance of these black lesions to
spots of tar (Santos et al. 2022). The common symptoms of tar spot disease in maize
plants are tar-like spots, clustered spots, and yellow halos. The disease can affect the
photosynthetic capacity of the plant by reducing the effective leaf area, which can
ultimately impact crop yield. While tar spot disease can vary in severity depending
on environmental conditions, it can be managed through cultural practices like crop
rotation, tillage, and the removal of infected plant debris. In severe cases, fungicides
may also be used to control the disease (Zhao et al. 2020). The disease has a life
cycle that involves overwintering infected maize debris from the previous growing
season, spore production, infection of healthy plants, and the continued production
and dispersal of spores during the growing season. Proper management practices are
crucial to minimize its impact on maize crop production.

Maize Rusts

Maize rust disease, also known as corn rust, is a plant disease that affects corn (Zea
mays) plants as shown in Fig. 19.11. Various kinds of fungus from the Puccinia
genus cause this illness, with Puccinia sorghi being one of the most prevalent
offenders (Schunck et al. 2021). The hallmark symptom of maize rust is the
appearance of small, circular elongated pustules on the leaves, stems, and sometimes
even the ears of corn plants. These pustules range in color from orange to reddish-
brown, resembling rust, and hence the name “corn rust.” The major causes of maize
rust disease in corn plants are fungal pathogens, overwintering (alternate hosts,
19 Guarding Maize: Vigilance Against Pathogens Early Identification,. . . 313

infected crop debris), weather conditions (moisture, temperature, and wind), and
susceptible hosts. Maize rust disease is a significant threat to corn crops and can lead
to yield, quality reductions, and economic losses for farmers.

Common Rust

Common rust disease, caused by the fungus Puccinia sorghi (Miao et al. 2021), is a
widespread and economically significant fungal infection that affects various plants,
especially those in the Poaceae family, such as corn (maize) (Zea mays), and does
not infect other plant species. The most noticeable symptom of common rust is the
appearance of raised, reddish-brown to orange pustules on the leaves, stems, and
sometimes even the husks and ears of infected plants (Qiu et al. 2022). These
pustules give the plant a rusty or blistered appearance, hence the name “common
rust.” The life cycle of the common rust fungus involves both sexual and asexual
reproduction. Urediniospores: these are the asexual spores produced in the rust
pustules on infected plants. They are transported due to wind and rain.
Urediniospores land on healthy corn plants and germinate to form specialized
structures called appressoria. These structures penetrate the plant’s epidermal cells,
initiating infection. The fungus grows within the plant, forming pustules that even-
tually rupture, releasing more urediniospores. As the season progresses, the fungus
undergoes sexual reproduction, producing teliospores within the pustules. Telio-
spores serve as the overwintering stage, allowing the fungus to survive the winter on
infected crop residues or alternate hosts. Common rust can lead to reduced photo-
synthetic capacity in infected plants, resulting in decreased yields. Severe infections
can cause premature leaf senescence, further reducing yield and crop quality. Early
detection and management are key to preventing the spread and impact of common
rust. Crop rotation, planting resistant varieties, and practicing good agricultural
hygiene are important preventative measures as shown in Fig. 19.12.

Polysora Rust

This disease is also known as oxidation in the southern hemisphere. Fungal infection
caused by Polysora spp. (Qiu et al. 2022) that affects maize (corn) primarily occurs
in tropical and subtropical regions. The symptoms include rust-colored pustules.

Fig. 19.12 Common rust-

infected corn leaf
314 K. Ahmed et al.

Fig. 19.13 Polysora rust (southern rust)-infected corn leaf

Similar to common rust, polysora rust also produces rust-colored pustules on the
leaves, stems, and sometimes husks and ears of infected maize plants as shown in
Fig. 19.13. Small yellow spots: early symptoms may include small, yellowish spots.
To manage fungal diseases in maize, farmers often use a combination of cultural
practices, such as crop rotation, choosing resistant maize varieties, and applying
fungicides when necessary. Proper irrigation and drainage management can also
help reduce the risk of fungal infections by minimizing moisture stress.

Bacterial Disease in Maize

Bacterial diseases in maize can significantly impact crop yield and quality. One of
the most common bacterial diseases affecting maize is bacterial leaf streak (BLS).
BLS in Maize Infectious leaf scar in maize occurs predominantly by the bacteria
Xanthomonas vasicola pv. Zeae. This bacterium is a pathogenic microorganism
responsible for the development of BLS (Oehme et al. 2022).
Symptoms BLS symptoms typically appear on the leaves of maize plants. They may
include long, narrow, yellow to tan streaks on the leaves, which often parallel the leaf
veins. These streaks may expand and turn brown or necrotic, leading to a “burned”
appearance. Lesions may coalesce, covering large areas of the leaf surface. In severe
cases, the leaves may become tattered and dry.
Disease Progression Bacterial leaf streak can spread rapidly, especially under
favorable conditions like high humidity and rainfall. The bacteria enter through
natural openings or wounds on the leaves. As the disease progresses, it can lead to
defoliation, reducing the plant’s photosynthetic capacity and ultimately affecting
crop yield.
Favorable Conditions BLS is more likely to occur when environmental conditions
favor bacterial growth and infection. These conditions include high humidity and
19 Guarding Maize: Vigilance Against Pathogens Early Identification,. . . 315

moisture, which promote bacterial growth and the spread of the disease. Warm
temperatures, typically between 68 and 86 °F (20 and 30 °C), are conducive to
bacterial multiplication. Rainfall or irrigation can provide the moisture necessary for
infection.
Management Strategies for BLS in Maize Include:
Crop Rotation: Avoid planting maize in the same field year after year to reduce
the buildup of the pathogen.
Resistant Varieties: Plant maize varieties that are resistant or tolerant to BLS.
Fungicide Application: In severe cases, fungicides may be used to control the
disease, although this is not always economically viable.
Sanitation: Remove and destroy infected crop residues to reduce the inoculum
source.
Proper Water Management: Avoid excessive irrigation, as it can create condi-
tions favorable for the disease.
Bacterial leaf streak is a significant concern for maize growers as it can lead to
yield losses and reduce the quality of harvested maize. Understanding the causal
agent, symptoms, disease progression, and management strategies for bacterial leaf
streak in maize is crucial for effective disease control and maintaining healthy maize
crops.

19.2.2 Maize Virus and Mollicute Diseases

Maize lethal necrosis disease (MLND) is a nasty disease that affects maize (corn)
plants. It is a relatively recent and destructive disease that has caused significant
economic losses in maize production, primarily in regions of East Africa, although
cases have been reported in other parts of the world as well (Klupczyńska and
Pawłowski 2021). MCMV and SCMV can cause this disease. These two viruses
work together synergistically to produce the devastating effects of MLND. Joint
infection of twins is caused by oxidation in the southern hemisphere MCMV and
SCMV. These viruses belong to different virus families. MLND is primarily trans-
mitted through vectors, particularly insect vectors. These vectors acquire viruses
when feeding on infected plants and then transmit them to healthy maize plants while
feeding. MLND-infected maize plants display a range of characteristic symptoms,
which typically become apparent during the growing season. The leaves of infected
plants show chlorosis or yellowing, particularly along the leaf margins. Infected
plants often exhibit stunted growth and reduced overall vigor.
Mollicute diseases in maize Mollicutes are a group of bacteria that lack a cell wall
and are responsible for causing various plant diseases (Farmers Weekly 2017),
including those in maize. One notable mollicute disease in maize is “corn stunt,”
caused by a bacterium known as Spiroplasma kunkelii.
316 K. Ahmed et al.

19.2.2.1 Corn Stunt

Corn stunt is caused by Spiroplasma kunkelii, a mollicute bacterium. Infected maize

plants exhibit stunting, reduced growth, and a characteristic “broom-like” appear-
ance of the leaves. The disease can also lead to reduced yields. Leafhoppers, which
act as vectors for the bacteria, are the main source of corn stunt transmission. When
they feed on infected plants, they acquire the pathogen and transmit it to healthy
plants. In both cases, prevention and control measures often involve planting
disease-resistant maize varieties, doing good weed control to reduce potential
hosts for vectors, and implementing integrated pest management strategies. It’s
important to note that particular symptoms and management practices could differ
contingent on the maize virus disease and the geographical region in which they
occur. Regular monitoring and early detection are key to managing these diseases
effectively.

19.3 Conclusion

Maize crop diseases pose significant challenges to farmers worldwide, impacting

crop yield, quality, and overall agricultural productivity. Effective identification and
management of these diseases are crucial for sustainable maize production. Early
detection is essential for effective disease management. Farmers should regularly
scout their maize fields to identify symptoms of diseases, including leaf discolor-
ation, wilting, necrosis, and unusual growth patterns. Integrated pest management
(IPM) approaches are recommended, combining various strategies to minimize
disease impact. Artistic practice, crop rotation, proper spacing, and sanitation help
reduce disease pressure. Planting disease-resistant maize varieties is a proactive
measure to prevent infections. Fungicides and pesticides should be used judiciously
and in accordance with recommended guidelines to minimize environmental impact.
Biological control methods, like beneficial microorganisms, can be employed to
suppress disease-causing pathogens. Farmers should be educated about common
maize diseases, their symptoms, and management options. Extension services and
agricultural institutions play a crucial role in disseminating knowledge and best
practices to farmers. By combining early detection, integrated management strate-
gies, education, and ongoing research, farmers can minimize the impact of diseases
and ensure sustainable maize production.

References

Afzaal M, Hameed S, Rasheed R, Din Khan WU (2022) Microalgal biofuels: a sustainable pathway
for renewable energy. In: Algal biotechnology. Elsevier, pp 187–222
19 Guarding Maize: Vigilance Against Pathogens Early Identification,. . . 317

Doebley JF, Gaut BS, Smith BD (2006) The molecular genetics of crop domestication. Cell 127(7):
1309–1321
Dotasara SK, Choudhary CS (2023) Fall armyworm in maize crop: challenges and management in
India. International Year of Millets
FAOSTAT (2023) https://www.fao.org/faostat/en/#data/QCL
Farmers Weekly (2017) https://www.farmersweekly.co.za/crops/field-crops/maize-production-
managing-critical-plant-growth-stages/
Forde BG, Clarkson DT (1999) Nitrate and ammonium nutrition of plants: physiological and
molecular perspectives. In: Advances in botanical research, vol 30. Academic Press, pp 1–90
Gao Z-F, Yang X, Mei Y, Zhang J, Chao Q, Wang B-C (2023) A dynamic phosphoproteomic
analysis provides insight into the C4 plant maize (Zea mays L.) response to natural diurnal
changes. Plant J 113(2):291–307
Hallauer AR, Carena MJ (2009) Maize. Springer
Klupczyńska EA, Pawłowski TA (2021) Regulation of seed dormancy and germination mecha-
nisms in a changing environment. Int J Mol Sci 22:1–18
Kominko H, Gorazda K, Wzorek Z (2021) Formulation and evaluation of organo-mineral fertilizers
based on sewage sludge optimized for maize and sunflower crops. Waste Manag 136:57–66
Kumar B, Rakshit S, Kumar S, Singh BK, Lahkar C, Jha AK, Kumar K et al (2022) Genetic
diversity, population structure and linkage disequilibrium analyses in tropical maize using
genotyping by sequencing. Plants Theory 11(6):799
Li Z, Guo R, Li M, Chen Y, Li G (2020) A review of computer vision technologies for plant
phenotyping. Comput Electron Agric 176:105672
Maitra S, Shankar T, Banerjee P (2020) Potential and advantages of maize-legume intercropping
system. In: Maize-production and use, pp 1–14
McHale LK, Haun WJ, Xu WW, Bhaskar PB, Anderson JE, Hyten DL, Gerhardt DJ, Jeddeloh JA,
Stupar RM (2012a) Structural variants in the soybean genome localize to clusters of biotic
stress-response genes. Plant Physiol 159(4):1295–1308
McHale LK, Haun WJ, Xu WW, Stinchcombe JR (2012b) Genomic features and evolution of corn
plants. Plant Cell 24(11):4815–4832
Miao T, Zhu C, Tongyu X, Yang T, Li N, Zhou Y, Deng H (2021) Automatic stem-leaf
segmentation of maize shoots using three-dimensional point cloud. Comput Electron Agric
187:106310
Monterrubio-Solís C, Barreau A, Ibarra JT (2023) Narrating changes, recalling memory: accumu-
lation by dispossession in food systems of indigenous communities at the extremes of Latin
America. Ecol Soc 28(1):3
Naseem M, Singh V, Ahmed K, Mahroof M, Ahamad G, Abbasi E (2022a) Architecture of
automatic irrigation system in hilly area using wireless sensor network: a review. In: In 2022
2nd International conference on emerging frontiers in electrical and electronic technologies
(ICEFEET). IEEE, pp 1–6
Naseem M, Alam M, Ahmad K, Singh V, Mahroof M, Ahamad G (2022b) Machine learning
approaches for automatic irrigation system in hilly areas using wireless sensor networks
Oehme LH, Reineke A-J, Weiß TM, Würschum T, He X, Müller J (2022) Remote sensing of maize
plant height at different growth stages using UAV-based digital surface models (DSM).
Agronomy 12(4):958
Qiu R, Zhang M, He Y (2022) Field estimation of maize plant height at jointing stage using an
RGB-D camera. Crop J 10(5):1274–1283
Ranum P, Peña-Rosas JP, Garcia-Casal MN (2014) Global maize production, utilization, and
consumption. Ann N Y Acad Sci 1312(1):105–112
Sadras V, Calderini D (2009) Crop physiology: applications for genetic improvement and agron-
omy. Academic Press
Santos D, Caio L, Abendroth LJ, Coulter JA, Nafziger ED, Suyker A, Jianming Y, Schnable PS,
Archontoulis SV (2022) Maize leaf appearance rates: a synthesis from the United States corn
belt. Front Plant Sci 13:872738
318 K. Ahmed et al.

Schunck D, Magistri F, Rosu RA, Cornelißen A, Chebrolu N, Paulus S, Léon J et al (2021)

Pheno4D: a spatio-temporal dataset of maize and tomato plant point clouds for phenotyping
and advanced plant analysis. PLoS One 16(8):e0256340
Smith AM, Zeeman SC (2006) Quantification of starch in plant tissues. Nat Protoc 1(3):1342–1345
Subedi B, Poudel A, Aryal S (2023) The impact of climate change on insect pest biology and
ecology: implications for pest management strategies, crop production, and food security. J
Agric Food Res 14:100733
ur Rehman F, Adnan M, Kalsoom M, Naz N, Husnain MG, Ilahi H, Ilyas MA, Yousaf G, Tahir R,
Ahmad U (2021) Seed-borne fungal diseases of maize (Zea mays L.): a review. Agrinula Jurnal
Agroteknologi dan Perkebunan 4(1):43–60
World Agricultural Production (2023) http://www.worldagriculturalproduction.com/crops/corn.
aspx
Xu H, Ming B, Wang K, Xue J, Hou P, Li S, Xie R (2023) Quantitative analysis of maize leaf collar
appearance rates. Plant Physiol Biochem 196:454–462
Zhang F, Cui Z, Chen X, Ju X, Shen J, Chen Q, Liu X et al (2012) Integrated nutrient management
for food security and environmental quality in China. Adv Agron 116:1–40
Zhao M, Tang S, Zhang H, He M, Liu J, Zhi H, Sui Y et al (2020) DROOPY LEAF1 controls leaf
architecture by orchestrating early brassinosteroid signaling. Proc Natl Acad Sci 117(35):
21766–21774
Chapter 20
Comprehensive Analysis of Deep Learning
Models for Plant Disease Prediction

Narendra Pal Singh Rathor, Praveen Kumar Bhanodia,

and Aditya Khamparia

Abstract The major crop across the world is wheat. The growth of the wheat crop is
significantly affected by various types of plant diseases. Technology advancements
are instrumental in the recognition and prediction of plant diseases by diagnosing the
health of plant leaves. To cut losses and achieve intelligent, healthy farming, the use
of computer vision and pattern recognition to identify disease has been researched.
The rapid and precise automatic detection of diseases is now possible with image
recognition techniques. This work focuses on developing methods for wheat plant
disease identification using deep learning models. There are many deep learning
models proposed by researchers, but the majority provide poor testing results if some
variation (rotation, tiling, and other abnormal image orientations) is there in the
images; moreover the models do not store relative spatial relationships among the
features captured. Thus the intent of this work is to implement the Whe-C-Net hybrid
model, which combines features of VGG16 and CapsNet. The VGG16 model is
initially employed for feature extraction. After that, misalignment issues with the
current deep learning models are dealt with using CapsNet layers. Later, dropouts,
sigmoid activation functions, and fully connected layers are employed. To avoid
overfitting and create a Whe-C-Net model that is more broadly applicable, dropouts
are used. On the dataset of wheat plant photos, the effectiveness of Whe-C-Net is
confirmed. Compared to competing models like pre-trained MobileNet, it obtains a
better validation accuracy of 98%, which is noteworthy. The accuracy rates for
Xception, ResNetAQ1, MobileNet and VGG16 were 96%, 96%, 65%, and 93%,
respectively.

Keywords Deep learning · Wheat · Plant disease · CNN · Whe-C-Net · Capsule net

N. P. S. Rathor · P. K. Bhanodia (✉)

Computer Science Engineering, Acropolis Institute of Technology and Research, Bhopal,
Madhya Pradesh, India
e-mail: [email protected]
A. Khamparia
Department of Computer Science, Babasaheb Bhimrao Ambedkar University, Satellite Centre,
Amethi, Uttar Pradesh, India

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 319
A. Khamparia et al. (eds.), Microbial Data Intelligence and Computational
Techniques for Sustainable Computing, Microorganisms for Sustainability 47,
https://doi.org/10.1007/978-981-99-9621-6_20
320 N. P. S. Rathor et al.

20.1 Introduction

India is primarily an agriculturally oriented nation, with agriculture employing 70%

of the people (http://www.fao.org/india/fao-in-india/india-at-a-glance/en/ n.d.).
Agriculture has grown far more significant in recent years than it was a few years
ago, when plants were just used to feed people and animals. This is because plants
are being employed to generate electricity and other forms of energy in order to
improve mankind’s living conditions. However, there are other plant diseases that
can cause significant economic and social impacts. It may possibly result in signif-
icant ecological damage. To avoid such losses, it is therefore preferable to detect
diseases precisely and promptly. Plant diseases can be detected using a variety of
methods, including both manual and computerized technologies. The majority of
plant illnesses manifest themselves as spots on the leaves that are more obvious to
the naked eye. On the other hand, some diseases do not appear on the leaves at all,
while others appear later after they have already caused significant damage to the
plants. In such cases, it is suggested that computerized systems would be the only
option for obtaining actual, timely exploitation of some pretty advanced algorithms
and analytical tools, ideally through the use of powerful microscopes and other
instruments (Mohanty et al. 2016). If the system was easy to use and accessible, it
may be a valuable tool for farmers in areas of the world where there is a lack of
infrastructure for providing agronomic and phytopathological guidance. Further-
more, on a large scale, the technology might be paired with autonomous agricultural
tractors to correctly and quickly find phytopathological concerns throughout the
cultivation field, utilizing continuous image capture. All of this is true, of course, if
the system can attain high levels of performance in detecting and diagnosing specific
diseases in real-world situations (Johannes et al. 2017). Machine learning-related
artificial intelligence applications have grown exponentially in recent years, thanks
to the development of computational systems, particularly Graphical Processing
Unit (GPU) embedded processors, resulting in the development of novel methodol-
ogies and models, which have now formed a new category, deep learning
(DL) (LeCun et al. 2015). A plant pathologist must have good observation skills
in recognizing distinctive symptoms in order to diagnose plant diseases accurately
(Riley et al. 2002). Because amateur gardeners and hobbyists may have more
difficulty diagnosing defected plants than trained plant pathologists, variations in
symptoms suggested by diseased plants may lead to an incorrect diagnosis. An
automated system that can identify plant illnesses based on the look and visual
symptoms of the plant might be extremely useful to both amateur gardeners and
skilled professionals as a disease diagnosis verification process (Sladojevic et al.
2016). Nowadays deep learning provides a very efficient way to extract features
from images, through which we can easily identify diseases in plant leaves. Here, the
work demonstrated various deep learning models to identify the plant leaf identifi-
cation and detection. In this work four different deep learning models are applied,
namely, VGG16, XceptionNet, MobileNet, and ResNet, using transfer learning for
the detection of plant leaves. All four model performances on the wheat leaf dataset
20 Comprehensive Analysis of Deep Learning Models for Plant Disease Prediction 321

in terms of accuracy and loss are compared. After that a hybrid model is used to
improve the performance to detect the plant leaf disease. In this work a dataset of
wheat leaves has been used, and all deep learning models are compared on the same
dataset. Our work mainly concentrated on a wheat leaf disease called leaf rust. After
applying all models, we have achieved the highest accuracy, i.e., 98.93%, through
our hybrid model. The details of the work carried out are mentioned in the coming
chapters.

20.2 History

Several researchers have tried on agricultural applications that use image processing,
pattern recognition techniques, machine learning, and deep learning throughout the
last decade. The disease can affect any part of the plant, including the roots, stem,
fruit, or leaves. When technology has updated in the last 5–10 years, everybody is
moving to small and handy devices. Everybody wants to process their results quickly
and efficiently. Earlier for disease detection in plant leaves we were dependent only
on physical scanning using our naked eyes, but now so many technologies are
available, using which we can easily and quickly detect diseases and provide
information about the diseases. Neural network training proves significant changes
in this area. Table 20.1 shows the summary of the work proposed in this domain.
We have used deep learning to recognize wheat plant diseases in our research,
which is motivated by the advancement of deep learning techniques and their
practical use. Researchers studied deep learning approaches for plant disease recog-
nition using leaf images, but no evidence of this was found in a search of the state-of-
the-art literature.

20.3 Research Methodology

DL enhances classical ML by having “depth” to the model and updating the data
with different functions that permit the representation of data hierarchically, across a
number of levels of abstraction (Patrick et al. 2019). Feature learning, or the
automated extraction of features from raw data, is a significant benefit of DL, with
higher-level features being generated by the composition of lower-level features
(Sarayloo and Asemani 2015b). Because of the more sophisticated models utilized,
which allow huge parallelization, DL can handle more complex problems especially
well and quickly (Lee et al. 2015). Models used in deep learning (DL) can improve
classification accuracy or minimize regression error when substantial datasets defin-
ing the problem are readily available. While there are many distinct types of deep
learning (e.g., unsupervised pre-trained networks, convolutional networks, recursive
networks, and recurrent networks), every DL uses some combination of convolu-
tions and fully connected layers as well as gates and memory cells. Classification and
Table 20.1 A comprehensive summary of noteworthy contributions in the plant disease analysis domain
322

Author Steps Attribute Plants Diseases Analysis

Zahra Sarayloo and 1: Segmentation Color, shape, Wheat Stem rust, leaf rust, tan No of images not
Davud Asemani (2015a) 2: Extract feature and texture spot, stripe rust, pink specified
3: Select features snow mold, septoria, and
4: Disease classification powdery mildew
Sue Han Lee, Chee Seng Convolutional neural – Forty-four plants but – Plant names missing and
Chan, Paul Wilkin, Paolo network specification not men- on which attributed
Remagnino (2015) 1: Layer/multilayer tioned by the author work has performed also
perceptron missing
2: Support vector
machine, radial basis
function
Yogesh Dandawate, Image acquisition of Color transfor- Soybean – On which soybean dis-
Radha Kokare (2015) soybean leaves. mation, back- ease has been carried out
Extraction of soybean ground subtrac- not mentioned. It just
Leaves from complex tion, shape shows
background analysis healthy vs. infected
Statistical analysis and images. Also worked on
Disease classification only 120 images
Srdjan Sladojevic, Marko Image preprocessing and Apple, grapevine, pair, Powdery mildew, downy Presented their work on
Arsenovic, Andras labeling, 2. Augmenta- peach, pear mildew, apple rust, apple fruits, leaves, and dif-
Anderla, Dubravko tion process. Neural net- powdery mildew, apple, ferent diseases
Culibrk and Darko work training. Performed pear, Erwinia amylovora
Stefanovic (2016) tests. Fine-tuning.
Equipment.
Alvaro Fuentes, Dong Dataset, labeling, disease Tomato Leaf mold, gray mold, Presented their work on
Sun Park, Sook Yoon, correlation canker, plague tomato leaf disease
Hong Youngki, Yujeong detection but not clearly
Lee (Fuentes et al. 2016) mentioned how many
N. P. S. Rathor et al.
 20

images and dataset have

been used for the work
Kadir Sabanci, Mustafa Database (Waikato envi- Wheat – Not properly mentioned
Akkaya (2016) ronment for knowledge on which wheat disease
analysis), multilayer work has been carried
perceptron algorithm, out. Work actually car-
K-nearest neighbors ried on wheat seed
algorithm, J48 decision
tree algorithm, naive
Bayes algorithm
Alvaro Fuentes, Sook Data collection, data Tomato Leaf mold, gray mold, Work carried out on
Yoon, Sang Cheol Kim annotation, data aug- canker, plague, miner, 5000 tomato images and
and Dong Sun Park mentation, disease and powdery mildew, nutri- used faster R-CNN
(2017) pest detection tional excess
Yang Lu, Shujuan Yi, Convolutional layer, sto- Rice Rice false smut (RFS), Work carried out on
Nianyin Zeng, Yurong chastic pooling layer, rice blast (RB), rice 500 images of rice and
Liud, Yong Zhang Softmax regression, brown spot (RBS), rice presented a simple CNN
(2017b) training algorithm seedling blight (RSEB), to detect the disease in
Rice rice leaf
Sheath blight (RSHB),
rice sheath rot (RSR),
rice bacterial leaf blight
(RBLB), rice bacterial
sheath rot (RBSR), and
rice bacterial wilt (RBW)
Bin Liu, Yun Zhang, Apple leaf pathological Apple leaf Mosaic, rust, brown spot, Work carried out on
DongJian He, and image acquisition, image and Alternaria leaf spot 13,689 images of apple
Comprehensive Analysis of Deep Learning Models for Plant Disease Prediction

Yuxiang Li (2018) processing and generat- leaf using AlexNet,

ing pathological images GoogLeNet, and
Softmax regression has
been used
323

(continued)
Table 20.1 (continued)
324

Author Steps Attribute Plants Diseases Analysis

Mohammed Brahimi, Preprocessing, feature Tomato Tomato yellow leaf curl Work carried out on
Kamel Boukhalfa & extraction, classification, virus, tomato mosaic 14,828 images, and
Abdelouahab Moussaoui virus, spider mites, machine learning algo-
(2017) Septoria spot, leaf mold, rithms K-nearest neigh-
late blight, bacterial spot bors algorithm (KNN)
and support vector
machine (SVM) have
been used
Jayme G.A. Barbedo – Olive tree, corn, apple, Anthracnose, tropical Simple CNN has been
(2018) tomato, potato, rice rust, southern corn rust, applied for the work.
scab, southern corn leaf Details of the steps
blight, Phaeosphaeria performed are also
leaf spot, Diplodia leaf missing
streak, Physoderma
brown spot, northern leaf
blight
Juncheng Maa, Keming Image datasets, CNN, Cucumber Anthracnose, downy Work carried out on
Dua,Feixiang Zhenga, evaluation of the DCNN mildew, powdery mil- cucumber leaves and
Lingxian Zhangb, dew, target leaf spots also used KNN and
Zhihong Gongc, Zhongfu SVM as classifiers
Suna (2018)
Artzai Picon, Aitor Deep convolutional neu- Wheat Tan spot, rust, Septoria ResNet-50 has been
Alvarez-Gila, ral networks (CNNs) used on wheat leaves for
Maximiliam Seitz, Amaia disease detection
Ortiz-Barredo, Jone
Echazarra,Alexander
Johannes (2019)
G. Prem Rishi Kranth, Decision tree, K-means Apple, strawberry Wilting, spot, powdery Machine learning algo-
M. Hema Lalitha, clustering, naive Bayes, mildew, galls, dryness rithm used for the
Laharika Basava, Anjali random forest, artificial detection of apple and
N. P. S. Rathor et al.

Mathur (2018) neural network strawberry diseases

 20

Md Mehedi Hasan, CNN Wheat Worked on wheat

Joshua P. Chopin, Hamid images and presented
Laga and Stanley simple CNN and
J. Miklavcic (2018a) performed work to
identify SPIKE in wheat
Konstantinos VGG, AlexNet Apple, banana, blue- Apple scab, black rot, Pre-trained model archi-
P. Ferentinos (2018) berry, cabbage, cassava, black Sigatoka, banana tecture have been used
celery, cherry, corn, speckle, powdery mil- to identify the plant
cucumber, eggplant, dew, downy mildew, leaves disease detection
gourd, grape, onion, esca, leaf blight,
orange, peach, soybean Huanglongbing, early
blight, cercospora
M. Akila, P. Deepan Data collection, image Sugarcane, cotton, – Faster region-based
(2018) annotation and augmen- potato, carrot, chilly, convolutional neural
tation, image analysis, brinjal, rice, wheat, network (faster R-CNN)
feature extraction banana and guava and region-based fully
convolutional network
(R-FCN) are mainly
used, but the disease
name on which work has
been carried out is not
mentioned specifically
Satwinder Kaur, Garima – – Apple, blueberry, corn, Apple scab, apple cedar GoogLeNet architecture
Joshi, Renu Vig (2019) cherry, grapes, orange, rust, cherry powdery has been used to identify
peach, pepper, potato, mildew, corn leaf spot, disease
soybean, squash, tomato, corn common rust, grape
strawberry black
Measles, peach bacterial
Comprehensive Analysis of Deep Learning Models for Plant Disease Prediction

spot, potato early blight,

squash powdery mildew,
tomato leaf mold, tomato
mosaic virus, tomato leaf
325

(continued)
Table 20.1 (continued)
326

Author Steps Attribute Plants Diseases Analysis

curl
Virus
Abirami Devaraj, Image acquisition, image Rice Alternaria alternata, bac- K-means and random
Karunya Rathan, preprocessing, image terial blight forest machine learning
Sarvepalli Jaahnavi and K segmentation, algorithm have been
Indira (2019) classification used. Not clearly men-
tioned in the dataset
details
Chowdhury Rafeed et al. Tea, apple, tomato, False smut, sheath blight, VGG16, ResNet-50,
(2021) grapevine, peach, and sheath rot, bacterial leaf inception-v3, Xception,
pear blight (BLB), neck blast inception-ResNet-v2
and brown spot pre-trained model archi-
tecture have been used
and mainly focused on
rice leaves
Justine Boulent, Samuel – – – – AlexNet, DenseNet,
Foucher, Jerome Theau1, inception, VGG, ResNet
and Pierre-Luc St-Charles model details have been
(2019) mentioned but not
clearly mentioned on
which plant and diseases
work has been carried
out
X.E. Pantazi, D. Moshou, Local binary patterns Vitis, Cucurbita pepo, Powdery mildew, black Support vector machine
A.A. Tamouridou (2019) (LBPs) for feature Cucumis sativus rot, downy learning algorithms have
extraction and one-class Mildew been used
classification for
classification
N. P. S. Rathor et al.
20 Comprehensive Analysis of Deep Learning Models for Plant Disease Prediction 327

prediction are especially effectively performed by DL models due to their highly

hierarchical structure, enormous learning capacity, and adaptability to many
extremely difficult tasks (from a data analysis standpoint) (Lee et al. 2015). Deep
learning is very useful nowadays in the agriculture domain. In deep learning various
models are proposed by authors for wheat leaf detection. This research covered
simple convolution neural networks, XceptionNet, MobileNet, ResNet, and VGG16.
All these models are available and trained at Keras. After deep learning models are
studied, move on to the wheat leaf diseases dataset. Dataset plays a key role in deep
learning to train the model. The dataset collection visited a number of recognized
and reputed dataset repositories like PlantVillage, Mendeley dataset, and Kaggle.
Mohammed, Assem (Mohammed 2020) presented a Wheat dataset which is
included in our research. The detailed description of the dataset is covered in the
next chapter. After that research moved on to transfer learning studies and imple-
mentation of deep learning models. Transfer learning (Lumini and Nanni 2019) is a
technique which provides a way through which we can reuse an already developed
model with our weights and trained model according to our problem. Covered
ResNet, VGG, MobileNet, and XceptionNet in this research use transfer learning.
Implemented a model which was trained with cutomized weights on Wheat dataset
and listed out accuracy, loss, Precision, Recall, and F-Score, to provide a complete
view of system’s performance (Goutte and Gaussier 2005). Then our research
moved to developing a new hybrid model which is better in terms of the performance
of existing implemented transfer learning models as shown in Fig. 20.1. Sabour et al.
(2017) presented a new architecture, i.e., Capsule Networks. Deep learning models
that use Capsule Networks as their fundamental building blocks have grown in
popularity since they were first introduced in 2017. A routing technique known as
“routing by agreement” is used in the most common version of CapsNets. The vector
output of this method substitutes the scalar output of CNNs and the pooling
technique of CNNs. Due to the fact that CNNs require large amounts of training
data and are unable to identify object position and deformation, the development of
Capsule Networks was made possible. Known as Capsule Networks, they are the
latest craze in deep learning. They have lived up to this expectation since their
performance in issues such as image recognition, natural language processing, object
identification, object segmentation, and language translation has consistently
outperformed that of convolutional neural networks (Patrick et al. 2019).

20.4 Deep Learning for Wheat Leaf Disease Detection

Deep learning methods are increasingly being used to solve machine vision chal-
lenges. Various researchers have previously investigated plant and leaf identification
using various methodologies. Initially, color information was employed to identify
the plant from the soil in these difficulties (Woebbecke et al. 1995; Gerhards and
Christensen 2003). The morphology of veins was used in some investigations (Sack
et al. 2008; Scoffoni et al. 2011). The leaf veins include a variety of texture and form
features that can help with plant identification using eyesight. The shape information
was also utilized in certain leaf-based research (Agarwal et al. 2006; Neto et al.
328 N. P. S. Rathor et al.

Fig. 20.1 Systematic steps

involved in plant disease
analysis and prediction

2006). Several researchers (Husin et al. 2012) integrated the shape and texture
information retrieved from the leaf, while others used color and texture information
(Pydipati et al. 2006). Larese et al. (2014a) offer a method for extracting the
morphological leaf vein information using computer vision techniques. Machine
learning methods are then utilized to predict the three different plant species using
the extracted data. Later, in a new study endeavor, these researchers enhanced this
strategy (Larese et al. 2014b). Plant genre classification uses spectroscopic
approaches as well. The feature extraction in these approaches is based on the
reflectance of the infrared, multispectral, and visible bands (Mattila et al. 2013;
Wang et al. 2007; Tyystjärvi et al. 2011). In their research, Muthevi and Uppu
20 Comprehensive Analysis of Deep Learning Models for Plant Disease Prediction 329

(2017) used Local Binary Patterns (LBP). For feature extraction, different types of
LBP extraction methods were used, including the signed component of CLBP
(SCLBP). Murat et al. (2017) used a combination of form descriptors to classify
the leaves of tropical shrub species. Flavia and Swedish leaf datasets were used to
test the suggested technique. View rotation invariant characteristics taken from the
fast Fourier transform and discrete wavelet transform were employed by Yousefi
et al. (2017). Yu et al. (2016) devised a strategy based on leaf venation and contour.
A method for invariant leaf detection was proposed by Horaisová and Kukal (2016).
Deep learning is a relatively recent approach to plant and leaf identification. To
work with leaf veins, Grinblat et al. (2016) presented a CNN model. Leaf veins are
retrieved as a binary mask and used to train an end-to-end CNN model. To create a
weed segmentation system, dos Santos Ferreira et al. (2017) integrated the
superpixel segmentation algorithm with CNN. On images segmented with the
superpixel method, CNNs are used in weed detection. For leaf identification, Barré
et al. (2017) developed a CNN architecture. For the classification, they used the
Foliage, LeafSnap, and Flavia datasets and developed a CNN model. In their work,
they developed an architecture that was similar to well-known CNN models like
AlexNet and CifarNet. Transfer learning from pre-trained CNN models was
employed by Jeon and Rhee (2017). For feature extraction, a pre-trained GoogLeNet
was employed. GoogLeNet was given leaf photos in various scales as input, and the
activation values of different layers were saved as features. Lu and Hu (Jiang et al.
2017a) proposed an automatic wheat leaf disease detection system and proposed a
CNN model to detect the disease. Hasan et al. (2018b) has also worked on wheat
spikes detection and analysis. They have worked on the ResNet-50 deep learning
model in their research.

20.5 Transfer Learning for Wheat Leaf Disease Detection

Transfer learning is typically a machine learning method, wherein the CNN models
trained on one task have been used as the base model for some other task (Lumini
and Nanni 2019). We can use a pre-trained network on big, labeled datasets, like
public picture datasets, to establish the weights instead of starting the training from
scratch by randomly initializing the weights. In the same fashion (Chen et al. 2020)
Junde Chen used pre-trained models on the ImageNet large dataset, which they
applied to the specific task established by the objective dataset. The main processes
of the transfer learning approach (Chen et al. 2020) are described as follows.

20.5.1 Identify Transfer Learning Base Model

Using the pre-trained CNN model, determine the transfer learning base networks and
give the network weights (W1, W2,. . . .W n) to the base networks. It is possible to
obtain the weights of bottom layers from a well-trained CNN (https://ker-as.io/
applications/) via the Ker-as API.
330 N. P. S. Rathor et al.

20.5.2 Create a New Neural Network

The network structure can be changed based on the bottom layers, which can include
altering layers, adding layers, and removing layers from networks, among other
things. It is possible to generate a new network structure in this way.

20.5.3 Perform Fine-Tuning on Created New Neural

Networks

In this step we are able to modify the layer parameters. Edna et al. (Too et al. 2019)
have performed fine-tuning on VGG16, done by truncating the original Softmax
layer and replacing it with their own.
Kathiresan et al. (2021) have also proposed the idea of transfer learning for the
detection of rice leaf diseases as shown in Fig. 20.2. Goyal et al. (2021) have worked
on wheat leaf and spike detection using transfer learning. They have included the
VGG16 and ResNet-50 base models for transfer learning.

20.6 Hybrid Approach

In the hybrid approach we apply pre-trained models using the transfer learning
method on the wheat leaf dataset.
The pre-trained models used for the classification and detection of plant diseases
are as follows.

Source ImageNet Dataset Target Dataset

VGGNet 16 Model Weight and

VGGNet 16 Model Parameters are
(Bottom layers)
freezed

Output of 1000 Perform the

Convolutional Finetuning of the
Classes Dense (Soft classes) model

Fig. 20.2 Process involved in transfer learning

20 Comprehensive Analysis of Deep Learning Models for Plant Disease Prediction 331

20.6.1 MobileNet

This is the first TensorFlow model designed specifically for mobile apps, and it uses
significantly separable convolutions. When compared to MobileNet’s use of con-
ventional convolutions on networks with the same depth, the number of parameters
is significantly reduced. It further leads to deep networks with low weight (Howard
et al. 2017).

20.6.2 XceptionNet

For the deep neural network it has 71 deep layers proposed by Francois Chollet
based on smart and thorough separation. In comparison to traditional convergence
depth this separable convolution model is more efficient and effective (Chollet
2017).

20.6.3 VGG16

The VGG16’s pre-training architecture consists of 13 convolutional layers, 5 addi-

tional layers, and 3 dense pool layers. A global average pooling layer and two dense
layers with the activation algorithms ReLu and Softmax are among the changes
made to the VGG16. The drop-out rate for both thick layers is 0.5 (Simonyan and
Zisserman 2014).

20.6.4 ResNet

Kaiming He et al. in 2015 introduced ResNet referred to as the Residual Network,

with this model the problem of vanishing gradient specifically in deep systems by
permitting the alternative for flow across gradient (He et al. 2015).
In this study the hybrid model is implemented using Google Colab, Keras, and
TensorFlow. Pre-trained VGG model is used for transfer learning (Hasan et al.
2018b), with dense and global average pooling layers; the activation function used
here is ReLu. The VGG model is followed by a capsule network with three added
routings. From the results obtained it is evident that Whe-C-Net superseded ResNet,
XceptionNet, MobileNet, and VGG model. Refer to Fig. 20.3 for the proposed
architecture of the Whe-C-Net model architecture.
The Whe-C-Net architecture has three distinct layers:
332 N. P. S. Rathor et al.

Output

VGG 16 Convolution Reshape Lambda

Capsule
Layer 2D

Fig. 20.3 The proposed Whe-C-Net model architecture

20.6.5 Extraction of Features Using VGG and Capsule

Network Layers

The VGG16 is a pre-trained CNN model consisting of 13 convolutional, 5 max

pooling, and 3 dense layers. In total out of 23, only 14 are weighted layers.

20.6.6 Classification Layers

In the output of capsule and VGG16 networks the first layer is known as the flattened
layer followed by the next layer, i.e., the dense layer, with ReLu activation function
and 0.5 dropout. The third and the last is also termed the dense layer, with the
Softmax activation function used for deduction in depth.

20.6.7 Spatial Transformer Layers

In the spatial transformer Lambda λ is used Default [-0.5: 0.5] to convert the wheat-
leaf image characteristics to a normal value of 0.0. For faster processing divide the
input into mini batches and apply batch normalization.

20.6.8 Implementation Steps Involved

Image dataset loaded in RAM for processing like the previous attempt when wheat
leaf image format pictures were saved in RGB. Defining the implementation archi-
tecture is shown in Fig. 20.4. The next step is to implement the metrics involved,
which includes threshold accuracy with β threshold and Fβ score. Thereafter the
model is employed for training and validation. The last step is validation of the
model. All steps are implemented in Figs. 20.5, 20.6.
20 Comprehensive Analysis of Deep Learning Models for Plant Disease Prediction 333

Fig. 20.4 Capsule model summary

Start
from keras import activations
from keras import utils
from keras.models import Model
Import Python from keras.layers import *
Libraries required from keras.optimizers import
for Capsule RMSprop, Adam, SGD, Nadam
Network import numpy as np

Capsule Function

def Capsule(num_capsule,dim_capsule,routings)

Layer 1: Conv2D(256, kernel_size=(9, 9), strides=(1, 1), activation=’relu’)(x)

Layer 2: PrimaryCapsule Layer (Layer1, dim_capsule=8, n_channels=32, kernel_size=9, strides=2, padding=’valid’)

Layer 3: CapsuleLayer (num_capsule=n_class, dim_capsule=16, routings=routings, name=’DiagnosisCaps’(Layer2)

Auxiliary layer to replace each capsule with its lenght out_caps = Length(name=’capsnet’)(Layer3)

Fig. 20.5 Capsule layer configuration

Table 20.2 shows that Whe-C-Net performed better for detecting plant disease by
diagnosing the wheat leaves with 98% accuracy. However, further training with
more epochs can enhance the proposed model performance.
334 N. P. S. Rathor et al.

Fig. 20.6 Hybrid model detailed architecture

20.7 Result and Discussion

The goal of the study is to use the deep learning models discussed to provide wheat
leaf disease detection. We used photos of normal leaves and leaves with rust to train
the models. To begin with, the ResNet-50 models have been used, which Artzai
Picon et al. (2019) also used to detect wheat leaf diseases. The validation accuracy
for the ResNet-50 model was 65% after training. Precision, recall, and f1-score are
20 Comprehensive Analysis of Deep Learning Models for Plant Disease Prediction 335

Table 20.2 Model performance analysis and parameters

Design f1- Validation No of
Dataset (model) Recall Precision score accuracy parameters
Mendeley dataset ResNet-50 0.58 0.57 0.57 65% Parameter:
(Kaur et al. 2019) 23,788,418
Trainable:
200,706
MobileNet 0.44 0.43 0.43 81% Parameter:
3,379,395
Trainable:
150,531
XceptionNet13 0.12 0.67 0.20 52% Parameter:
21,271,082
Trainable:
409,602
VGG 0.87 0.98 0.92 93% Parameter:
16,824,130
Trainable:
2,109,442
Whe-C-net 0.99 0.98 0.98 98% Parameter:
25,339,968
Trainable:
17,704,704

additional parameters that the author has not addressed but eventually are crucial to
analyzing the classifier. Thereafter, the MobileNet model was used to forecast the
diseases, with an accuracy rate of 81%. Despite the fact that we have achieved better
accuracy, it is not a reliable indicator of model performance.
Further in comparison to ResNet-50, the MobileNet model has higher accuracy
but worse precision, recall, and f1-score. Later we began training using a new model,
XceptionNet13, for the detection and prediction of possible disease across wheat
crops but, regrettably, did not attain the desired outcome. The results obtained were
lower in precision, recall, and f1-score. The VGG was the next model trained on the
dataset, and the results obtained in terms of accuracy, precision, recall, and f1-score
were anticipated. Finally, we used a hybrid model that offered greater precision,
recall, accuracy, and f1-score compared to the rest of the models trained and tested.
With the hybrid model, we can forecast the diseases that affect wheat leaves with a
sizable degree of accuracy, precision, recall, and f1-score. The results are shown in
Fig. 20.7.

20.8 Conclusion

The aim of this work is to propose Whe-C-Net a hybrid deep learning model for the
detection of diseases in wheat leaves. For the experimental study the technique is
applied to a standard dataset of wheat plants downloaded from Mendeley. The model
336 N. P. S. Rathor et al.

Fig. 20.7 Evaluative comparative analysis of different CNN-based classification models

Whe-C-Net has performed fairly significantly and generated an optimum validation

accuracy of 98%, compared to ResNet-50 (65%), Xception13 (52%), MobileNet
(81%), and VGG (93%). A further prerequisite is the gathering of actual images from
the fields in order to increase the effectiveness of the Whe-C-Net model and utility in
the area of agriculture for farmers to detect diseases in wheat plants using leaves.
Thus this hybrid model is used to boost accuracy without lengthening training time
because the simple CNN typically performs poorly in data augmentation. The
outcome of the suggested hybrid model Whe-C-Net describes the deep learning
model that can be used to enhance the diagnosis of plant disease using the leaves in
comparison to the conventional manual approaches.

References

Agarwal G, Belhumeur P, Feiner S, Jacobs D, Kress WJ, Ramamoorthi R, Bourg A, Dixit N,

Ling H, Mahajan D et al (2006) First steps toward an electronic field guide for plants. Taxon
55(3):597–610
Akila M, Deepan P (2018) Detection and classification of plant leaf diseases by using deep learning
algorithm. Int J Eng Res Technol 6(07)
Barbedo JG (2018) Factors influencing the use of deep learning for plant disease recognition.
Biosyst Eng 172:84–91
Barré P, Stöver BC, Müller KF, Steinhage V (2017) Leafnet: a computer vision system for
automatic plant species identification. Ecol Informat 40:50–56
Boulent J, Foucher S, Théau J, St-Charles PL (2019) Convolutional neural networks for the
automatic identification of plant diseases. Front Plant Sci 10:941
Brahimi M, Boukhalfa K, Moussaoui A (2017) Deep learning for tomato diseases: classification and
symptoms visualization. Appl Artif Intell 31(4):299–315
20 Comprehensive Analysis of Deep Learning Models for Plant Disease Prediction 337

Chaudhari SB, Wagaskar V, Shaikh M, Shelke O, Shirsath V (2021) Plant disease detection
implementation using tensorflow. Int Res J Mod Eng Technol Sci 3(6)
Chen J, Chen J, Zhang D, Sun Y, Nanehkaran YA (2020) Using deep transfer learning for image-
based plant disease identification. Comput Electron Agric 173:105393., ISSN 0168-1699.
https://doi.org/10.1016/j.compag.2020.105393
Chollet F (2017) Xception: deep learning with depthwise separable convolutions. In: Proceedings
of the IEEE conference on computer vision and pattern recognition, pp 1251–1258
Dandawate, Y., & Kokare, R. (2015). An automated approach for classification of plant diseases
towards development of futuristic Decision Support System in Indian perspective. In 2015
International conference on advances in computing, communications and informatics (ICACCI)
(pp. 794–799). IEEE
Devaraj, A., Rathan, K., Jaahnavi, S., & Indira, K. (2019). Identification of plant disease using
image processing technique. In 2019 International Conference on Communication and Signal
Processing (ICCSP) (pp. 0749–0753). IEEE
Dos Santos Ferreira A, Freitas DM, da Silva GG, Pistori H, Folhes MT (2017) Weed detection in
soybean crops using convnets. Comput Electron Agric 143:314–324
FAO (n.d.). http://www.fao.org/india/fao-in-india/india-at-a-glance/en/
Ferentinos KP (2018) Deep learning models for plant disease detection and diagnosis. Comput
Electron Agric 145:311–318
Fuentes A, Lee Y, Hong Y, Yoon S, Park D (2016) Characteristics of Tomato Plant Diseases—A
study for tomato plant disease identification
Fuentes A, Yoon S, Kim SC, Park DS (2017) A robust deep-learning-based detector for real-time
tomato plant diseases and pests recognition. Sensors 17(9):2022
Gerhards R, Christensen S (2003) Real-time weed detection, decision making and patch spraying in
maize, sugarbeet, winter wheat and winter barley. Weed Res 43(6):385–392
Goutte C, Gaussier E (2005) A probabilistic interpretation of precision, recall and F-score, with
implication for evaluation. Lect Notes Comput Sci 3408:345–359. https://doi.org/10.1007/978-
3-540-31865-1_25
Goyal L, Sharma CM, Singh A, Singh PK (2021) Leaf and spike wheat disease detection &
classification using an improved deep convolutional architecture. Inform Med Unlocked 25:
100642
Grinblat GL, Uzal LC, Larese MG, Granitto PM (2016) Deep learning for plant identification using
vein morphological patterns. Computer Electron Agric 127:418–424
Hasan MM, Chopin JP, Laga H, Miklavcic SJ (2018a) Detection and analysis of wheat spikes using
convolutional neural networks. Plant Methods 14(1):1–13
Hasan MM, Chopin JP, Laga H, Miklavcic SJ (2018b) Detection and analysis of wheat spikes using
convolutional neural networks. Plant Methods (14):100. https://doi.org/10.1186/s13007-018-
0366-8. Erratum in: Plant Methods 2019 Mar 20;15:27. PMID: 30459822; PMCID:
PMC6236889
He K, Zhang X, Ren S, Sun J (2015) Spatial pyramid pooling in deep convolutional networks for
visual recognition. IEEE Trans Pattern Anal Mach Intell 37(9):1904–1916
Horaisová K, Kukal J (2016) Leaf classification from binary image via artificial intelligence.
Biosyst Eng 142:83–100
Howard AG et al (2017) Mobilenets: efficient convolutional neural networks for mobile vision
applications. arXiv:1704.04861
Husin Z, Shakaff A, Aziz A, Farook R, Jaafar M, Hashim U, Harun A (2012) Embedded portable
device for herb leaves recognition using image processing techniques and neural network
algorithm. Comput Electron Agric 89:18–29
Jeon W-S, Rhee S-Y (2017) Plant leaf recognition using a convolution neural network. Int J Fuzzy
Logic Intell Syst 17(1):26–34
Johannes A, Picon A, Alvarez-Gila A, Echazarra J, Rodriguez-Vaamonde S, Navajas AD, Ortiz-
Barredo A (2017) Automatic plant disease diagnosis using mobile capture devices, applied on a
wheat use case. Comput Electron Agric 138:200–209
338 N. P. S. Rathor et al.

Kathiresan G, Anirudh M, Nagharjun M, Karthik R (2021) Disease detection in rice leaves using
transfer learning techniques. J Phys Conf Ser 1911(1):012004. IOP Publishing
Kaur S, Joshi G, Vig R (2019) Plant disease classification using deep learning Google net model. Int
J Innovat Technol Explor Eng 8(9):319–322
Kranth GPR, Lalitha MH, Basava L, Mathur A (2018) Plant disease prediction using machine
learning algorithms. Int J Comput App:1–7
Larese MG, Namías R, Craviotto RM, Arango MR, Gallo C, Granitto PM (2014a) Automatic
classification of legumes using leaf vein image features. Pattern Recogn 47(1):158–168
Larese MG, Baya AE, Craviotto RM, Arango MR, Gallo C, Granitto PM (2014b) Multiscale
recognition of legume varieties based on leaf venation images. Expert Syst Appl 41(10):
4638–4647
LeCun Y, Bengio Y, Hinton G (2015) Deep learning. Nature 521:436–444. https://doi.org/10.1038/
nature14539
Lee, S. H., Chan, C. S., Wilkin, P., & Remagnino, P. (2015). Deep-plant: plant identification with
convolutional neural networks. In 2015 IEEE International Conference on Image Processing
(ICIP) (pp. 452–456). IEEE
Liu B, Zhang Y, He D, Li Y (2018) Identification of apple leaf diseases based on deep convolutional
neural networks. Symmetry 10(1):11
Lu J, Hu J, Zhao G, Mei F, Zhang C (2017a) An in-field automatic wheat disease diagnosis system.
Comput Electron Agric 142:369379
Lu Y, Yi S, Zeng N, Liu Y, Zhang Y (2017b) Identification of rice diseases using deep
convolutional neural networks. Neurocomputing 267:378–384
Lumini A, Nanni L (2019) Deep learning and transfer learning features for plankton classification.
Ecol Inform 51:33–43
Ma J, Du K, Zheng F, Zhang L, Gong Z, Sun Z (2018) A recognition method for cucumber diseases
using leaf symptom images based on deep convolutional neural network. Comput Electron
Agric 154:18–24
Mattila H, Valli P, Pahikkala T, Teuhola J, Nevalainen OS, Tyystjärvi E (2013) Comparison of
chlorophyll fluorescence curves and texture analysis for automatic plant identification. Precision
Agric. 14(6):621–636
Mohammed A (2020) Wheat rust images for diseases map. V1. https://doi.org/10.17632/
25g6cm8vhb.1
Mohanty SP, Hughes DP, Salathé M (2016) Using deep learning for image-based plant disease
detection. Front Plant Sci. https://doi.org/10.3389/fpls.2016.01419. Article: 1419, 7
Murat M, Chang S-W, Abu A, Yap HJ, Yong K-T (2017) Automated classification of tropical shrub
species: a hybrid of leaf shape and machine learning approach. PeerJ 5:e3792
Muthevi, A., Uppu, R.B., 2017. Leaf classification using completed local binary pattern of
textures. In: 2017 IEEE 7th International Advance Computing Conference (IACC).IEEE,
pp. 870–874
Neto JC, Meyer GE, Jones DD, Samal AK (2006) Plant species identification using elliptic Fourier
leaf shape analysis. Comput Electron Agric 50(2):121–134
Pantazi XE, Moshou D, Tamouridou AA (2019) Automated leaf disease detection in different crop
species through image features analysis and one class classifiers. Comput Electron Agric 156:
96–104
Patrick MK, Adekoya AF, Mighty AA, Edward BY (2019) Capsule networks—a survey. J King
Saud Univ Comput Inf Sci
Picon A, Alvarez-Gila A, Seitz M, Ortiz-Barredo A, Echazarra J, Johannes A (2019) Deep
convolutional neural networks for mobile capture device-based crop disease classification in
the wild. Comput Electron Agric 161:280–290
Pydipati R, Burks T, Lee W (2006) Identification of citrus disease using color texture features and
discriminant analysis. Comput Electron Agric 52(1–2):49–59
M. B. Riley, M. R. Williamson, and O. Maloy, “Plant disease diagnosis. The Plant Health
Instructor,” 2002
20 Comprehensive Analysis of Deep Learning Models for Plant Disease Prediction 339

Sabanci K, Akkaya M (2016) Classification of different wheat varieties by using data mining
algorithms. Int J Intell Syst App Eng 4(2):40–44
S. Sabour, N. Frosst, G.E. Hinton, Dynamic routing between capsules, 31st Conference on Neural
Information Processing Systems (2017)
Sack L, Dietrich EM, Streeter CM, Sánchez-Gómez D, Holbrook NM (2008) Leaf palmate venation
and vascular redundancy confer tolerance of hydraulic disruption. Proc Nat Acad Sci 105(5):
1567–1572
Sarayloo, Z., & Asemani, D. (2015a). Designing a classifier for automatic detection of fungal
diseases in wheat plant: by pattern recognition techniques. In 2015 23rd Iranian Conference on
Electrical Engineering (pp. 1193–1197). IEEE
Sarayloo Z, & Asemani, D. (2015b). Designing a classifier for automatic detection of fungal
diseases in wheat plant: By pattern recognition techniques. In 2015 23rd Iranian Conference
on Electrical Engineering (pp. 1193–1197). IEEE
Scoffoni C, Rawls M, McKown A, Cochard H, Sack L (2011) Decline of leaf hydraulic conduc-
tance with dehydration: relationship to leaf size and venation architecture. Plant Physiol 156(2):
832–843
Simonyan, K, and Zisserman A. Very deep convolutional networks for large-scale image recogni-
tion arXiv:1409.1556 (2014)
Sladojevic S, Arsenovic M, Anderla A, Culibrk D, Stefanovic D (2016) Deep neural networks
based recognition of plant diseases by leaf image classification. Comput Intell Neurosci 2016
Too EC, Yujian L, Njuki S, Yingchun L (2019) A comparative study of fine-tuning deep learning
models for plant disease identification. Comput Electron Agric 161:272–279. https://doi.org/10.
1016/j.compag.2018.03.032. ISSN 0168-1699; (https://www.sciencedirect.com/science/article/
pii/S0168169917313303)
Tyystjärvi E, Nørremark M, Mattila H, Keränen M, Hakala-Yatkin M, Ottosen C-O, Rosenqvist E
(2011) Automatic identification of crop and weed species with chlorophyll fluorescence induc-
tion curves. Precision Agric 12(4):546–563
Wang N, Zhang N, Wei J, Stoll Q, Peterson D (2007) A real-time, embedded, weed detection
system for use in wheat fields. Biosyst Eng 98(3):276–285
Woebbecke DM, Meyer GE, Von Bargen K, Mortensen D (1995) Color indices for weed identi-
fication under various soil, residue, and lighting conditions. Trans ASAE 38(1):259–269
Yousefi E, Baleghi Y, Sakhaei SM (2017) Rotation invariant wavelet descriptors, a new set of
features to enhance plant leaves classification. Comput Electron Agric 140:70–76
Yu, X., Xiong, S., Gao, Y., Zhao, Y., Yuan, X., 2016. Multiscale crossing representation using
combined feature of contour and venation for leaf image identification. In: 2016 International
Conference on Digital Image Computing: Techniques and Applications (DICTA). IEEE,
pp. 1–6
Chapter 21
Enhancing Single-Cell Trajectory Inference
and Microbial Data Intelligence

Bhargavi Posinasetty, Mukesh Soni, Sagar Dhanraj Pande,

Krishnendu Adhikary, and Dhirendra Kumar Tripathi

Abstract The utilization of single-cell trajectory inference methods to deduce cell

differentiation trajectories from single-cell transcriptomic or proteomic data holds
significant importance in comprehending the developmental mechanisms of healthy
tissues and offering valuable insights into pathological conditions. Nonetheless, the
enhancement of accuracy and resilience in existing algorithms for inferring single-
cell trajectories presents a persistent obstacle, mostly attributable to the interference
caused by the identification of unrelated genes during single-cell sequencing. In
order to effectively tackle this matter and expand the suitability of these methods to a
wider array of biological data, we introduce iterTIPD, a trajectory inference method
that utilizes iterative feature selection. The Iterative Topological Feature Selection
(iterTIPD) algorithm is a widely employed approach in the field of genomics for the
purpose of detecting differentially expressed genes. It is specifically designed to be
applied repeatedly on linear or branching single-cell RNA sequencing data. The
described iterative approach involves the selection of a subset of genes that make a
significant contribution to the construction of the differentiation trajectory. This
selection leads to enhanced precision and robustness in the ordering of cell
pseudo-time. Furthermore, iterTIPD exhibits its efficacy not only in conventional
single-cell data analysis but also in the realm of microbial data intelligence. The

B. Posinasetty
Department of Masters in Public Health, The University of Southern Mississippi, Hattiesburg,
MS, USA
M. Soni (✉)
Department of CSE, University Centre for Research & Development Chandigarh University,
Mohali, Punjab, India
S. D. Pande
School of Computer Science and Engineering, VIT-AP University, Amaravati, Andhra Pradesh,
India
K. Adhikary
Centurion University of Technology and Management, Bhubaneswar, Odisha, India
D. K. Tripathi
Sri Satya Sai University of Technology and Medical Sciences, Sehore, MP, India

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 341
A. Khamparia et al. (eds.), Microbial Data Intelligence and Computational
Techniques for Sustainable Computing, Microorganisms for Sustainability 47,
https://doi.org/10.1007/978-981-99-9621-6_21
342 B. Posinasetty et al.

experimental findings obtained from the study of four scRNA-seq data sets demon-
strate that iterTIPD significantly improves the accuracy and reliability of single-cell
trajectory inference techniques. This enhancement renders iterTIPD a valuable
resource for researchers across many fields, including the analysis of microbial
data. Additionally, iterTIPD not only improves the efficiency of trajectory inference
methods but also has robust generalization abilities. The iterTIPD method effectively
reconstructs the differentiation track of neural stem cells, demonstrating a strong
alignment with established brain progenitor cell differentiation pathways. Moreover,
the present research demonstrates that Top2a and Gjal may serve as promising novel
markers for characterizing activated neural progenitor cell subgroups. This finding
underscores the algorithm’s capacity to uncover biologically significant information.

Keywords Single-cell trajectory inference · iterTIPD · Microbial data intelligence ·

Gene selection · Pseudo-time estimation · Neural stem cells

21.1 Introduction

Single-cell RNA sequencing (scRNA-seq) has become a powerful tool for studying
cellular heterogeneity (Karbalayghareh et al. 2019; Gan et al. 2022) and early phases
of embryonic development (Ni et al. 2022; Chen et al. 2020), thanks to its ability to
analyse the transcriptome at the level of individual cells. The method shown here is
well suited to answering important questions in biology. Quality control (Ni et al.
2022; Chen et al. 2020), batch effect correction (Wu and Ma 2023; Lee et al. 2015),
data standardization (Lee et al. 2015; Zeng et al. 2021), feature selection (Lee et al.
2015; Zeng et al. 2021), dimensionality reduction (Zhang et al. 2023; Lu et al. 2021),
clustering (Ma et al. 2020; Seabolt et al. 2022), identifying cell subpopulations
(Ma et al. 2020; Wassan et al. 2019), and trajectory inference (Ma et al. 2020;
Cawley et al. 2006) are just a few of the many methods available for analysing
single-cell RNA sequencing (scRNA-seq) data. Inferring a cell’s specific path of
development is one of the many active research areas nowadays. Introducing genetic
markers into cells and then following their offspring’s development has always been
the method of choice for tracing lineages (Li et al. 2021). However, the existence of
only a few genetic markers may mask the diversity that actually occurs within
cellular populations. Single-cell transcriptomic data or protein expression profiles
are analysed to infer the course of cellular differentiation.
Reconstruct the lineage relationship (differentiation trajectory) of cells automat-
ically using omics data, categorize cells depending on their position in the differen-
tiation trajectory, and use this information to probe the dynamic expression of genes.
Pseudo-times are assigned to each cell using these methods of calculation; sorting
cells based on pseudo-time is known as pseudo-time sorting. A single cell’s pro-
gression through the many transitional stages of differentiation can be described in
terms of pseudo-time, which is the sequence of cells that follows a dynamic process
of continuous development in biological systems. Therefore, cells at the start and
21 Enhancing Single-Cell Trajectory Inference and Microbial Data Intelligence 343

finish points of a trajectory, as well as cells at intermediate points, can all be

identified by pseudo-temporal sorting (Zeng et al. 2020).
Trajectory inference methods based on single-cell sequencing data are mainly
used in the following two aspects:
1. Identify transition states: In many biological systems, cells exhibit continuous
state transitions, characterized by different changes in transcription, morphology,
epigenome, and surface markers (Lee and Friderikos 2022). Using trajectory
analysis based on single-cell sequencing data can provide a more direct and
unbiased method to identify and correctly sequence different transition stages,
that is, reconstruct the lineage trajectory of cells based on scRNA-seq data and
discover new transition states (Wang et al. 2022).
2. Identify key regulatory factors: In addition to revealing gene expression dynamics
between cells, single-cell trajectory inference can also help identify key regula-
tory factors that trigger state transitions. For example, in the study of human
definitive endoderm cell development (Zhang and Zhang 2020), single-cell
trajectory inference was used to arrange cells along developmental trajectories,
successfully reconstructing known differentiation pathways, and also discovering
new candidates.
Essentially, the application of trajectory inference methods to single-cell RNA
sequencing (scRNA-seq) data may reveal connections made over the course of cell
differentiation, providing important new understandings of the variability of gene
expression and cell dynamics. Therefore, it is of great scholarly significance to
explore trajectory inference methods using single-cell RNA sequencing (scRNA-
seq) data. However, considering the intrinsic characteristics of scRNA-seq data—
high noise and heterogeneity (Afshar et al. 2016)—as well as the dropout effect,
which occurs when low-expression genes are difficult to find owing to technological
limitations (Li et al. 2021), a more accurate and robust method must be developed.
Techniques for cell differentiation trajectories remain challenging. Currently, there
are more than 70 different methods available in the field of single-cell trajectory
inference (Li et al. 2023). To deal with the problem of high-dimensional noise in
scRNA-seq data, single-cell trajectory inference frequently uses feature dimension-
ality reduction and feature selection algorithms.
Monocle (Ma et al. 2020) is considered the inaugural and archetypal single-cell
trajectory inference technique within the realm of methods employing feature
dimensionality reduction. Nevertheless, due to the advancements in contemporary
single-cell RNA sequencing (scRNA-seq) techniques, which enable the simulta-
neous measurement of several cells, the construction of a minimal spanning tree
(MST) connecting a substantial number of cells has become intricate and challeng-
ing to comprehend. Consequently, Monocle, a computational tool, lacks the capa-
bility to accurately forecast such complexities. In 2017, the proposal of Monocle2
(Wassan et al. 2019) introduced a method that generated a lineage tree within a
low-dimensional space. This approach effectively preserved a greater amount of
information regarding the differentiation trajectory present in the original data, hence
addressing the challenge of forecasting branch trajectories. The Monocle and
344 B. Posinasetty et al.

Monocle2 algorithms have been employed to deduce diverse forms of cellular

differentiation trajectories. These include the inference of neuronal lineage trajecto-
ries (Muller et al. 1989; Dai et al. 2018), the differentiation of progenitor and stem
cells within the haematopoietic system (Wassan et al. 2019), as well as investigations
into the development of the placenta (Iriya et al. 2020) and other related studies. The
SCUBA (single-cell clustering using bifurcation analysis) method, as described in
reference (Geard and Wiles 2005), employs t-distributed stochastic neighbour
embedding (t-SNE) (Hu et al. 2016) to effectively reduce the dimensionality of the
data. Subsequently, the dimensionally reduced data is subjected to a smoothing
process. The utilization of SCUBA has been employed to determine crucial genes
involved in the process of stem cell differentiation within the small intestine of mice
(Zhang and Zhang 2021). The Slingshot algorithm utilizes the dimensionally
reduced data and clustering outcomes as its input to create a minimum spanning
tree (MST). Subsequently, it enhances the tree structure by using smooth curve
fitting techniques to each branch of the MST. Ultimately, each cell is projected onto
the curve that is nearest to it, leading to the emergence of a lineage trajectory that is
organized and exhibits branching patterns. The utilization of slingshot has been
employed in recent studies to forecast the cellular destiny and branching sites within
the lineage trajectories of olfactory stem cells (Deconinck et al. 2021).
Different from feature dimensionality reduction methods, using feature selection
methods can reduce the dimensionality of genes and identify the most relevant
features, improving signal-to-noise ratio and computational efficiency of down-
stream analysis (such as clustering or pseudo-time inference). Effective feature
selection methods can allow researchers to focus on the analysis of key genes,
thereby improving the performance of trajectory inference methods that pseudo-
temporally sort cells based on gene expression profile similarity. Some current
feature selection methods used in single-cell trajectory inference mainly use the
average expression of genes in all cells to screen highly variable genes (HVGs),
thereby reducing the dimensionality of the data. However, due to the dropout effect
in single-cell data, screening highly variable genes directly from the data may miss
some important low-expressed genes carrying effective information. In bulk RNA
sequencing (bulk RNA-seq) experiments, screening differentially expressed genes is
usually used as a feature selection method, but scRNA-seq data generally does not
have information about cell subpopulations (in fact, this is also the goal of analysing
scRNA-seq data), so screening of differentially expressed genes is difficult to
directly apply to scRNA-seq data. Based on the problems mentioned above, this
chapter proposes an iterative feature selection optimization method (iterative trajec-
tory inference based on probability distribution, iterTIPD) based on a trajectory
inference method based on probability distribution—TIPD algorithm (Liu et al.
2022). The premise assumptions are:
If a gene plays a role in the final constructed single-cell differentiation trajectory,
it should be differentially expressed along the cell’s trajectory (Wassan et al. 2019).
The work and contributions of this chapter can be summarized as follows:
21 Enhancing Single-Cell Trajectory Inference and Microbial Data Intelligence 345

1. iterTIPD was run on four single-cell data sets, and appropriate iteration param-
eters were determined experimentally. Moreover, in order to explore the effec-
tiveness of this iterative feature selection method, this chapter compares iterTIPD,
TIPD, and two other commonly used feature selection methods in terms of
accuracy, robustness, and ability to detect gold standard genes.
2. The differentiation trajectory of neural stem cells (NSCs) was successfully
reconstructed based on the optimized iterTIPD algorithm, and through compar-
ison, it was found that the differentiation trajectory was highly consistent with the
known differentiation trajectory of neural stem cells.
3. The genes Top2a and Gja1 were found to be differentially expressed in three
different states of activated neural stem cell populations, which may be new
markers of activated neural stem cell populations.

21.2 iterTIPD

iterTIPD is an iterative feature selection algorithm that can be widely used to screen
differentially expressed genes on single-cell RNA sequencing data with linear or
branched structures. By filtering out the subset of genes that contribute the most to
the constructed differentiation trajectory, the accuracy and robustness of pseudo-
time sorting of cells are improved, thereby improving the performance of the single-
cell trajectory inference algorithm. The following explains the iterTIPD algorithm
proposed in this chapter based on the TIPD (Liu et al. 2022) algorithm.

21.2.1 TIPD Algorithms

The TIPD method quantifies the cellular differentiation potential by evaluating the
signalling entropy (SR) of individual cells, enabling the assessment of diverse states
within the cell population. Single-cell RNA sequencing (scRNA-seq) can be
employed to provide a quantitative assessment of the differentiation state of indi-
vidual cells. A high SR value indicates a heightened capacity for differentiation,
whereas a low SR value signifies a diminished aptitude for differentiation. The TIPD
method integrates signal entropy and clustering outcomes of gene expression data to
characterize the probability distribution of diverse states within a population of cells.
The Jensen-Shannon divergence (JSD), also known as the symmetric JS divergence,
is employed to quantify the dissimilarity between probability distributions of cell
clusters. Subsequently, the minimum spanning tree (MST) is computed on the full
network formed by the centres of these cell clusters and their corresponding prob-
ability distribution distances. The aforementioned procedures establish the funda-
mental framework of the cellular development trajectory. Subsequently, all cells are
mapped onto the curve-fitted framework, resulting in the derivation of the pseudo-
time sequence of cells.
346 B. Posinasetty et al.

The MST thus constructed is calculated based on system-level information in the

context of signal entropy, so it can reflect the real biological process globally,
making the TIPD algorithm show higher performance than other trajectory inference
methods.

21.2.2 Iterative Feature Selection of iterTIPD Algorithm

The output of the single-cell trajectory inference method is sorted according to the
pseudo-time of cell differentiation. It can be considered that individual cells are in
different states. Therefore, differentially expressed genes can be calculated in
pseudo-time order as a feature selection method for scRNA-seq data. The TIPD
algorithm is used here to calculate pseudo-time (the part marked by the green dotted
box in Fig. 21.1).

Gene

Cell

TIPD t-SNE Single Entropy

K-means H-State

Pseudo Time Calculation

The highest accuracy of the New Feature Sub

output is the corresponding time

Tobit Model Pair Genetic Modeling

GAM Model Fitting

Calculate the Corrected p-Value of

Screen Differentially Expressed

Genes

Fig. 21.1 iterTIPD algorithm flow chart

21 Enhancing Single-Cell Trajectory Inference and Microbial Data Intelligence 347

Existing studies have shown that the generalized additive model (GAM)
(Ma et al. 2020) is used to calculate differentially expressed genes in pseudo-time
and detect gold standard genes. Feature selection can be performed by fitting a GAM
to the relationship between a gene’s expression and pseudo-time. First, the expres-
sion level of each gene in the cell is modelled through the Tobit model (Wei et al.
2021), and then the relationship between the expression of each gene and pseudo-
time is fitted using the GAM model, where the expression level of each gene Y
depends on the latent variable Y
, the expression is:

Y
, Y
⩾ λ
Y= λ, Y
< λ
ð21:1Þ

where λ is the detection threshold, and λ is set to 0.1 by default. The latent
variable Y
depends on the variable xi in the GAM model, where xi represents the
pseudo-time value corresponding to each cell, i 2 {1, 2, . . ., n}. Therefore, the
expression of the GAM model of differentially expressed genes in pseudo-time is
calculated. The formula is defined as:

EðY Þ = sðψ ðbx , gi ÞÞ þ ε ð21:2Þ

Among them, gi represents a single cell, i 2 {1, 2, ⋯, n}, ψ t(bx, gi) represents the
pseudo-time value assigned to cell gi, and s represents a cubic smooth function with
an effective degree of freedom of 3. The error term ε is normally distributed with
mean 0. Finally, differentially expressed genes were iteratively screened through
eq. (3).

Gt = a j FDRpta- 1 < 0:05, a 2 Gt - 1 ð21:3Þ

Among them, Gt represents the gene set after iterative screening t times; pta- 1 is
the p-value of gene a calculated using the approximate χ 2 likelihood ratio test on
Gt - 1; and FDRpta- 1 is the Benjamini-Hochberg test of pat - 1 p-value of gene a
obtained after FDR correction (Flexman et al. 2006). Genes with FDRpta- 1 less than
0.05 in each iteration were retained for the next iteration, and these genes were
considered to be differentially expressed genes. The implementation of the GAM
model and related test functions in the above process is provided by the “VGAM”
package (Wei et al. 2019).
The differentially expressed genes in pseudo-time order calculated in the above
steps will form a new feature subset for a new round of calculation. The number of
iterations is controlled by the specified parameters, and the default is 100, which can
be determined according to the accuracy change trend of different data sets during
the iteration process (see Sect. 2.3). After the algorithm performs multiple iterations
with reference to the iteration parameters, it stops iteration when the accuracy shows
an obvious downward trend and outputs the pseudo-time sorting results of cells with
the highest accuracy in all calculations and the corresponding gene subsets.
348 B. Posinasetty et al.

(a) Linear Trajectory

(b) Branch Trajectory

Fig. 21.2 Schematic diagram of pseudo-time sequencing of cells in (a) linear trajectories and (b)
branch trajectories

If the trajectory constructed by the trajectory inference method is a linear trajec-

tory, as shown in Fig. 21.2a, the pseudo-time sequence of differentially expressed
genes is directly calculated as a feature subset for screening, if it is a branch
trajectory, as shown in Fig. 21.2b, it contains. For the differentiation trajectory of
a branch structure, calculate the differentially expressed genes in pseudo-time order
for each lineage (such as a-b-c and a-b-d), and finally find the union of these
differentially expressed genes as the feature subset for screening. Because the feature
subset used in each round is a gene subset calculated based on the gene set of the
previous round, as the number of iterations increases, the number of screened genes
gradually decreases, and some genes containing effective information may be lost,
so the number of iterations should not be too many.

21.3 Experiment

21.3.1 Data Set Description

This study utilizes four single-cell RNA sequencing (scRNA-seq) data sets, which
exhibit variations in terms of their sizes and encompass both linear and branched
structures. Furthermore, these data sets encompass species such as humans and mice.
The individuals in question can be identified as follows:
1. Mouse lung alveolar type 2 cells, often known as AT2 cells. The data set known
as AT2 comprises 101 cells of alveolar type 2 in mice. These cells were obtained
from the lungs of embryonic mice at four different time points: E14.5, E16.5,
E18.5, and adult. This information was taken from a previous study (Dai et al.
2018). The trajectory of differentiation seen in this study exhibits a linear shape.
2. Human skeletal muscle myoblasts (HSMM). The HSMM data set has a total of
271 human skeletal myoblasts, encompassing three distinct cell types:
21 Enhancing Single-Cell Trajectory Inference and Microbial Data Intelligence 349

proliferating cells, developing myoblasts, and interstitial mesenchymal cells

(Ma et al. 2020). The trajectory of differentiation observed in this context exhibits
a linear shape.
3. Mouse embryonic stem cells (ESC) are a type of pluripotent cells derived from
the inner cell mass of mouse embryos. The ESC data set consists of 2717 cells
that were collected at four different time intervals (d0, d2, d4, and d7) following
the withdrawal of leukaemia inhibitory factor (LIF). The study utilized mouse
embryonic stem cells (Schiebinger 2021). The trajectory of differentiation
exhibits a linear structure.
Mouse conventional dendritic cells (cDCs) are a type of dendritic cell seen
in mice.
The cDC collection is made up of 251 normal dendritic cells (cDCs) from mice.
This type of cDC is made up of three different types of cells: pre-dendritic cells
(pre-DCs), macrophage dendritic cell progenitors (MDPs), and common dendritic
cell progenitors (CDPs) (Geard and Wiles 2005). There are two main types of
conventional dendritic cells (cDCs) in mice. These are cDC1 and cDC2 (Rezk
et al. 2023; Nguyen et al. 2022). Three different types of cell lines are controlled
by different transcriptional programmes. These programmes start when macro-
phages, monocytes, and dendritic cells (MDPs) differentiate. The next step is the
formation of common dendritic cell progenitors (CDPs) and then pre-dendritic cells
(pre-DCs). These cells eventually differentiate into cDC1 and cDC2 lines (Chang
et al. 2017). There are branches in the process of differentiation.

21.3.2 Evaluation Index

When evaluating the efficacy of algorithms in the field of single-cell differentiation

trajectory inference, researchers commonly employ accuracy, robustness, and the
ability to detect gold standard genes as complete metrics (Seabolt et al. 2022). This
chapter also employs these three variables to assess the efficacy of the trajectory
inference technique. This chapter uses pseudo-time sorting of cells to evaluate
accuracy. Accuracy is assessed by calculating the Kendall rank correlation coeffi-
cient between the externally recorded information in the experimental stage and the
data set (Sahayasheela et al. 2022; Chang et al. 2019). The assessment of resilience
can be conducted by computing a robustness score for the pseudo-time ordering
of both the original and perturbed data (Seabolt et al. 2022). The generation of
perturbed data sets involved the random subsampling of 90%, 80%, and 70% of cells
from the original data set. This process was repeated 50 times to provide several
duplicates. The mathematical expression representing the robustness score is as
follows:
350 B. Posinasetty et al.

2
Sπ1 ,π2 = hðπ 1 , π 2 , i, jÞ ð21:4Þ
j A A j - 1Þ i, j2A; i ≠ j

Among them, π 1 and π 2 represent the ordering of two pseudo-times, A is the

union of all cells in π 1 and π 2, and jAj represents the cardinality of the set A. For a
pair of cells i and j, if their order in π 1 is the same as in π 2, then h(π 1, π 2, i, j) = 1,
otherwise h(π 1, π 2, i, j) = 0. The higher the robustness score, the more similar the
two pseudo-time rankings are.
Each feature selection method’s ability to improve the detection of gold standard
genes is measured in this study using the mean ranking of gold standard genes
(Seabolt et al. 2022). For a given data set of gene expression measurements, the
“gold standard” genes are found by carefully combing through the corresponding
scholarly literature. Differentiation and development of cells, as well as other
biological events, are known to cause these genes to be expressed in unique ways.
After applying false discovery rate (FDR) correction, the list of differentially
expressed genes is ranked by their p-values (Wei et al. 2019). A measure of the
algorithm’s performance in detecting known differentially expressed genes is the
location of gold standard genes within the set of differentially expressed genes. As
the ranking and corresponding numerical value rise, the algorithm’s ability to detect
known differentially expressed genes improves.

21.3.3 Confirmation of iterTIPD Iteration Parameters

The iterTIPD technique can screen candidate feature genes in a recursive fashion. If
the number of iterations through the method is too little, the screened gene subset
may contain many false positives. Important feature genes could be lost if the
process is repeated too many times. In this piece, we use iterTIPD on the four
single-cell data sets described in Sect. 3.1. At the outset, we run 100 iterations.
Variations in the correlation between iterations and precision across data sets are
tracked. Figure 21.3 displays the obtained data. When the number of iterations is set
to 0, the associated accuracy is the accuracy attained by TIPD. The number of
iterations where the accuracy improved the most is indicated by the yellow dot.
The experimental findings were fitted with a polynomial equation, yielding the red
dotted line as the trend line. The accuracy of the four single-cell data sets showed a
general trend of first growing and then dropping as the number of iterations rose, as
seen in Fig. 21.3, during the iteration process. Because the accuracy is also
influenced by the dimensionality reduction phase in the iterTIPD algorithm, it varies.
The accuracy trend first rises, then falls. This is because the accuracy of the
algorithm is enhanced by increasing the number of iterTIPD iterations when the
data contains more redundant features, which in turn reduces the feature subset.
When accuracy is the highest, valid information-carrying genes are the first to be
filtered out, leading to a decline in accuracy as the number of features is reduced
21 Enhancing Single-Cell Trajectory Inference and Microbial Data Intelligence 351

Fig. 21.3 (a) Number of iterTIPD Iterations and Accuracy. (b) Number of iterTIPD iterations and
accuracy

further. The declining accuracy trend for the four single-cell data sets becomes
noticeable as the number of repeats approaches 100. For this reason, the initial
value for iterTIPD is 100. In the future, researchers can use this parameter as a
starting point for inferring cell differentiation trajectories from different data sets
using iterTIPD. After the iteration is complete, iterTIPD will return the results of the
most accurate pseudo-time sorting of cells throughout the iteration, along with the
subsets of genes that were sorted. On the AT2 and cDC data sets, the highest
accuracy is reached at the 40th, 56th, 43rd, and 62nd iterations (the number of
iterations to reach the greatest accuracy and the associated highest accuracy are
marked in brackets) (see Fig. 21.3a, b). Overall, iterTIPD’s optimization effect is
positive across all four single-cell data sets, but it’s most pronounced on the ESC
data set, where it improves accuracy by 0.09.
352 B. Posinasetty et al.

21.3.4 Analysis of the Effectiveness of iterTIPD

This chapter selected the feature selection method provided by two popular software
packages, the “Monocle” package (Ma et al. 2020) and the “Seurat” package (Demir
and Eren 2022) to screen highly variable genes (HVGs), combined with the TIPD
algorithm (named M_HVGs respectively). +TIPD and S _ HVGs + TIPD to compare
the optimization effect of iterTIPD’s iterative feature selection method. Among
them, the function parameters used to screen HVGs in Monocle and Seurat use
default parameters, and the process of screening HVGs corresponds to the gene
selection step in iterTIPD and TIPD (select the normalized gene expression level to
be greater than 1 in at least 30% of the cells genes.); subsequent steps, such as
dimensionality reduction, remain consistent. In addition, in order to explore the
generalization ability of iterTIPD’s feature selection method, this chapter applies it
to the gene selection step in Monocle2 (Wassan et al. 2019), and this method is
named iter+Monocle2.

21.3.4.1 Accuracy Analysis

Table 21.1 shows that the accuracy of iterTIPD on AT2, HSMM, ESC, and cDC data
sets is higher than that of TIPD. The accuracy of the algorithm has been improved by
0.028, 0.062, 0.089, and 0.040, respectively, indicating that the iterTIPD iterative
feature selection method has effectively improved TIPD’s accuracy degree, and the
accuracy is the highest on the other three single-cell data sets, except the ESC data
set. The M_HVGs+TIPD method only improves the accuracy of TIPD on the ESC
data set, and the S_HVGs+TIPD method only improves the accuracy of TIPD on the
ESC and cDC data sets. Comparing the experimental results of Monocle2 and iter
+Monocle2, it can be seen that on the AT2, ESC, and cDC data sets, Monocle2 using
the iterTIPD feature selection method improved the accuracy of Monocle2 by 0.031,
0.054, and 0.013, respectively. On the HSMM data set, it was only the accuracy
difference is 0.018, which shows that the feature selection method of iterTIPD has
good generalization performance. The iterTIPD feature selection method improves
the accuracy of TIPD and Monocle2 on the ESC data set by 0.089 (the gap between
TIPD and iterTIPD) and 0.054 (the gap between Monocle2 and iter+Monocle2),
respectively, which is larger than that on the other three data sets, which also shows
that on large sample size data sets, excellent feature selection methods can

Table 21.1 Accuracy com- Method AT2 HSMM ESC cDC

parison using different feature
iterTIPD 0.852 0.602 0.658 0.741
selection methods
TIPD 0.821 0.533 0.570 0.689
M_HVGs_TIPD 0.415 0.478 0.579 0.667
S_HVGs_TIPD 0.698 0.516 0.688 0.735
Monocle2 0.718 0.561 0.419 0.585
Iter+Monocle2 0.752 0.540 0.481 0.595
21 Enhancing Single-Cell Trajectory Inference and Microbial Data Intelligence 353

effectively reduce the noise in the data set and improve the accuracy of the algo-
rithm. Overall, iterTIPD is better than the other two feature selection methods.

21.3.4.2 Robustness Analysis

Theoretically, iterTIPD attains maximum accuracy by eliminating the genes that are
deemed irrelevant by progressively decreasing the feature subset throughout the
course of iterations. Thus, whether the iterTIPD approach can reliably “target” a
subset of genes that yield valuable information more than once indicates how robust
the strategy is. In order to determine whether iterTIPD’s iterative feature selection
may increase TIPD’s robustness and whether it is superior to the other two feature
selection approaches, this chapter evaluates the robustness of the aforementioned six
methods.
All perturbation data sets combined, iterTIPD’s average robustness score is
higher than TIPD’s, and its score fluctuation range is more constrained. All pertur-
bation data sets show consistent robustness scores for the iter+Monocle2 technique.
It has been discovered that iterative feature selection techniques improve the algo-
rithm’s robustness, as shown by its better performance when compared to Monocle2.
However, the feature selection strategies used by Monocle and Seurat do not always
improve the robustness of the algorithm. The robustness score of the M_HVGs
+TIPD technique is generally lower than that of TIPD. However, the M_HVGs
+TIPD technique shows a slightly higher robustness score than TIPD on the
perturbation data set that covers 80% of the sample size on the cDC data set.
The S_HVGs+TIPD method demonstrates superior robustness compared to TIPD
in certain instances, specifically when applied to perturbed data sets with 90% and
70% sample sizes on the ESC data set, as well as perturbed data with a 70% sample
size on the cDC data set. However, in other cases, the robustness of the S_HVGs
+TIPD method does not significantly differ from that of TIPD. Hence, the experi-
mental results demonstrate that the iterative feature selection approach employed by
iterTIPD significantly enhances the algorithm’s resilience, surpassing the feature
selection strategies employed by Monocle and Seurat.

21.3.4.3 Analysis of the Ability to Detect Gold Standard Genes

This chapter also conducted comparative experiments to test whether each feature
selection method can improve the ability to detect gold standard genes and used the
average ranking of gold standard genes as a measurement indicator. The values in
Table 21.2 represent the average ranking of gold standard genes in gene rankings
obtained by different algorithms.
Table 21.2 shows that the average ranking of gold standard genes on iterTIPD is
lower than that of TIPD. The better an algorithm is at finding gold standard genes,
the lower the average ranking of gold standard genes. The iter+Monocle2 method
also does better in AT2, HSMM, and ESC. It’s also lower than Monocle2 on the data
354 B. Posinasetty et al.

Table 21.2 Average ranking Method AT2 HSMM ESC cDC

of gold standard genes using
iterTIPD 2151 3342 6044 1592
different feature selection
methods TIPD 2480 3929 6464 2049
M_HVGs_TIPD 3077 3542 7939 1726
S_HVGs_TIPD 2579 3618 7981 1463
Monocle2 3639 4275 8039 3312
Iter+Monocle2 3105 3983 7220 3731

set, which shows that iterTIPD’s method for selecting features over and over again
can help the programme find gold standard genes better. The iterTIPD gold standard
gene ranks lower than M_HVGs+TIPD and iter+Monocle2 in four sets of data. It
also ranks lower than S_HVGs+TIPD in AT2, HSMM, and ESC data sets, with
differences of 438, 270, and 1934 points, respectively. On the cDC data set, it ranks a
little higher than S_HVGs+TIPD (the difference is only 138). Also, on the HSMM
data set and the cDC data set, the average score of gold standard genes for the
M_HVGs+TIPD and S_HVGs+TIPD methods is only lower than TIPD. This means
that these two feature selection methods can’t really make TIPD better at finding
gold standard genes. In general, iterTIPD’s feature selection method makes TIPD
and Monocle2 even better at finding gold standard genes; it is also better than
Monocle and Seurat’s feature selection methods.

21.3.5 Analysis of Results

In this chapter, an optimization method iterTIPD using iterative feature selection is

proposed based on the TIPD algorithm. In order to explore the generalization ability
of iterTIPD’s feature selection method, this chapter applies it to the gene selection
step of Monocle2 and compares it with Monocle2. It also chooses two other classic
feature selection methods to combine with TIPD. In terms of accuracy and robust-
ness, comparative experiments were conducted on three indicators including the
ability to detect gold standard genes. Experimental results show that iterTIPD’s
iterative feature selection can improve the accuracy and robustness of the algorithm
and its ability to detect gold standard genes. Based on the experimental results on
four single-cell data sets, although M_HVGs and S_HVGs use different methods to
play the role of feature selection, due to the separation of the cell differentiation
trajectory inference algorithm, the effect is usually not optimal. In contrast, iterTIPD
combines the feature selection task with the cell differentiation trajectory inference
task and performs iteratively. In each iteration, the cell differentiation trajectory
inference algorithm is used as the evaluator to screen genes so that the selected
feature subset can better meet the needs of cell differentiation trajectory inference,
and the feature subset is continuously reduced in multiple iterations so that the
inferred cell differentiation trajectories are obtained with optimal accuracy.
21 Enhancing Single-Cell Trajectory Inference and Microbial Data Intelligence 355

In experiments comparing robustness, iterTIPD maintained the highest robust-

ness scores on the four single-cell data sets. This is because iterTIPD continuously
narrows down the gene subset and continuously sorts the cell pseudo-time sorting
results during the iterative process. Corrections will eventually “lock” on a set of
genes that provide useful information to achieve the highest accuracy. The TIPD
algorithm has randomness in the dimensionality reduction step, which usually leads
to unstable results. iterTIPD’s iterative feature selection method can improve the
robustness of the algorithm to a certain extent.
However, iterTIPD still has value worthy of further exploration, such as the
following:
(1) In the dimensionality reduction step, the values of the retained dimensions
have a greater impact on the algorithm results. You can consider flexibly trying other
dimensionality reduction algorithms based on the characteristics of the data set;
(2) in this chapter, iterTIPD uses 4 public dataset. It has been verified on single-cell
data sets, and its generalization ability can be tested on larger-scale data sets in the
future; (3) iterTIPD’s iterative feature selection method can also be used on other
machine learning models by optimizing feature subsets to improve the accuracy of
the algorithm.

21.4 Application of iterTIPD in Neural Stem Cell

Differentiation

21.4.1 Adult Stem Cell Lineage

The processes of stem cell quiescence and activation are very important for keeping
many body systems and tissues healthy, renewing them, and making them adaptable.
They are also very important for how quickly we age and get diseases. It is possible
for quiescent stem cells to successfully use both external and internal signals to
either stay in a dormant state or divide and differentiate to make new cells when they
are activated (Flexman et al. 2006). It’s important that neural stem cells (NSCs) are
present in the adult brain because they are a key source of regenerative cells that can
help treat neurodegenerative diseases and damage to neurons. Neurological stem cell
(NSC) development processes vary greatly. This is very important to look into if we
want to fully understand the functional traits and gene regulatory networks
connected to NSC populations. On the other hand, using population-based methods
might hide the natural variety that exists within neural stem cell (NSC) lineages. As a
result, this could make it harder to find new, unusual cell types or intermediate states
and also harder to fully understand how complex transcriptional processes work. The
iterTIPD algorithm is used in this study to find adult neural stem cell lineages and
look into how NSC populations vary and how genes are expressed in single cells.
Most neural stem cells (NSCs) are in a state called “quiescence,” which means
they are not actively dividing. This type of NSCs, which are also called quiescent
356 B. Posinasetty et al.

Quiescent Neural Neurons

Stem Cells (qNSC)

Astrocytes

Nerve cells
(NPC) Oligodendrocytes

Activated Neural Stem Cells

(NSC)

Fig. 21.4 Schematic diagram of neural stem cell differentiation lineage

neural stem cells (qNSCs) and don’t divide, can change into activated neural stem
cells (aNSCs), which do divide (Mansoor et al. 2015). Neuronal stem cells (aNSCs)
in adults can make neural progenitor cells (NPCs). NPCs are a group of cells that
divide quickly and show traits of early brain development. Finally, as shown in
Fig. 21.4, the neural progenitor cells (NPCs) change into specific cell types such as
neurons, astrocytes, and oligodendrocytes. There is still a lot we don’t know about
the different types of neural stem cells and how their genes change over time, even
though new single-cell correlation studies have shed light on the complex makeup of
neural stem cell (NSC) populations in certain neurogenic areas of the adult brain
(Mikhailov and Sankai 2018; Flexman et al. 2007).

21.4.2 Data Sets and Preprocessing

The neural stem cell differentiation data set used in this chapter is PRJNA324289
(Mikhailov and Sankai 2018), which is in the standardized FPKM format and
contains 79 quiescent neural stem cells (qNSC), 152 activated neural stem cells
(aNSC), and 64 neural progenitor cells (NPC) and 34 astrocytes.
The preprocessing step refers to the processing method of this chapter (Chang
et al. 2017): for cell filtration, contaminated oligodendrocytes and outlier cells
(observed in two-dimensional space using PCA (principal component analysis))
are removed. For gene filtering, retain genes expressed more than ten times in at
least five samples. Since this chapter hopes to explore the lineage relationship
between qNSCs, aNSCs, and NPCs, astrocytes were removed, leaving 250 samples
for subsequent analysis.
21 Enhancing Single-Cell Trajectory Inference and Microbial Data Intelligence 357

21.4.3 Neural Stem Cell Differentiation Trajectory

Constructed by iterTIPD Analysis

21.4.3.1 iterTIPD Reconstructs the Differentiation Trajectory of Neural

Stem Cells

The neural stem cell differentiation data set that had already been cleaned up was fed
into the iterTIPD method to figure out the paths of differentiation. iterTIPS predic-
tion of the path of neural stem cell development fits with what other studies have
found (Mansoor et al. 2015). First, iterTIPD finds cell states that are not the same.
Then, the minimum spanning tree is found by using the JS divergence distance to
find the distance between the probability distributions of these different states. Then,
iterTIPD picks the qNSC population with the highest average entropy value as the
starting point. This lets us figure out that the qNSC-aNSC-NPC differentiate in a
straight line. Finally, the simultaneous master curve method (Cawley et al. 2006)
was used to make all the cells fit, and each cell’s pseudo-time value was found.

21.4.3.2 Analysis to Identify Intermediate States of aNSC Populations

This chapter selected relevant marker genes for three different populations of qNSC,
aNSC, and NPC to explore their expression patterns in pseudo-chronological order.
Id3 is a key gene related to the quiescent state of neural stem cell differentiation,
which was previously reported as a marker of qNSC population (Nguyen et al.
2022); Egfr is related to the activation process; Cdk4 and Cdk1 are related to the
cell cycle; Dlx2 and Dcx are related to the differentiation process of neurons, among
which Dlx2 is a proneural transcription factor known to promote neural differenti-
ation (Demir and Eren 2022).
The iterTIPD approach calculates the expression of the above marker genes on
the pseudo-time sequence. In order to clearly demonstrate the continuous dynamic
changes in marker gene expression, the sequential number of the cell pseudo-time is
used instead of the pseudo-time value. Id3 is first highly expressed in the qNSC
population, and then the NSC differentiation process is activated and Egfr is
upregulated. Cell cycle-related genes Cdk4 and Cdk1 are highly expressed in
aNSC populations. Dlx2 is highly expressed in a subset of aNSC populations,
suggesting that a subpopulation of aNSCs may exhibit early transcriptomic signa-
tures of neural stem cell differentiation. The NPC population ranks last in pseudo-
temporal order and highly expresses the important neurogenesis regulator Dcx. The
expression pattern of the above marker genes is consistent with existing research
(Jiang et al. 2014), which also proves that iterTIPD correctly infers the pseudo-time
sequence of single cells.
The dynamic expression of these marker genes along pseudo-time sequence
shows successively different differentiation states. This chapter refers to the
pseudo-time stage (red) that contains the majority of the qNSC population as the
358 B. Posinasetty et al.

qNSC-like population and the pseudo-time stage (blue) that contains the majority of
the NPC population as the NPC-like population. In aNSC-like populations (green),
by observing the specific expression patterns of marker genes such as Egfr, Cdk1,
Dlx2, and Dcx, this chapter divides aNSC-like populations into three continuous
intermediate states and names them aNSC early stage, aNSC mid-stage, and aNSC
late stage. Among them, Egfr is low expressed in the qNSC-like population; aNSC
has high expression of Egfr and low expression of Cdk1 in the early stage; aNSC
has high expression of Cdk1 and low expression of Dlx2 in the middle stage; aNSC
has high expression of Cdk1 and Dlx2 in the late stage; and the NPC-like population
has high expression of Egfr and Dcx. The expression of genes Cdk1 and Dlx2 in the
aNSC population shows different changes, so it is speculated that there may be an
intermediate state in the aNSC population.
In order to verify the above conjecture, this chapter then calculated the average
expression levels of the specifically expressed marker genes Cdk1 and Dlx2 in the
early, middle, and late stages of aNSC to observe their trends and patterns. The
results are shown in Fig. 21.5a, b. There was a large difference in levels, and there
was also a significant difference in the expression of Dlx2 on aNSC mid-stage and
aNSC late stage. The specific expression of Cdk1 and Dlx2 further suggests that
novel subpopulations may exist within the aNSC population.
In summary, iterTIPD correctly inferred the pseudo-temporal sequence of neural
stem cells, which revealed the dynamic characteristics of gene expression. Based on
this sequence, we observed the specifically expressed genes Cdk1 and Dlx2 in the
aNSC population and calculated that these two genes were differentially expressed in
the aNSC population, suggesting that new intermediate states of activation and
differentiation may exist in the aNSC population.

21.4.3.3 Analyse and Discover Markers That Define the Intermediate

State of aNSC

The dynamic expression of marker genes reveals the possible existence of new
subpopulations in aNSCs. Based on the specific expression of Cdk1 and Dlx2 marker
genes, this chapter divides the aNSC-like population into three different continuous
intermediate states: early aNSC, mid-stage aNSC, and late-stage aNSC. In order to
find new markers for aNSC populations, this chapter calculated the differentially
expressed genes of NSC populations in pseudo-time order and used the Pearson
correlation coefficient (PCC) to calculate the similarity between Cdk1 and Dlx2 and
other genes respectively. aNSC subpopulation markers were found based on the
following criteria: genes with the highest similarity and the smallest corrected p-
value in differential expression analysis. Two genes were finally identified: one was
Top2a, which was most similar to Cdk1 and had the smallest corrected p-value; the
other was Gja1, which was most similar to Dlx2 and had the smallest corrected p-
value.
The expression of Top2a and Gja1 in three different states of aNSC population is
shown in Fig. 21.6a, b. Top2a is differentially expressed in early aNSC and
21 Enhancing Single-Cell Trajectory Inference and Microbial Data Intelligence 359

Fig. 21.5 (a) Average expression level distribution of Cdk1 aNSC population status. (b) Average
expression level distribution of Dlx2 aNSC population status

mid-stage aNSC, and Gja1 is differentially expressed in mid-stage aNSC and late
aNSC, and both show fold changes. Top2a is known to be extremely important for
the process of cell division and proliferation, and a recent study showed that Top2a
may be a new characteristic molecule expressed by neural stem cells (Mansoor et al.
2015). Gja1 is known to be a marker of astrocytes (Mikhailov and Sankai 2018) and
is also associated with the process of neural differentiation. Therefore, Top2a and
Gja1 are likely to be new markers for aNSC subpopulations.
360 B. Posinasetty et al.

Fig. 21.6 (a) Average expression level distribution of Top2a aNSC population status. (b) Average
expression level distribution of Gja1 aNSC population status

21.5 Conclusion

The majority of genes identified in single-cell RNA sequencing (scRNA-seq) data

are not pertinent to investigating the fundamental biological mechanisms. However,
by employing feature selection techniques to decrease the dimensionality of genes,
21 Enhancing Single-Cell Trajectory Inference and Microbial Data Intelligence 361

the signal-to-noise ratio and computing efficiency of subsequent analyses can be

enhanced. This work presents an algorithm called iterTIPD, which is an iterative
feature selection optimization algorithm derived from the TIPD algorithm. The novel
aspect of iterTIPD consists in its iterative application of a commonly employed
feature selection method for bulk RNA-seq data to scRNA-seq data. This approach
enhances the accuracy of cell pseudogene identification by effectively filtering out a
subset of genes that have the greatest impact on the generated differentiation
trajectory. This study performed experiments on four distinct single-cell data sets,
where the parameters of iteration were defined. Subsequently, the performance of
iterTIPD was compared to two other feature selection approaches, both in combi-
nation with TIPD, with respect to accuracy, robustness, and the capacity to identify
gold standard genes. A comparative experiment was conducted. This study aims to
assess the generalization capability of iterTIPD’s feature selection approach by
implementing it in the gene selection phase of Monocle2 and conducting a compar-
ative analysis with Monocle2. The experimental findings demonstrate that the
iterative feature selection strategy employed by iterTIPD yields enhanced perfor-
mance for TIPD, exhibits superior generalization capabilities, and outperforms two
other conventional feature selection methods.
The present study subsequently employs the iterTIPD algorithm to generate the
differentiation pathway of brain stem cells. Through the examination of marker gene
expression patterns in a pseudo-chronological sequence, this study has identified the
presence of three distinct intermediate states within the population of activated
neural stem cells. This finding suggests the existence of additional subpopulations
characterized by different activation and differentiation states within the larger
population of activated neural stem cells. This study employed a combination of
gene correlation analysis and differential expression analysis to identify the differ-
entially expressed genes Top2a and Gja1 in three distinct states of activated neural
stem cell populations. Based on a comprehensive review of relevant literature, it has
been shown that Top2a exhibits potential as a novel molecular marker expressed by
neural stem cells. Conversely, Gja1 has been established as a recognized marker of
astrocytes and is additionally implicated in the neural differentiation process. The
aforementioned finding has the potential to offer novel insights into the examination
of specific subsets of activated neural stem cell populations. Despite the algorithm in
this study demonstrating favourable outcomes in accuracy, robustness, and its
capacity to identify gold standard genes, there remain certain constraints that
necessitate additional investigation.

References

Afshar S, Salimi E, Braasch K, Butler M, Thomson DJ, Bridges GE (2016) Multi-frequency DEP
cytometer employing a microwave sensor for dielectric analysis of single cells. IEEE Trans
Microw Theory Tech 64(3):991–998. https://doi.org/10.1109/TMTT.2016.2518178
362 B. Posinasetty et al.

Cawley GC, Talbot NLC, Janacek GJ, Peck MW (2006) Sparse bayesian kernel survival analysis
for modeling the growth domain of microbial pathogens. IEEE Trans Neural Netw 17(2):
471–481. https://doi.org/10.1109/TNN.2005.863452
Y. -H. Chang et al., Human induced pluripotent stem cell region recognition in microscopy images
using Convolutional Neural Networks, 2017 39th Annual International Conference of the IEEE
Engineering in Medicine and Biology Society (EMBC), Jeju, Korea (South), 2017,
pp. 4058–4061, doi: https://doi.org/10.1109/EMBC.2017.8037747.
Y. -H. Chang et al., Human induced pluripotent stem cell reprogramming prediction in microscopy
images using LSTM based RNN, 2019 41st Annual International Conference of the IEEE
Engineering in Medicine and Biology Society (EMBC), Berlin, Germany, 2019,
pp. 2416–2419, doi: https://doi.org/10.1109/EMBC.2019.8857568.
Chen Z et al (2020) A branch point on differentiation trajectory is the bifurcating event revealed by
dynamical network biomarker analysis of single-cell data. IEEE/ACM Trans Comput Biol
Bioinform 17(2):366–375. https://doi.org/10.1109/TCBB.2018.2847690
Dai C et al (2018) Automated non-invasive measurement of single Sperm’s motility and morphol-
ogy. IEEE Trans Med Imaging 37(10):2257–2265. https://doi.org/10.1109/TMI.2018.2840827
Deconinck L, Cannoodt R, Saelens W, Deplancke B, Saeys Y (2021) Recent advances in trajectory
inference from single-cell omics data. Curr Opin Syst Biol 27:100344. ISSN 2452-3100
Demir MH, Eren B (2022) Output voltage control of double chambers microbial fuel cell using
intelligence-based optimized adaptive neuro fuzzy inference controller. Int J Hydrog Energy
47(45):19837–19849. ISSN 0360-3199.
J. A. Flexman, S. Minoshima, Y. Kim and D. J. Cross, Magneto-Optical Labeling of Fetal Neural
Stem Cells for in vivo MRI Tracking, 2006 International Conference of the IEEE Engineering in
Medicine and Biology Society, New York, NY, USA, 2006, pp. 5631–5634, doi: https://doi.org/
10.1109/IEMBS.2006.259982.
J. A. Flexman, D. J. Cross, Y. Kim and S. Minoshima, Morphological and parametric estimation of
fetal neural stem cell migratory capacity in the rat brain. 2007 29th Annual International
Conference of the IEEE Engineering in Medicine and Biology Society, Lyon, France, 2007,
pp. 4464–4467, doi: https://doi.org/10.1109/IEMBS.2007.4353330.
Gan Y, Li N, Guo C, Zou G, Guan J, Zhou S (2022) TiC2D: trajectory inference from single-cell
RNA-Seq data using consensus clustering. IEEE/ACM Trans Comput Biol Bioinform 19(4):
2512–2522. https://doi.org/10.1109/TCBB.2021.3061720
Geard N, Wiles J (2005) A gene network model for developing cell lineages. Artif Life 11(3):
249–267. https://doi.org/10.1162/1064546054407202
Hu Q, Merchante C, Stepanova AN, Alonso JM, Heber S (2016) Genome-wide search for translated
upstream open Reading frames in Arabidopsis Thaliana. IEEE Trans Nanobioscience 15(2):
148–157. https://doi.org/10.1109/TNB.2016.2516950
Iriya R et al (2020) Rapid antibiotic susceptibility testing based on bacterial motion patterns with
long short-term memory neural networks. IEEE Sensors J 20(9):4940–4950. https://doi.org/10.
1109/JSEN.2020.2967058
Jiang X, Xu W, Park EK, Li G (2014) Selecting protein families for environmental features based
on manifold regularization. IEEE Trans Nanobioscience 13(2):104–108. https://doi.org/10.
1109/TNB.2014.2316744
Karbalayghareh A, Braga-Neto U, Dougherty ER (2019) Classification of single-cell gene expres-
sion trajectories from incomplete and Noisy data. IEEE/ACM Trans Comput Biol Bioinform
16(1):193–207. https://doi.org/10.1109/TCBB.2017.2763946
Lee J, Friderikos V (2022) Interference-aware path planning optimization for multiple UAVs in
beyond 5G networks. J Commun Netw 24(2):125–138. https://doi.org/10.23919/JCN.2022.
000006
Lee KS, Kim TJ, Pratx G (2015) Single-cell tracking with PET using a novel trajectory reconstruc-
tion algorithm. IEEE Trans Med Imaging 34(4):994–1003. https://doi.org/10.1109/TMI.2014.
2373351
21 Enhancing Single-Cell Trajectory Inference and Microbial Data Intelligence 363

Li M et al (2021) A deep learning-based method for identification of bacteriophage-host interaction.

IEEE/ACM Trans Comput Biol Bioinform 18(5):1801–1810. https://doi.org/10.1109/TCBB.
2020.3017386
Li X et al (2023) A clustering method unifying cell-type recognition and subtype identification for
tumor heterogeneity analysis. IEEE/ACM Trans Comput Biol Bioinform 20(2):822–832.
https://doi.org/10.1109/TCBB.2022.3203185
Liu R, Pisco AO, Braun E, Linnarsson S, Zou J (2022) Dynamical systems model of RNA velocity
improves inference of single-cell trajectory, pseudo-time and gene regulation. J Mol Biol
434(15):167606. ISSN 0022-2836
Lu F, Lin Y, Yuan C, Zhang X-F, Ou-Yang L (2021) EnTSSR: a weighted ensemble learning
method to impute single-cell RNA sequencing data. IEEE/ACM Trans Comput Biol Bioinform
18(6):2781–2787. https://doi.org/10.1109/TCBB.2021.3110850
Ma F, Lian L, Ji P, Yin Y, Chen W (2020) Fault diagnosis scheme based on microbial fuel cell
model. IEEE Access 8:224306–224317. https://doi.org/10.1109/ACCESS.2020.3044354
Mansoor A, Patsekin V, Scherl D, Robinson JP, Rajwa B (2015) A statistical modeling approach to
computer-aided quantification of dental biofilm. IEEE J Biomed Health Inform 19(1):358–366.
https://doi.org/10.1109/JBHI.2014.2310204
A. Mikhailov and Y. Sankai, Donation of neural stem cells? Post mortal viability of spinal cord
neuronal cells. 2018 40th Annual International Conference of the IEEE Engineering in Medicine
and Biology Society (EMBC), Honolulu, HI, USA, 2018, pp. 5333–5337, doi: https://doi.org/
10.1109/EMBC.2018.8513487.
Muller G, Piel H, Roth RW, Aune B, Magne C, Veyssiere A (1989) Field-emission loading of
superconducting accelerator cavities at L- and S-band frequencies. IEEE Trans Electr Insul
24(6):1013–1017. https://doi.org/10.1109/14.46329
Nguyen VT, Ta QTH, Nguyen PKT (2022) Artificial intelligence-based modeling and optimization
of microbial electrolysis cell-assisted anaerobic digestion fed with alkaline-pretreated waste-
activated sludge. Biochem Eng J 187:108670. ISSN 1369-703X
Ni X, Geng B, Zheng H, Shi J, Hu G, Gao J (2022) Accurate estimation of single-cell differentiation
potency based on network topology and gene ontology information. IEEE/ACM Trans Comput
Biol Bioinform 19(6):3255–3262. https://doi.org/10.1109/TCBB.2021.3112951
Rezk H, Olabi AG, Abdelkareem MA, Sayed ET (2023) Artificial intelligence as a novel tool for
enhancing the performance of urine fed microbial fuel cell as an emerging approach for
simultaneous power generation and wastewater treatment. J Taiwan Inst Chem Eng 148:
104726. ISSN 1876-1070.
Sahayasheela VJ, Lankadasari MB, Dan VM, Dastager SG, Pandian GN, Sugiyama H (2022)
Artificial intelligence in microbial natural product drug discovery: current and emerging role.
Nat Prod Rep 39(12):2215–2230. ISSN 0265-0568.
Schiebinger G (2021) Reconstructing developmental landscapes and trajectories from single-cell
data. Curr Opin Syst Biol 27:100351. ISSN 2452-3100
Seabolt EE et al (2022) Functional genomics platform, a cloud-based platform for studying
microbial life at scale. IEEE/ACM Trans Comput Biol Bioinform 19(2):940–952. https://doi.
org/10.1109/TCBB.2020.3021231
Wang H-Y, Zhao J-P, Su Y-S, Zheng C-H (2022) scCDG: a method based on DAE and GCN for
scRNA-Seq data analysis. IEEE/ACM Trans Comput Biol Bioinform 19(6):3685–3694. https://
doi.org/10.1109/TCBB.2021.3126641
Wassan JT, Wang H, Browne F, Zheng H (2019) Phy-PMRFI: phylogeny-aware prediction of
metagenomic functions using random Forest feature importance. IEEE Trans Nanobioscience
18(3):273–282. https://doi.org/10.1109/TNB.2019.2912824
Wei J, Zhou T, Zhang X, Tian T (2019) SCOUT: a new algorithm for the inference of pseudo-time
trajectory using single-cell data. Comput Biol Chem 80:111–120. ISSN 1476-9271.
Wei J, Zhou T, Zhang X, Tian T (2021) DTFLOW: inference and visualization of single-cell
Pseudotime trajectory using diffusion propagation. Genomics Proteomics Bioinformatics 19(2):
306–318. ISSN 1672-0229.
364 B. Posinasetty et al.

Wu W, Ma X (2023) Network-based structural learning nonnegative matrix factorization algorithm

for clustering of scRNA-Seq data. IEEE/ACM Trans Comput Biol Bioinform 20(1):566–575.
https://doi.org/10.1109/TCBB.2022.3161131
Zeng Q, Ma X, Cheng B, Zhou E, Pang W (2020) GANs-based data augmentation for citrus disease
severity detection using deep learning. IEEE Access 8:172882–172891. https://doi.org/10.1109/
ACCESS.2020.3025196
Zeng S, Zhang H, Di B, Song L (2021) Trajectory optimization and resource allocation for OFDMA
UAV relay networks. IEEE Trans Wirel Commun 20(10):6634–6647. https://doi.org/10.1109/
TWC.2021.3075594
Zhang L, Zhang S (2020) Comparison of computational methods for imputing single-cell
RNA-sequencing data. IEEE/ACM Trans Comput Biol Bioinform 17(2):376–389. https://doi.
org/10.1109/TCBB.2018.2848633
Zhang Z, Zhang X (2021) Inference of high-resolution trajectories in single-cell RNA-seq data by
using RNA velocity. Cell Rep Methods 1(6):100095. ISSN 2667-2375
Zhang Y et al (2023) Automated dissection of intact single cell from tissue using robotic micro-
manipulation system. IEEE Robot Autom Lett 8(8):4705–4712. https://doi.org/10.1109/LRA.
2023.3287364
Chapter 22
AI-Assisted Methods for Protein Structure
Prediction and Analysis

Divya Goel , Ravi Kumar , and Sudhir Kumar

Abstract Proteins are the workhorses of cells. Their sequence is determined by the
genetic code embedded in the DNA, which translates it faithfully into a string of
amino acids known as the primary structure of proteins. But for proteins to achieve
functional mode, they must be correctly folded into a three-dimensional structure
commonly known as their tertiary structure. Determining the tertiary structure of the
proteins is often an expensive and time-consuming process. Protein structure predic-
tion has been in play for several decades now. But recent developments in the fields of
computational hardware, software, and artificial intelligence have led to the simulta-
neous development of methods capable of protein secondary and tertiary structure
prediction. Machine learning-based methods have recently emerged as aids of choice
for the prediction of protein structures. Programs like AlphaFold and AlphaFold2
have revolutionized the structure prediction landscape. This chapter presents a com-
prehensive account of the basics of AI-associated methods, along with the evolution
of protein structure prediction and its subsequent analysis, helping in related appli-
cations in diverse fields ranging from drug discovery to enzyme design.

Keywords Artificial Intelligence · Machine Learning · Deep Learning · Protein

Structure Prediction · AlphaFold

D. Goel
Department of Biotechnology, H.N.B. Garhwal University, Srinagar Garhwal, Uttarakhand,
India
R. Kumar
Department of Computer Science Engineering, Lovely Professional University, Jalandhar,
Punjab, India
Department of Computer Science Engineering, Jawaharlal Nehru Government Engineering
College, Sundernagar, Himachal Pradesh, India
S. Kumar (✉)
Department of Biotechnology, H.N.B. Garhwal University, Srinagar Garhwal, Uttarakhand,
India
Special Centre for Molecular Medicine, Jawaharlal Nehru University, New Delhi, India

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024 365
A. Khamparia et al. (eds.), Microbial Data Intelligence and Computational
Techniques for Sustainable Computing, Microorganisms for Sustainability 47,
https://doi.org/10.1007/978-981-99-9621-6_22
366 D. Goel et al.

22.1 Introduction

22.1.1 Overview of Protein Structure Prediction and Analysis

Proteins are fundamental macromolecules that are pivotal in all biological processes.
Elucidation of their three-dimensional (3D) structure is essential for comprehending
their functions and interactions. However, experimental determination of protein
structures, while more accurate, is often expensive and time consuming. As a result,
protein structure prediction and analysis via computational methods have emerged as
invaluable tools in modern biology. The amino acid sequence and protein structure
have always been connected, and they need to be analyzed critically, which would
enable the prediction of function from genome sequence data and facilitate the
intentional modification of annotated protein functions by crafting amino acid
sequences with targeted structures.
Protein structure prediction helps to deduce the three-dimensional protein struc-
ture solely from the sequence of amino acids, without experimental data. This field
encompasses two main approaches: template-based methods and ab initio methods.
Template-based methods rely on known protein structures with sequence simi-
larity to the target protein. They use these templates to generate a model for the target
protein. On the other hand, ab initio methods predict protein structures from scratch
without relying on known templates. They utilize principles from physics and
optimization algorithms to explore the conformational space of protein folding.
The last 2 years have been phenomenal for the development and advancement of
AI and AI-assisted techniques, especially in protein structure determination.

22.1.2 Role of AI in Advancing Protein Structure Research

Artificial intelligence (AI) has revolutionized various scientific domains, and protein
structure research is no exception (Paul et al. 2021). AI-driven approaches have
significantly advanced our understanding of protein structures and their functions.
AI-driven approaches, such as homology modeling and ab initio methods, have
significantly improved protein structure prediction. Machine learning algorithms can
learn from large databases of experimentally determined protein structures to iden-
tify patterns and relationships between protein sequences and their corresponding
structures. These recognized patterns can then be applied to previse the structures of
novel proteins (AlQuraishi 2021). AI-driven classification methods, such as cluster-
ing algorithms and neural networks, have facilitated the categorization of proteins
into structural families based on their fold and function. These automated approaches
can handle large datasets and uncover new relationships between protein structures
(Vijayan et al. 2022). AI-based function prediction methods use deep learning
models that are trained on vast databases of studied protein functions to devise the
functions of novel proteins (Kandathil et al. 2023). AI has also been used to finetune
22 AI-Assisted Methods for Protein Structure Prediction and Analysis 367

experimentally determined protein structures. By combining experimental data with

AI-driven approaches, such as molecular dynamics (MD) simulations and neural
network-based models, researchers can obtain more accurate and reliable protein
structures (AlQuraishi 2019a). AI-driven protein structure research has opened new
opportunities in drug design and personalized medicine. Virtual screening of poten-
tial drug compounds against the predicted protein structures has become expeditious,
facilitating drug discovery. Moreover, AI algorithms can analyze protein structures
that are part of individual patients’ profile, which enables researchers to design
personalized treatments tailored to their unique genetic makeup (Johnson et al.
2021). This chapter will provide a background of protein structure fundamentals,
basics of machine learning, secondary structure prediction, and tertiary structure
prediction with a special mention of AlphaFold.

22.2 Fundamentals of Protein Structure

Proteins undergo a process of adopting stable three-dimensional configurations,

known as conformations, which are dictated by their specific arrangement of
amino acids. This process, known as protein folding, is central to the proper
functioning of proteins and is a strictly regulated process (Diaz-Villanueva et al.
2015; Englander et al. 2007). The entire architecture of a protein is generally
characterized at four separate levels of intricacy: primary, secondary, tertiary, and
quaternary structure (Sun et al. 2004).
At the core of protein structure analysis lies the primary structure, which points
toward the sequence of amino acids that constitute a protein. The sequence is
encoded by the genetic information in the DNA, and variations in the sequence
give rise to the vast diversity of proteins found in nature. The primary structure forms
the foundation for higher-order structures and is essential for determining a protein’s
functional properties (Anfinsen 1973).
The secondary structure is the next level of the structural hierarchy of a protein,
which arises from local interactions within the polypeptide chain. These interactions
lead to the formation of distinct repetitive patterns, primarily the alpha-helix, beta-
sheet, and turns. The alpha-helix resembles a right-handed coil, while the beta-sheet
adopts a pleated, sheet-like arrangement. Turns and loops serve as connectors,
linking helices and β-sheets together. Hydrogen bonding between the backbone
atoms stabilizes these secondary structure elements, influencing the protein’s overall
shape and properties (Sun et al. 2004).
Tertiary structure represents the three-dimensional arrangement of all secondary
structures in terms of their accurate spatial arrangement and location of functional
groups in a single polypeptide chain. This folding is guided by long-range interac-
tions between different amino acids that may not be adjacent in the linear sequence.
Various forces, such as hydrogen bonding, disulfide bridges, van der Waals inter-
actions, and hydrophobic effect, impart equilibrium and stability to the tertiary
structure. The compact, folded conformation is critical for the proper functioning
368 D. Goel et al.

of proteins, as it positions the active sites and functional regions in the correct
orientation (Novak 2021). The tertiary structure of the protein can be further
subdivided into domains, folds, and motifs. A domain is a structurally independent
compact folding unit, usually consisting of 50–300 amino acids that may be con-
served in nature (Sun et al. 2004). A specific arrangement of secondary structure
elements in space depicts a fold which is consistent with a number of different
proteins (Minami et al. 2014). Protein motifs are short, conserved sequences or
patterns of amino acids that are found in specific protein families or functional
domains (Parras-Moltó et al. 2013).
Proteins can also exhibit quaternary structure, which involves the precise assem-
bly of multiple polypeptide chains (subunits) to form a multimeric functional com-
plex of a protein. The subunits might be similar or distinct, and their interaction is
mediated by non-covalent bonds, like hydrogen bonding and hydrophobic interac-
tions (Fig. 22.1) (Alberts et al. 2022).
Any prediction of the protein structure must first accurately identify and predict
these structural features (Deng et al. 2018). Learning from the vast data available
from databases like PDB, NCBI, and UniProt, AI-assisted techniques have come a
long way from predicting the smaller secondary structural elements to modeling full-
length proteins with increasing accuracy (Pakhrin et al. 2021).

22.3 Machine Learning Basics for Protein Structure

Analysis

Prediction of protein structure has remained a fundamental challenge in biochemis-

try, particularly because there is a direct relation between protein sequence, its
structure, and function (AlQuraishi 2021). A branch of artificial intelligence, i.e.,
machine learning, has emanated as a formidable aid in protein structure research,
enabling researchers to make accurate predictions and gain deeper insights into
protein structure-function relationships.
Designing algorithms and drafting models for statistical analysis are important
components of machine learning that empower computers to recognize motifs and
project the protein structure from data even when it is not programmed emphatically.
Machine learning is divided into three main categories: supervised learning,
unsupervised learning, and reinforcement learning (RL).
22 AI-Assisted Methods for Protein Structure Prediction and Analysis 369

Fig. 22.1 Hierarchical representations of protein structure starting from primary structure (amino
acid sequence), secondary structure (α-helix, β-sheets), tertiary structure, and finally quaternary
structure. Protein structures are derived from PDB id 3P47 (Kumar et al. 2011), and the images are
made using the PyMOL program (Educational version) (Schrodinger, LLC 2015)

22.3.1 Supervised, Unsupervised, and Reinforcement

Learning

Supervised Learning
In supervised learning, the algorithm is trained on categorized data, along with the
provision of input and compatible output. The objective is to comprehend mapping
from corresponding inputs and outputs and estimate new and unseen data, as shown
in Fig. 22.2.
370 D. Goel et al.

Fig. 22.2 Supervised learning

Fig. 22.3 Unsupervised learning

Unsupervised Learning
In unsupervised learning, the algorithm is trained on unlabeled data, and the aim is to
identify data patterns or groupings without explicit guidance, as shown in Fig. 22.3.
Reinforcement Learning
Reinforcement learning involves training an agent to work in an environment which
reaps supreme results. The agent gathers information from the assessment that is
gained in the form of incentives or consequences, as shown in Fig. 22.4.

22.3.1.1 Supervised Learning Algorithms

Supervised learning algorithms constitute a cornerstone of machine learning,

leveraging labeled training data to infer correlations between input features and
corresponding output labels. When working with protein structure prediction,
22 AI-Assisted Methods for Protein Structure Prediction and Analysis 371

Fig. 22.4 Reinforcement learning

these algorithms learn from known protein structures to predict the structure of a
given amino acid sequence. Regression algorithms, such as linear regression and
support vector regression, attempt to predict continuous structural properties like the
distance between amino acids. Classification algorithms, like random forests and
support vector machines, can predict secondary structure elements (helices, strands,
coils) or solvent accessibility (Kumar et al. 2011).

Regression Algorithms and Distance Prediction

One of the prime objectives in protein structure prediction involves estimating inter-
atomic distances within a protein’s structure (Schrodinger, LLC 2015). Regression
algorithms, such as linear regression and support vector regression, are instrumental
in predicting these distances. By training on datasets containing known protein
structures and their corresponding inter-atomic distances, these algorithms learn
the intricate relationships between amino acid sequences and spatial arrangements
(Zhu et al. 2021). For instance, given a protein sequence, a trained regression model
can estimate the distances between key amino acid pairs, elucidating the protein’s
three-dimensional conformation.

Classification Algorithms and Secondary Structure Prediction

The secondary structure of a protein—comprising alpha-helices, beta strands, and

coils—plays a pivotal role in determining its function. Classification algorithms like
random forests and support vector machines excel at predicting these secondary
structure elements. Utilizing annotated datasets containing experimentally deter-
mined secondary structure assignments, these algorithms learn to discriminate
sequences based on the propensity of adopting particular secondary structures
(Pearce and Zhang 2021). Consequently, when provided with a protein sequence,
the model can accurately deduce the distribution of secondary structure elements,
facilitating insights into the protein’s folding patterns.
372 D. Goel et al.

Solvent Accessibility Prediction with Classification Algorithms

Understanding the solvent accessibility of amino acids within a protein aid in

deciphering their roles in binding sites and interactions (Kuhlman and Bradley
2019). Classification algorithms also come to the fore in predicting solvent accessi-
bility. Through exposure to datasets enriched with structural information and
corresponding solvent accessibility annotations, these algorithms acquire the capac-
ity to predict the extent to which specific amino acids are exposed to the solvent
environment. Thus, given a protein sequence, the model can anticipate regions with
varying degrees of solvent accessibility, contributing to a holistic understanding of
the protein’s structural landscape.

22.3.1.2 Unsupervised Learning Algorithms

Protein dynamics includes the variations in the three-dimensional configuration of a

protein necessary for it to execute its biological role. These variations occur in
response to external influences, such as temperature fluctuations and interactions
with other molecules, compelling the protein to assume multiple conformations
throughout its existence and enabling rapid transitions between these conformations
(Chen et al. 2009; Ahmad et al. 2003). Unsupervised learning algorithms are
employed to extract meaningful information from vast datasets of protein structures,
ultimately aiding in the understanding of protein folding, function, and interactions.
The primary technique used by unsupervised learning is clustering.

Clustering

Clustering is an unsupervised classification approach utilized to unveil the inherent

structure within data by uncovering concealed patterns. The primary aim of cluster-
ing techniques is to autonomously partition a dataset comprising instances into
clusters or groups characterized by their similarity while simultaneously identifying
clusters that exhibit dissimilarity (Chen et al. 2009). By identifying clusters of
similar protein structures, researchers can gain insights into protein families, evolu-
tionary relationships, and functional similarities (Teletin et al. 2018). Clustering can
further be of many types:
(a) K-Means Clustering: K-means is an algorithm that partitions protein structures
into a predetermined number of clusters. It is one of the most used clustering
algorithms. It employs a method of iteratively allotting each structure to the
closest cluster centroid and then upgrading the centroids established from the
average of the structures within each cluster (Tokuriki and Tawfik 2009).
(b) Hierarchical Clustering: Hierarchical clustering constructs a hierarchical tree-
like structure of clusters, thereby enabling researchers to investigate the
22 AI-Assisted Methods for Protein Structure Prediction and Analysis 373

inconsistencies of different stages in their data. The hierarchical tree can be

agglomerative (bottom-up) or divisive (top-down) (Zaslavsky et al. 2016).
(c) Density-Based Clustering: Density-based clustering methods, such as
DBSCAN, recognize clusters based on regions of higher data density. This
approach is particularly useful when clusters have irregular shapes and varying
densities (Melman and Roshan 2018).
(d) Dimensionality Reduction: Methods like Principal Component Analysis (PCA)
and t-Distributed Stochastic Neighbor Embedding (t-SNE) are utilized in con-
junction with clustering to reduce the dimensionality of protein structure data,
making it more manageable and informative (Jisna and Jayaraj 2021).
(e) Graph-Based Clustering: Graph-based methods represent protein structures as
nodes in a graph, with edges denoting structural similarity. Community detection
algorithms are then applied to identify clusters (Dang et al. 2021).

Principal Component Analysis

PCA is an elementary unsupervised machine learning technique widely applied for

protein structure prediction. It plays a pivotal role in alleviating the multiple dimen-
sionalities of complex protein structure data and retaining critical information at the
same time (Zhu et al. 2016). Its principal goal is to modify high-dimensional data
into a lower-dimensional representation, known as principal components, while
minimizing the loss of information (Russo and Borras 2022). In the context of
protein structure prediction, PCA operates by finding linear combinations of the
original variables (atom coordinates) that maximize variance, effectively capturing
the most significant structural variations in the data (Chen et al. 2009; Jolliffe and
Cadima 2016).

Application in Protein Structure Prediction

Dimensionality Reduction: Protein structures are inherently high-dimensional, with

each atom’s coordinates representing a separate dimension. PCA addresses this
challenge by projecting the information onto a reduced set of orthogonal axes, the
primary elements, which capture the essential structural features. This transforma-
tion significantly simplifies subsequent analysis and visualization.
Noise Reduction: Protein structure data may contain noise or redundant informa-
tion. PCA identifies and emphasizes the principal components that contribute most
significantly to the structural variance, effectively filtering out irrelevant details and
noise.
Visualization: PCA enables the visualization of protein structures in a lower-
dimensional space, facilitating it to discern patterns, similarities, and differences
among structures. Researchers can visualize and explore structural variations along
the principal components to gain insights into the protein’s conformational
flexibility.
374 D. Goel et al.

Clustering: PCA can be used in conjunction with clustering techniques to group

similar protein structures into clusters based on their principal component represen-
tations. This facilitates the identification of structural subgroups within a dataset.

22.3.1.3 Reinforced Learning

Reinforcement learning (RL) is a component of machine learning which is impli-

cated in sequential decision-making in environments where an agent (algorithm that
learns from trial and error) interacts with the surroundings. The agent is trained to act
to intensify a cumulative reward signal over time. In protein structure prediction, RL
is employed to optimize the folding by iteratively adjusting the conformation of the
protein until it reaches its most stable or biologically relevant state.
Unlike traditional machine learning approaches, in reinforcement learning,
desired outcome is not explicitly defined. Instead, features are entered into the
environment, and rewards are assigned. These rewards can be either positive or
negative, based on the agent’s actions and decisions (David and Jacobs 2014).

22.4 AI-Assisted Secondary Structure Prediction

22.4.1 Traditional Methods

Protein secondary structure prediction includes the accurate predictions of secondary

structures, including alpha-helices, beta-sheets, and coil regions. Traditional
methods for protein secondary structure prediction primarily rely on empirical
rules and physicochemical properties of amino acids. Methods such as Chou-
Fasman (Kitao 2022), Garnier-Osguthorpe-Robson (GOR) (Arora et al. 2018), and
Dictionary of Secondary Structure of Proteins (DSSP) (Ashok Kumar 2013) were
used over the years for secondary structure prediction. Chou-Fasman is an early
method based on the propensity of amino acids to form secondary structures. It
assigns secondary structure elements by analyzing the local sequence properties,
such as the presence of specific amino acid residues that favor the formation of
alpha-helices or beta-sheets. While simple, Chou-Fasman suffers from limited
accuracy due to its reliance on fixed parameters. GOR or Garnier-Osguthorpe-
Robson method utilizes statistical analysis of amino acid sequences to predict
secondary structure. It calculates probabilities for each amino acid to belong to
alpha-helices, beta-sheets, or coil regions based on databases of known protein
structures. GOR offers improved accuracy compared to Chou-Fasman but is still
hampered by its dependency on predefined datasets. (Dictionary of Secondary
Structure of Proteins) DSSP is a commonly employed method for secondary struc-
ture assignment in experimentally determined protein structures. It defines second-
ary structures by assigning helical or strand regions based on hydrogen bonding
22 AI-Assisted Methods for Protein Structure Prediction and Analysis 375

patterns and geometrical criteria. DSSP is highly accurate but is limited to known
protein structures.
While these methods were well established because of their simplicity and
transparency, needing only limited computational power, they suffered many limi-
tations, especially when dealing with novel protein sequences.
Combining the simplicity and interpretability of traditional methods with the
predictive power of AI can lead to more robust and accurate predictions. Addition-
ally, ongoing advancements in AI and deep learning algorithms, as well as the
availability of larger and more diverse protein databases, will continue to enhance
the accuracy and applicability of AI-based approaches.

22.4.2 Deep Learning Models for Secondary Structure

Prediction

Deep learning constitutes a specific domain within machine learning which revolves
around artificial neural networks. It places a strong emphasis on employing
interconnected layers to convert input data into features that are conducive to
predicting corresponding outputs. When provided with a dataset that is sufficiently
expansive, pairing inputs with their respective outputs, a training algorithm can be
applied to autonomously acquire the ability to map inputs to outputs (Kouza et al.
2017). This is achieved by adjusting a group of specifications at all the levels within
the neural network (Kabsch and Sander 1983; Taye 2023). Deep learning models,
particularly convolutional neural networks (CNNs) and recurrent neural networks
(RNNs), exhibited an exemplary success in protein secondary structure prediction.
These models have been specifically drafted to automatically decipher intricate
motifs and relationships between amino acid sequences, enabling accurate predic-
tions (Goodfellow et al. 2016).

22.4.2.1 Convolutional Neural Networks (CNNs)

CNNs are structural designs intended for the handling of data that exhibits consistent
spatial organization. A CNN layer uses a smart trick by using the same small filters
repeatedly across data, which is helpful in two ways: it prevents overfitting because
there are fewer things to adjust compared to the input layer, and it works well even if
the data shifts around. A CNN module typically has many of these layers stacked
together. As you move deeper into these layers, they can understand more compli-
cated features (Taye 2023).
In 2016, Wang and colleagues introduced a convolutional neural network (CNN)
approach for protein secondary structure prediction (PSSP), achieving an impressive
accuracy rate of up to 84% (Wang et al. 2016a, b). Their method, known as “Deep
Convolutional Neural Fields” (DeepCNF), comprised two key components: a
376 D. Goel et al.

conditional random fields (CRF) module, originally developed in their earlier work
from 2011, and a deep convolutional neural network (DCNN) module that handled
the input data leading up to the CRF, as described in their 2016 publication (Ismi
et al. 2022). Both one-dimensional CNN (CNN-1D) and two-dimensional CNN
(CNN-2D) were used previously in PSSP models. CNN-1D utilized sequences as
inputs and thus was used more often. Nevertheless, some models have incorporated
CNN-2D into their architectures to enhance the extraction of time- and space-related
features from input sequences. Specifically, feature vectors, such as position-specific
scoring matrix (PSSM) and one-hot encoding, representing a fixed-length residue
window, were assigned as input in studies for the two-dimensional CNN
(Goodfellow et al. 2016; Wang et al. 2016a, b).
CNNs bring two advantages in the prediction of secondary structures of proteins:
(a) Local Sequence Patterns: CNNs are well suited to recognize short-range inter-
actions and local sequence features crucial for secondary structure prediction.
They apply a set of convolutional filters covering all the different positions in the
amino acid sequence, enabling them to learn and extract relevant features
effectively (Guo et al. 2018).
(b) Translation Invariance: CNNs exhibit a remarkable property called translation
invariance. This means they can identify patterns even when the data is slightly
shifted or moved, making them suitable for the flexible and dynamic nature of
protein structures (Liu et al. 2017).

22.4.2.2 Recurrent Neural Networks (RNNs)

Recurrent neural networks (RNNs) design enables them to learn the global patterns
from sequential data. They provide a potent means for handling sequential data by
virtue of their intrinsic capability to “retain” past inputs within the sequence through
hidden states (Gelman et al. 2021). In the process of handling an input sequence, an
RNN module employs an internal state vector to condense the knowledge acquired
after refining the different components of the sequence. This module encompasses a
parameterized sub-component that receives both the previous internal state vector
and the present input element of the sequence as inputs, resulting in the generation of
the current internal state vector. Ultimately, this final state vector serves as a concise
representation, summarizing the entirety of the input sequence (Taye 2023).
One of the key advantages of RNNs in protein structure prediction is their ability
to handle variable-length input sequences. Proteins can vary significantly in length,
but RNNs can process sequences of different lengths without the need for fixed-size
inputs. This flexibility is essential for modeling the diversity of protein sequences
encountered in practice (Alzubaidi et al. 2021). Furthermore, RNNs can capture
long-range dependencies in protein sequences. Proteins often exhibit complex
folding patterns that result from interactions between distant amino acids in the
sequence. RNNs, with their recurrent connections, can capture these long-range
dependencies, allowing them to learn intricate structural motifs and patterns
(Gelman et al. 2021). A very simple RNN-based model known as LocalNet achieved
22 AI-Assisted Methods for Protein Structure Prediction and Analysis 377

an accuracy of 85.15% in predicting all three types of secondary structures (Helix,

sheets, turns) using only local amino acid information (Lee 2023).
In recent years, researchers have made significant strides in applying RNNs to
protein structure prediction. The development of specialized RNN variants, such as
long short-term memory (LSTM) and gated recurrent unit (GRU), has improved the
ability of RNNs to capture long-range dependencies (Taye 2023).
Long short-term memory (LSTM) networks are a specialized type of recurrent
neural network (RNN) architecture modulated to deal with some of the shortcomings
of traditional RNNs when it comes to detecting long-range dependencies in sequen-
tial data. LSTMs were introduced by Hochreiter and Schmidhuber in 1997 and have
since become a crucial component in various machine learning applications, includ-
ing natural language processing, time series analysis, and, as you mentioned, protein
structure prediction (Liu 2017). LSTMs are particularly advantageous in the context
of protein structural prediction because they can handle long sequences as they have
memory cells that can store and retrieve information over long sequences (Yang
et al. 2021). They incorporate gating mechanisms, like input gate, forget gate, and
output gate, which control the direction of information in and out of the memory
cells. It provides the network with the fine-grained control required for modeling
complex dependencies in protein sequences and structures (Hochreiter and
Schmidhuber 1997). An orchestrated model with dual loss function established for
the prediction of protein secondary structure consisting of five sub-models
connected via a Bi-LSTM layer achieved 84.3% in Q3 accuracy and 81.9% in
segment overlap measure (SOV) score (Yang et al. 2021). A deep bidirectional
long short-term memory (DBLSTM) was able to achieve reasonable level of accu-
racy compared to the state-of-the-art approaches like bidirectional long short-term
memory (BLSTM) and DeepCNF (Hu et al. 2019). Another study developed a
method called the optimized convolutional and long short-term memory neural
network model (OCLSTM) for predicting protein structure using an optimized
convolutional neural network to find patterns between nearby amino acids and a
bidirectional long short-term memory neural network to discover relationships
between distant amino acids. This model showed accuracy ranging from 82.36%
to 85.08% on different datasets (Ahmed et al. 2023).
Additionally, the integration of observation techniques has enhanced the model’s
ability to concentrate on relevant parts of the input sequence. Prediction of protein
structure with RNNs has also been achieved via transfer learning. Pretrained models
on large protein databases can be refined on particular assignments, thereby decreas-
ing the requirement of large quantity of task-specific data.
Gated recurrent unit (GRU) is another type of recurrent neural network (RNN)
architecture, related to long short-term memory (LSTM), modeled to explore the
diminishing gradient issue and capture long-range dependencies in sequential data
(Liu et al. 2017; Hochreiter and Schmidhuber 1997). Compared to LSTMs, GRUs
have fewer parameters because they lack separate memory cells. This makes them
computationally more efficient, particularly in scenarios with limited data and
computational resources. However, this simplicity can also make them less expres-
sive for assignments that are needed to decipher very complex long-term
378 D. Goel et al.

dependencies. A modified GRU-based method was recently utilized for the recog-
nition of phosphorylation sites in host cells infected with SARS-CoV-2 (Hattori et al.
2017).
Many researchers have used the hybrid RNN methods to achieve better accura-
cies on protein secondary structure predictions. Hybrid recurrent neural networks
(RNNs) direct toward a branch of neural network architectures that combine differ-
ent types of recurrent units or modules within a single network. Romana Rahman
Ema et al. used gated recurrent unit (GRU), long short-term memory (LSTM),
bidirectional gated recurrent unit (BGRU), and bidirectional long short-term mem-
ory (BLSTM) neural networks and improved the accuracies to a range of >90%
(Ema et al. 2022) (Zhao and Liu 2021). A webserver ET-GRU was also developed
using the 1D CNN, GRU, and PSSM profiles capable of identifying the electron
transport proteins among the protein sequence samples (Zhang et al. 2023).

22.4.3 Case Studies

DeepPrime2Sec, developed by Ehsaneddin Asgari et al. utilizing the CNN-BLSTM

(convolutional neural network-bidirectional long short-term memory) network and
combining top-k ensemble models, obtained accuracy rates of 69.9% and 70.4% for
predicting eight-class protein secondary structures on the CB513 dataset, which is
known to be one of the most demanding datasets for this type of prediction
(Ehsaneddin et al. 2019).
A reductive deep learning model called MLPRNN, composed of one bidirectional
gate current unit (BGRU) and two multi-layer perceptron (MLP) blocks, has been
introduced for predicting protein secondary structures, either in three-state or eight-
state classifications. The accuracy of MLPRNN’s predictions on the widely used
CB513 benchmark (70.07%) dataset is on par with the performance of other cutting-
edge models like DeepCNF (68–73%) (Le et al. 2019).
Sharma and Srivastava used a bidirectional long short-term memory (BLSTM)
model to project the secondary structures of proteins using the local contextual
information captured with character n-gram and achieved a Q3 accuracy of
86.69% on the CASP9 dataset (Ehsaneddin et al. 2019).
Chi-Hua Yu et al. also reported the development of a deep learning model which
used convolutional and recurrent architectures along with natural language models to
predict alpha-helix and beta-sheets in proteins. Their model could decipher the
obscured patterns in protein sequences and use this knowledge to predict the
secondary structures (Lyu et al. 2021).
SPIDER2 employs three rounds of deep learning neural network iterations to
enhance the efficiency of predictions for multiple structural characteristics, like
three-state secondary structure, angle-based representation of backbone structure,
and solvent accessible surface area, concurrently. When evaluated on an indepen-
dent test set comprising 1199 proteins, SPIDER2 attains the following performance
metrics: 82% accuracy for secondary structure prediction, a correlation coefficient of
22 AI-Assisted Methods for Protein Structure Prediction and Analysis 379

0.76 between predicted and actual solvent accessible surface area, mean absolute
errors of 19° and 30° for backbone φ and ψ angles, respectively, and mean absolute
errors of 8° and 32° for Cα-based θ and τ angles, respectively (Sharma and
Srivastava 2021).
Xin Jin et al. proposed CGAN-PSSP, a novel protein secondary structure predic-
tion (PSSP) model built upon a conditional generative adversarial network (CGAN),
which can be used for the prediction of eight-state (Q8) and three-state (Q3) protein
secondary structures. It was able to achieve 71.3% Q8 and 84.8% Q3 accuracy on
CASP11 dataset (Yu et al. 2022).

22.5 Predicting Tertiary Structure with AI

The tertiary structure of proteins represents the highest level of their structural
organization, defining the intricate 3D arrangement of atoms within a single poly-
peptide chain. This spatial arrangement is crucial in determining a protein’s function
and interactions with other molecules. Tertiary structure results from the complex
folding and bonding of secondary structures (alpha-helices, beta-sheets) and addi-
tional irregular segments. Forces responsible for maintaining the tertiary structure
include hydrogen bonds, disulfide bonds, hydrophobic interactions, and van der
Waals forces. The precise formation of unique 3D conformation of a protein is
essential for it to perform its functions inside a cell effectively, such as enzyme
catalysis, molecular recognition, and structural support within cells and organisms
(Alberts et al. 2022). Understanding and predicting protein tertiary structure are vital
endeavors in biochemistry and molecular biology, with implications ranging from
drug design to the elucidation of fundamental biological processes. Homology-based
methods have been a huge favorite among the researchers to project the protein
tertiary structure.

22.5.1 Homology Modeling and Comparative Protein

Structure Prediction

Homology modeling, or comparative modeling, is the method to construct a pro-

tein’s three-dimensional structure by utilizing structural insights derived from
known configurations of related proteins. This modeling strategy capitalizes on the
principle that a protein’s structural conformation tends to exhibit greater conserva-
tion across evolution compared to its specific sequence of amino acids. Conse-
quently, minor-to-moderate alterations in the amino acid sequence typically
correspond to minimal deviations in the 3D structure. A critical prerequisite for
the effective application of homology modeling is the presence of substantial
380 D. Goel et al.

similarity, typically exceeding 30%, between the protein sequences under

consideration.

22.5.2 Template-Based Structure Prediction

Homology modeling starts with identifying the suitable template and aligning the
query and target sequences. It then proceeds to the generation of the model including
backbone and side chains. Finally, the model is optimized and validated. The query
sequence is subjected to BLAST (Yang et al. 2017) against the PDB database. The
best structure with the lowest e value and maximum overall coverage is then selected
as the template structure. This structure then helps to determine the structure of the
query sequence, which is then refined individually for loops, side chains, and
secondary structures, giving an output structure. This method is very popular
because of its ease and accuracy, but the latter depends on the quality of the reference
structure. The extent to which the query sequence shows similarity or identity with
the reference sequence and its coverage is the critical step in this type of modeling
and is also its major limitation.
MODELLER is a popular program used for homology modeling (Jin et al. 2022).
A simple script is needed for modeller along with a sequence alignment file (Query
vs template) and the atomic coordinates of the templates (PDB file). MODELLER
then computes a model that includes all non-hydrogen atoms. It can also execute
various supplementary tasks, such as assigning folds, aligning two protein
sequences, or aligning their profiles (Altschul et al. 1990). There are several other
software or web-based programs like SWISS-Model (Webb and Sali 2016) and
I-TASSER (Eswar et al. 2006) that can be used for template-based homology
modeling.

22.5.3 Threading or Fold Recognition-Based Methods

Threading, also known as fold recognition, is a computational method used for

protein tertiary structure prediction when there are no close homologous proteins
with known structures available in the database. This approach assumes that even
distantly related proteins may share similar structural folds, and it seeks to identify
the best structural match or “thread” for a target protein sequence within the archive
of known protein folds. Although it follows similar steps like homology modeling,
fold identification and fold fitting steps are different. This method compares the
query sequence with a collection of structural templates containing the different
protein folds and allows the best fitting of the folds as judged by the scoring
functions. After generating the initial model, some threading methods may perform
model assessment and refinement steps to improve the quality of the predicted
structure. This can involve energy minimization, molecular dynamics simulations,
22 AI-Assisted Methods for Protein Structure Prediction and Analysis 381

or other optimization techniques. Examples include Phyre2 (Waterhouse et al.

2018), RaptorX (Yang and Zhang 2015), and HHpred (Kelley et al. 2015).
Threading methods have limitations when dealing with highly novel or extremely
divergent protein sequences, as the chances of finding a suitable template decrease.

22.5.4 Ab Initio Modeling

Ab initio protein modeling, also referred to as de novo modeling, is based on the

fundamental laws of thermodynamics and physics. By minimizing the free energy of
a protein’s conformation, it seeks to discover the 3D structure of the molecule. This
refinement can achieve its aim of the protein’s lowest-energy state or its most stable
configuration out of an infinite number of potential conformations.
Ab initio modeling often entails a conformational search procedure directed by a
carefully designed energy function. The final models are often selected from a
variety of structural decoys, or alternative conformations, that this process typically
generates (Kallberg et al. 2012). Energy functions used here are either physics based
or knowledge based, depending upon their working. The efficiency of ab initio
modeling thus depends on three important factors: the availability of a precise energy
function that can identify a most thermodynamically stable protein state, the appli-
cation of a quick search technique to identify low-energy conformational states, and
the quick identification of a near-native structural model (Zimmermann et al. 2018).
In physics-based techniques, parameters related to bond lengths, angles, torsional
angles, van der Waals interactions, and electrostatics are combined into a molecular
mechanics force field. Herein the protein is considered as a sophisticated atomic
model (Lee et al. 2009). Amber which started in 1970 is a suite of biomolecular
simulation programs using physics-based force fields and molecular dynamics
(MD) simulations to derive ab initio protein structure (Lee et al. 2017). Another
program, CHARMM (Chemistry at HARvard Macromolecular Mechanics), devel-
oped by Martin Karplus and his group, also uses empirical energy functions for
molecular simulations but also has the capacity for new model building (Wang et al.
2004; Case et al. 2005). The high cost of physics-based energy computations limited
their use. However, programs like UNRES and ASTRO-FOLD attempted to reduce
this problem. UNRES represented the polypeptide chain as a sequence of α-carbon
atoms (Cα) linked by virtual bonds of length 3.8 Å and the sidechains represented as
a single interaction site, effectively reducing the number of atoms under consider-
ation by a factor of 10 and consequently decreasing the time required for structure
prediction of proteins up to 100 residues long (Brooks et al. 2009). ASTRO-FOLD
calculates the free energy function from the oligopeptide overlapping and interaction
between hydrophobic residues to predict the secondary structures in the proteins.
This data is then transformed into a whole sequence, leading to the complete model
of the protein (Brooks et al. 1983). Other methods based on these energy functions
increased the length of the successfully predicted model to about 150 residues
(Zimmermann et al. 2018).
382 D. Goel et al.

22.5.5 Knowledge-Based Energy Function

The statistical analysis of recognized protein structures yields knowledge-based

energy functions, also known as statistical potential functions. These programs
estimate the likelihood or favorability of interactions or structural properties in
each protein conformation using data from databases of experimentally structured
proteins (such as the Protein Data Bank [PDB]). Knowledge-based energy functions
have inbuilt propensities for secondary structures, but they may not be enough for
local structure modeling. A fragment-based approach utilizes secondary structure
fragments taken from aligning sequences for protein structure modeling. Several
software programs deploy this method to reduce the entropy of conformational
search and achieve accuracy in local structure modeling. Examples include Rosetta
and TASSER/I-TASSER programs. Simons et al. developed Rosetta based on the
fragment assembly approach given by Bowie and Eisenberg (Bowie and Eisenberg
1994; Simons et al. 1997). To explore the relevant conformational space of the target
sequence and rank the resulting models on the basis of their energy function, Rosetta
employs Metropolis Monte Carlo sampling approaches combined with knowledge-
based energy functions (Bowie and Eisenberg 1994). Rosetta was able to predict a
70-amino-acid-long protein structure to near atomic resolution in CASP6 (Simons
et al. 1997). TASSER was a purely knowledge-based program for 3D model
construction of proteins. To investigate potential folding patterns, the query
sequence is first passed through a selection of typical protein structures. The
threaded and aligned sections are then used to extract continuous segments with at
least five residues. These segments are then used to reconstruct entire models of the
protein, whereas non-aligned sections are built using a lattice-based ab initio model-
ing strategy (Kallberg et al. 2012; Kaufmann et al. 2010). I-TASSER is a graded
template-based application for the prediction of the structure and function of a
protein, which was ranked at the top of many CASP events (Eswar et al. 2006).
These methods provided a stable benchmark for the AI-associated techniques to
evolve.

22.5.6 AI-Driven De Novo Protein Structure Prediction

Before delving into the world of AlphaFold, evolutionary algorithms and Monte
Carlo Methods need to be understood. Monte Carlo simulations (MCS) represent a
widely used approach for calculating the trajectories and thermodynamic character-
istics of proteins. Monte Carlo simulation employs stochastic sampling and statisti-
cal modeling to approximate mathematical functions and replicate the behavior of
intricate systems (Bradley et al. 2005). In a simulation run, a sequence of random
movements is made within the conformational space, each of which introduces
alterations to certain molecular degrees of freedom. The acceptance of each step is
contingent upon the likelihood associated with the alteration in the energy function’s
22 AI-Assisted Methods for Protein Structure Prediction and Analysis 383

value. An energy landscape is a topographical representation across the space of

conformations, illustrating the potential energy associated with every conceivable
molecular conformation (Zhang et al. 2002).
The theory of the energy landscape says that although there are n number of
possible conformations of the protein structure during folding, it can be effectively
represented using only one or a limited set of reaction coordinates without sacrificing
a remarkable volume of kinetic information (Harrison 2010). The application of
Monte Carlo and conformational space learning techniques enabled the progressive
protein structure prediction software to perform better in a limited set of time. Protein
segments can be constructed incrementally, adding one amino acid at a time, with
each consecutive step tailored to promote the adoption of low-energy conformations
(Edwards et al. 2012). Such methods have been developed and used previously for
biological macromolecular structure prediction (Travaglini-Allocatelli et al. 2009;
Wong et al. 2018; Tang et al. 2014).

22.5.7 Deep Learning Models for Tertiary Structure

Prediction

CASP 13 (Critical Assessment of Protein Structure Prediction) revolutionized the

AI-associated structure prediction landscape with the introduction of AlphaFold, a
deep learning-based template free modeling (FM) program (Wick and Siepmann
2000). This has been described as watershed moment in the field of structural
prediction for complex macromolecules like proteins (Zhang et al. 2007). CASP, a
biennial event that is organized with the goal of evaluating and assessing the
accuracy of protein structure predictions on a global scale, has been held since
1994. CASP13, which was held in December 2018, brought the real force of AI to
the fore of protein structure prediction. DeepMind’s AlphaFold was ranked first
among the template free modeling (FM) programs. It doubled the accuracy achieved
during the earlier CASP event and increased the prediction range up to 100 (Senior
et al. 2020).
The success of AlphaFold comes from the groundwork prepared by several
generations of evolution in protein research. It is a co-evolution-dependent method
which, in the first step, generates a large repertoire of multiple sequence alignments
(MSAs) of evolutionary varied homologous proteins in the range of 105–106
sequences. It then extracts the information from the residues that seem to
co-evolve, sharing a similar conformational space. AlphaFold takes these binary
contact distributions as a protein-specific statistical potential function, which is
directly optimized to facilitate protein folding (AlQuraishi 2020). AlphaFold’s
core concept is that a distribution of distances between protein residues can be
transformed into a continuous function, which can then be minimized to predict
protein structures accurately. The major portion of the success of AlphaFold was
attributed to the use of very advanced neural networks with “dilated convolutions”
384 D. Goel et al.

capable of capturing long-distance interactions in the protein sequence (Senior et al.

2020).

22.5.8 AlphaFold2

In CASP 14 in 2020, Jumper et al. submitted their newer version of the AlphaFold
under the name of AlphaFold2 (AF2) while claiming that it is completely different
from its previous version (AlQuraishi 2019b). To provide accurate end-to-end
structure prediction, AF2 proposes a unique architecture that simultaneously inte-
grates pairwise characteristics and multiple sequence alignments (MSAs), as well as
an updated output representation and loss function. An attention mechanism-based
transformer (Sippl 1990) was used in AF2, which is a recently developed deep neural
network that uses the self-attention process to collect intrinsic properties and exhibits
considerable potential for widespread AI applications. This transformer system is
made up of two modules—a decoder module and an encoder module—each of
which contains several transformer blocks with the identical structure. A feed-
forward neural network, a shortcut connection, a multi-head attention layer, and
layer normalization are all contained within each transformer block. This is termed
as “Evoformer” in AF2, which can model the prediction of protein structure as a
three-dimensional graph inference issue, where the edges of the graph are deter-
mined by the closeness of residues (AlQuraishi 2019b).
Proteins have been known to conserve their structure during evolution, which
means that even though the sequence of a particular protein may undergo changes as
much as 80%, its 3D structure may still be the same. If two residues show conser-
vation among themselves, it generally means that there is some sort of interaction
among them. This information becomes the basis for AF2. The AF2 architecture
consists of three layers: the first is the input module, the second is Evoformer, and the
third is the structure module. The input module takes the amino acid sequence, and
MSA is fabricated by searching for homologous sequences and their corresponding
structures, if any. It then generates a pairwise distance matrix between the amino
acids. Evoformer accepts the MSA as input and subjects it to deep neural learning
layers which outputs the updated MSA and an updated pair representation. Forty-
eight such layers are present in AF2, which are independent of each other and
contain two pathways: one for the MSA and the other for the pairwise representation
which represents attention as a triangle of residues. Here each triangle means that
any two residues can influence the third one. The third module, which is like a
decoder, decodes the information generated by the Evoformer and translates it into
the 3D atomic coordinates for the query sequence (Jumper et al. 2021).
The way AF2 was trained is a key factor in its success. The creators used a
method called self-distillation, which combined data from the Protein Data Bank
(PDB) with a new dataset of predicted protein structures. In this training process,
25% of the examples came from well-known PDB structures, while 75% came from
the new self-distillation dataset. The goal was to help AF2 replicate challenging
22 AI-Assisted Methods for Protein Structure Prediction and Analysis 385

protein structure predictions by using various training data enhancement techniques.

By using this combination of datasets, including the data predicted by AF2, the
model’s performance was significantly improved.

22.5.9 Applications of AlphaFold2

AlphaFold has revolutionized the protein structure landscape, and the major bene-
ficiary is structural biology community. There is a growing divide between the
proteomic data and the experimental protein structures available, as it is both time
consuming and costly to determine the structure of a protein through wet lab
procedures. However, advanced predictive methods such as AF2 still need experi-
mental validation. Other areas that have potential applications of AF include protein-
protein interactions, protein drug-interactions, enzyme mechanisms, drug discovery,
and protein design (Jumper et al. 2021). AF has been able to accurately decipher the
structure of ubiquitin-specific peptidase 7 (USP7), a complex enzyme with multiple
domains showing dynamic behavior. The predictive model was able to match the
interactions and movements of various domains as shown by the experimentally
derived models of the same protein (Vaswani et al. 2017; Yang et al. 2023). Mehmet
Akdel et al. did a comprehensive analysis of the AF2 predictions on a large dataset
comprising reference proteomes for 11 species. They compared the results with the
PDB data for various parameters, including typical structural elements, missense
variants, ligand binding, and interaction analysis. They found out that AF2 can
predict about 25% more residues with confidence as compared to template-based
homology modeling. It was also successful in predicting the protein disorder more
accurately and was able to predict structural folds that were rare even for PDB (Kim
et al. 2016). AF2 has since then found applications in antibiotic discovery (Perrakis
and Sixma 2021), disordered protein structure prediction (Akdel et al. 2022),
predicting novel human proteins with knots (Wong et al. 2022), and viral research
(Ruff and Pappu 2021), to name a few. It has also been combined with X-ray
crystallography, NMR, mass spectrometry, and CryoEM for solving the experimen-
tal data-based structures (Perlinska et al. 2023; Gutnik et al. 2023; Allison et al.
2022; Arantes et al. 2022; Laurents 2022).

22.6 Conclusion and Future Scope

Protein structure has always fascinated scientists as it allows the visualization of a

biological machinery at its finest. There has been constant effort by the researchers to
accurately predict the structure of proteins from their sequences. The collective effort
of studying the intrinsic properties of proteins has enabled the scientists to now
achieve the accuracy of near-native structures using advanced AI tools. However, it
is still far from complete, as there are several limitations to these tools which need to
386 D. Goel et al.

be addressed. Specifically, the effects of epigenetic modifications on proteins, point

mutations, post-translationally modified proteins (Yang et al. 2023; Nagaratnam
et al. 2022; Paul et al. 2022; Azzaz et al. 2022) (Buel and Walters 2022), etc., are
some of the areas where AF2 is still in its infancy. Recently protein language models
have been developed which promise better and more accurate protein folding
predictions (Bertoline et al. 2023). Whatever the case may be, it is a fast-changing
landscape, with new developments coming every day.

References

Ashok Kumar T (2013) CFSSP: Chou and Fasman secondary structure prediction server. Wide
Spectr Res J 1:15–19. https://doi.org/10.5281/zenodo.50733. (ISSN 2250-2815)
Ahmad S, Gromiha MM, Sarai A (2003) Real value prediction of solvent accessibility from amino
acid sequence. Proteins 50(4):629–635. https://doi.org/10.1002/prot.10328
Ahmed SF, Alam MSB, Hassan M, Rozbu MR, Ishtiak T, Rafa N et al (2023) Deep learning
modelling techniques: current progress, applications, advantages, and challenges. Artif Intell
Rev 56(11):13521–13617. https://doi.org/10.1007/s10462-023-10466-8
Akdel M, Pires DEV, Pardo EP, Janes J, Zalevsky AO, Meszaros B et al (2022) A structural biology
community assessment of AlphaFold2 applications. Nat Struct Mol Biol 29(11):1056–1067.
https://doi.org/10.1038/s41594-022-00849-w
Alberts B, Johnson A, Lewis J, Raff M, Roberts K, Walter P (2022) The shape and structure of
proteins. In: Alberts B (ed) Molecular biology of the cell, 4th edn. Garland Science, New York
Allison TM, Degiacomi MT, Marklund EG, Jovine L, Elofsson A, Benesch JLP et al (2022)
Complementing machine learning-based structure predictions with native mass spectrometry.
Protein Sci 31(6):e4333. https://doi.org/10.1002/pro.4333
AlQuraishi M (2019a) End-to-end differentiable learning of protein structure. Cell Systems 8(4):
292–301.e3. https://doi.org/10.1016/j.cels.2019.03.006
AlQuraishi M (2019b) AlphaFold at CASP13. Bioinformatics 35(22):4862–4865. https://doi.org/
10.1093/bioinformatics/btz422
AlQuraishi M (2020) A watershed moment for protein structure prediction. Nature 577(7792):
627–628. https://doi.org/10.1038/d41586-019-03951-0
AlQuraishi M (2021) Machine learning in protein structure prediction. Curr Opin Chem Biol 65:1–
8. https://doi.org/10.1016/j.cbpa.2021.04.005
Altschul SF, Gish W, Miller W, Myers EW, Lipman DJ (1990) Basic local alignment search tool. J
Mol Biol 215(3):403–410. https://doi.org/10.1016/S0022-2836(05)80360-2
Alzubaidi L, Zhang J, Humaidi AJ, Al-Dujaili A, Duan Y, Al-Shamma O et al (2021) Review of
deep learning: concepts, CNN architectures, challenges, applications, future directions. J Big
Data 8(1):53. https://doi.org/10.1186/s40537-021-00444-8
Anfinsen CB (1973) Principles that govern the folding of protein chains. Science 181(4096):
223–230. https://doi.org/10.1126/science.181.4096.223
Arantes P, Nierzwicki L, Belato H, D’Ordine A, Jogl G, Lisi G et al (2022) Assessing structure and
dynamics of AlphaFold2 prediction of GeoCas9. Biophys J 121:45a. https://doi.org/10.1016/j.
bpj.2021.11.2474
Arora D, Mishra D, Budhlakoti N, Srivastava S, Singh A, Kumar S (2018) Introduction of
reinforcement learning in bioinformatics. Biotech Today 8:25. https://doi.org/10.5958/
2322-0996.2018.00019.4
Azzaz F, Yahi N, Chahinian H, Fantini J (2022) The epigenetic dimension of protein structure is an
intrinsic weakness of the AlphaFold program. Biomol Ther 12(10). https://doi.org/10.3390/
biom12101527
22 AI-Assisted Methods for Protein Structure Prediction and Analysis 387

Bertoline LMF, Lima AN, Krieger JE, Teixeira SK (2023) Before and after AlphaFold2: an
overview of protein structure prediction. Front Bioinform 3:1120370. https://doi.org/10.3389/
fbinf.2023.1120370
Bowie JU, Eisenberg D (1994) An evolutionary approach to folding small alpha-helical proteins
that uses sequence information and an empirical guiding fitness function. Proc Natl Acad Sci U
S A 91(10):4436–4440. https://doi.org/10.1073/pnas.91.10.4436
Bradley P, Malmsträm L, Qian B, Schonbrun J, Chivian D, Kim DE et al (2005) Free modeling with
Rosetta in CASP6. Proteins 61(S7):128–134. https://doi.org/10.1002/prot.20729
Brooks BR, Bruccoleri RE, Olafson BD, States DJ, Swaminathan S, Karplus M (1983) CHARMM:
a program for macromolecular energy, minimization, and dynamics calculations. J Comput
Chem 4(2):187–217. https://doi.org/10.1002/jcc.540040211
Brooks BR, Brooks CL 3rd, Mackerell AD Jr, Nilsson L, Petrella RJ, Roux B et al (2009)
CHARMM: the biomolecular simulation program. J Comput Chem 30(10):1545–1614.
https://doi.org/10.1002/jcc.21287
Buel GR, Walters KJ (2022) Can AlphaFold2 predict the impact of missense mutations on
structure? Nat Struct Mol Biol 29(1):1–2. https://doi.org/10.1038/s41594-021-00714-2
Case DA, Cheatham TE 3rd, Darden T, Gohlke H, Luo R, Merz KM Jr et al (2005) The Amber
biomolecular simulation programs. J Comput Chem 26(16):1668–1688. https://doi.org/10.
1002/jcc.20290
Chen C, Chen L, Zou X, Cai P (2009) Prediction of protein secondary structure content by using the
concept of Chous pseudo amino acid composition and support vector machine. Protein Pept Lett
16:27–31. https://doi.org/10.2174/092986609787049420
David CC, Jacobs DJ (2014) Principal component analysis: a method for determining the essential
dynamics of proteins. Methods Mol Biol 1084:193–226. https://doi.org/10.1007/978-1-62703-
658-0_11
Deng H, Jia Y, Zhang Y (2018) Protein structure prediction. Int J Mod Phys B 32(18):1840009.
https://doi.org/10.1142/S021797921840009X
Diaz-Villanueva JF, Diaz-Molina R, Garcia-Gonzalez V (2015) Protein folding and mechanisms of
Proteostasis. Int J Mol Sci 16(8):17193–17230. https://doi.org/10.3390/ijms160817193
Edwards SA, Wagner J, Gräter F (2012) Dynamic Prestress in a globular protein. PLoS Comput
Biol 8(5):e1002509. https://doi.org/10.1371/journal.pcbi.1002509
Ehsaneddin A, Nina P, Alice CM, Mohammad RKM (2019) DeepPrime2Sec: deep learning for
protein secondary structure prediction from the primary sequences. bioRxiv:705426. https://doi.
org/10.1101/705426
Ema RR, Khatun A, Hossain MA, Akhond MR, Hossain N, Arafat MY (2022) Protein secondary
structure prediction using hybrid recurrent neural networks. J Comput Sci 18(7):599. https://doi.
org/10.3844/jcssp.2022.599.611
Englander SW, Mayne L, Krishna MM (2007) Protein folding and misfolding: mechanism and
principles. Q Rev Biophys 40(4):287–326. https://doi.org/10.1017/S0033583508004654
Eswar N, Webb B, Marti-Renom MA, Madhusudhan MS, Eramian D, Shen MY et al (2006)
Comparative protein structure modeling using Modeller. Curr Protoc Bioinformatics Chapter 5:
Unit-5 6. https://doi.org/10.1002/0471250953.bi0506s15
Gelman S, Fahlberg SA, Heinzelman P, Romero PA, Gitter A (2021) Neural networks to learn
protein sequence-function relationships from deep mutational scanning data. Proc Natl Acad Sci
U S A 118(48). https://doi.org/10.1073/pnas.2104878118
Goodfellow I, Bengio Y, Courville A (2016) Deep learning. MIT Press
Guo Y, Wang B, Li W, Yang B (2018) Protein secondary structure prediction improved by
recurrent neural networks integrated with two-dimensional convolutional neural networks. J
Bioinforma Comput Biol 16(05):1850021. https://doi.org/10.1142/S021972001850021X
Gutnik D, Evseev P, Miroshnikov K, Shneider M (2023) Using AlphaFold predictions in viral
research. Curr Issues Mol Biol 45(4):3705–3732. https://doi.org/10.3390/cimb45040240
Harrison RL (2010) Introduction to Monte Carlo simulation. AIP Conf Proc 1204:17–21. https://
doi.org/10.1063/1.3295638
388 D. Goel et al.

Hattori L, Benítez M, Lopes H. A deep bidirectional long short-term memory approach applied to
the protein secondary structure prediction problem. 2017 IEEE Latin American Conference on
Computational Intelligence (LA-CCI), 1–6.2017
Hochreiter S, Schmidhuber J (1997) Long short-term memory. Neural Comput 9(8):1735–1780.
https://doi.org/10.1162/neco.1997.9.8.1735
Hu H, Li Z, Elofsson A, Xie S (2019) A Bi-LSTM based ensemble algorithm for prediction of
protein secondary structure. Appl Sci 9. https://doi.org/10.3390/app9173538
Ismi DP, Pulungan R, Afiahayati. (2022) Deep learning for protein secondary structure prediction:
pre and post-AlphaFold. Computational and structural. Biotechnol J 20:6271–6286. https://doi.
org/10.1016/j.csbj.2022.11.012
Jin X, Guo L, Jiang Q, Wu N, Yao S (2022) Prediction of protein secondary structure based on an
improved channel attention and multiscale convolution module. Front Bioeng Biotechnol 10:
901018
Jisna VA, Jayaraj PB (2021) Protein structure prediction: conventional and deep learning perspec-
tives. Protein J 40(4):522–544. https://doi.org/10.1007/s10930-021-10003-y
Johnson KB, Wei WQ, Weeraratne D, Frisse ME, Misulis K, Rhee K et al (2021) Precision
medicine, AI, and the future of personalized health care. Clin Transl Sci 14(1):86–93. https://
doi.org/10.1111/cts.12884
Jolliffe IT, Cadima J (2016) Principal component analysis: a review and recent developments.
Philos Trans R Soc A Math Phys Eng Sci 374(2065):20150202. https://doi.org/10.1098/rsta.
2015.0202
Jumper J, Evans R, Pritzel A, Green T, Figurnov M, Ronneberger O et al (2021) Highly accurate
protein structure prediction with AlphaFold. Nature 596(7873):583–589. https://doi.org/10.
1038/s41586-021-03819-2
Kabsch W, Sander C (1983) Dictionary of protein secondary structure: pattern recognition of
hydrogen-bonded and geometrical features. Biopolymers 22(12):2577–2637. https://doi.org/
10.1002/bip.360221211
Kallberg M, Wang H, Wang S, Peng J, Wang Z, Lu H et al (2012) Template-based protein structure
modeling using the RaptorX web server. Nat Protoc 7(8):1511–1522. https://doi.org/10.1038/
nprot.2012.085
Kandathil SM, Lau AM, Jones DT (2023) Machine learning methods for predicting protein
structure from single sequences. Curr Opin Struct Biol 81:102627. https://doi.org/10.1016/j.
sbi.2023.102627
Kaufmann KW, Lemmon GH, Deluca SL, Sheehan JH, Meiler J (2010) Practically useful: what the
Rosetta protein modeling suite can do for you. Biochemistry 49(14):2987–2998. https://doi.org/
10.1021/bi902153g
Kelley LA, Mezulis S, Yates CM, Wass MN, Sternberg MJ (2015) The Phyre2 web portal for
protein modeling, prediction and analysis. Nat Protoc 10(6):845–858. https://doi.org/10.1038/
nprot.2015.053
Kim RQ, van Dijk WJ, Sixma TK (2016) Structure of USP7 catalytic domain and three
Ubl-domains reveals a connector alpha-helix with regulatory role. J Struct Biol 195(1):11–18.
https://doi.org/10.1016/j.jsb.2016.05.005
Kitao A (2022) Principal component analysis and related methods for investigating the dynamics of
biological macromolecules. J 5:298. https://doi.org/10.3390/j5020021
Kouza M, Faraggi E, Kolinski A, Kloczkowski A (2017) The GOR method of protein secondary
structure prediction and its application as a protein aggregation prediction tool. Methods Mol
Biol 1484:7–24. https://doi.org/10.1007/978-1-4939-6406-2_2
Kuhlman B, Bradley P (2019) Advances in protein structure prediction and design. Nat Rev Mol
Cell Biol 20(11):681–697. https://doi.org/10.1038/s41580-019-0163-x
Kumar S, Raj I, Nagpal I, Subbarao N, Gourinath S (2011) Structural and biochemical studies of
serine acetyltransferase reveal why the parasite Entamoeba histolytica cannot form a cysteine
synthase complex. J Biol Chem 286(14):12533–12541. https://doi.org/10.1074/jbc.M110.
197376
22 AI-Assisted Methods for Protein Structure Prediction and Analysis 389

Laurents D (2022) AlphaFold 2 and NMR spectroscopy: partners to understand protein structure,
dynamics and function. Front Mol Biosci 9:9. https://doi.org/10.3389/fmolb.2022.906437
Le NQK, Yapp EKY, Yeh H-Y (2019) ET-GRU: using multi-layer gated recurrent units to identify
electron transport proteins. BMC Bioinformatics. 20(1):377. https://doi.org/10.1186/s12859-
019-2972-5
Lee M (2023) Recent advances in deep learning for protein-protein interaction analysis: a compre-
hensive review. Molecules 28(13). https://doi.org/10.3390/molecules28135169
Lee J, Wu S, Zhang Y. Ab initio protein structure prediction 2009:3–25. doi: https://doi.org/10.
1007/978-1-4020-9058-5_1
Lee J, Freddolino PL, Zhang Y. Ab initio protein structure prediction. 2017:3–35. doi: https://doi.
org/10.1007/978-94-024-1069-3_1
Liu X (2017) Deep recurrent neural network for protein function prediction from sequence.
arXiv:701.08318. https://doi.org/10.48550/arXiv.1701.08318
Liu Y, Cheng J, Ma Y, Chen Y. Protein secondary structure prediction based on two dimensional
deep convolutional neural networks. 2017 3rd IEEE International Conference on Computer and
Communications (ICCC), 2017. p. 1995–9
Lyu Z, Wang Z, Luo F, Shuai J, Huang Y (2021) Protein secondary structure prediction with a
reductive deep learning method. Front Bioeng Biotechnol 9:687426. https://doi.org/10.3389/
fbioe.2021.687426
Melman P, Roshan U. K-means-based feature learning for protein sequence classification. 2018
Minami S, Sawada K, Chikenji G (2014) How a spatial arrangement of secondary structure
elements is dispersed in the universe of protein folds. PLoS One 9(9):e107959. https://doi.
org/10.1371/journal.pone.0107959
Nagaratnam N, Martin-Garcia JM, Yang JH, Goode MR, Ketawala G, Craciunescu FM et al (2022)
Structural and biophysical properties of FopA, a major outer membrane protein of Francisella
tularensis. PLoS One 17(8):e0267370. https://doi.org/10.1371/journal.pone.0267370
Novak WRP (2021) Tertiary structure domains, folds, and motifs. In: Bell E (ed) Molecular life
sciences: an encyclopedic reference. Springer New York, New York, NY, pp 1–5
Pakhrin SC, Shrestha B, Adhikari B, Kc DB (2021) Deep learning-based advances in protein
structure prediction. Int J Mol Sci 22(11). https://doi.org/10.3390/ijms22115553
Parras-Moltó M, Campos-Laborie FJ, García-Diéguez J, Rodríguez-Griñolo MR, Pérez-Pulido AJ
(2013) Classification of protein motifs based on subcellular localization uncovers evolutionary
relationships at both sequence and functional levels. BMC Bioinformatics. 14(1):229. https://
doi.org/10.1186/1471-2105-14-229
Paul D, Sanap G, Shenoy S, Kalyane D, Kalia K, Tekade RK (2021) Artificial intelligence in drug
discovery and development. Drug Discov Today 26(1):80–93. https://doi.org/10.1016/j.drudis.
2020.10.010
Paul B, Weeratunga S, Tillu VA, Hariri H, Henne WM, Collins BM (2022) Structural predictions of
the SNX-RGS proteins suggest they belong to a new class of lipid transfer proteins. Front Cell
Dev Biol 10:826688. https://doi.org/10.3389/fcell.2022.826688
Pearce R, Zhang Y (2021) Toward the solution of the protein structure prediction problem. J Biol
Chem 297(1):100870. https://doi.org/10.1016/j.jbc.2021.100870
Perlinska AP, Niemyska WH, Gren BA, Bukowicki M, Nowakowski S, Rubach P et al (2023)
AlphaFold predicts novel human proteins with knots. Protein Sci 32(5):e4631. https://doi.org/
10.1002/pro.4631
Perrakis A, Sixma TK (2021) AI revolutions in biology. EMBO Rep 22(11):e54046. https://doi.org/
10.15252/embr.202154046
Ruff KM, Pappu RV (2021) AlphaFold and implications for intrinsically disordered proteins. J Mol
Biol 433(20):167208. https://doi.org/10.1016/j.jmb.2021.167208
Russo A, Borras A (2022) Comparison of dimension reduction techniques applied to the analysis of
airborne radionuclide activity concentration. J Environ Radioact 244-245:106813. https://doi.
org/10.1016/j.jenvrad.2022.106813
Schrodinger, LLC. The PyMOL molecular graphics system, Version 1.8. 2015
390 D. Goel et al.

Senior AW, Evans R, Jumper J, Kirkpatrick J, Sifre L, Green T et al (2020) Improved protein
structure prediction using potentials from deep learning. Nature 577(7792):706–710. https://doi.
org/10.1038/s41586-019-1923-7
Sharma KA, Srivastava R (2021) Protein secondary structure prediction using character bi-gram
embedding and Bi-LSTM. Curr Bioinforma 16(2):333–338. https://doi.org/10.2174/
1574893615999200601122840
Simons KT, Kooperberg C, Huang E, Baker D (1997) Assembly of protein tertiary structures from
fragments with similar local sequences using simulated annealing and Bayesian scoring func-
tions. J Mol Biol 268(1):209–225. https://doi.org/10.1006/jmbi.1997.0959
Sippl MJ (1990) Calculation of conformational ensembles from potentials of mean force. An
approach to the knowledge-based prediction of local structures in globular proteins. J Mol
Biol 213(4):859–883. https://doi.org/10.1016/s0022-2836(05)80269-4
Sun PD, Foster CE, Boyington JC (2004) Overview of protein structural and functional folds. Curr
Protoc Protein Sci Chapter 17(1):Unit 17 1. https://doi.org/10.1002/0471140864.ps1701s35
Tang K, Zhang J, Liang J (2014) Fast protein loop sampling and structure prediction using distance-
guided sequential chain-growth Monte Carlo method. PLoS Comput Biol 10(4):e1003539.
https://doi.org/10.1371/journal.pcbi.1003539
Taye MM (2023) Understanding of machine learning with deep learning: architectures, workflow,
applications and future directions. Computers 12. https://doi.org/10.3390/computers12050091
Teletin M, Czibula G, Albert S, Bocicor M-I (2018) Using unsupervised learning methods for
enhancing protein structure insight. Procedia Comput Sci 126:19–28. https://doi.org/10.1016/j.
procs.2018.07.205
Tokuriki N, Tawfik DS (2009) Protein dynamism and evolvability. Science 19359577(5934):
203–207
Travaglini-Allocatelli C, Ivarsson Y, Jemth P, Gianni S (2009) Folding and stability of globular
proteins and implications for function. Curr Opin Struct Biol 19(1):3–7. https://doi.org/10.1016/
j.sbi.2008.12.001
Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez AN et al (2017) Attention is all you
need. In: Guyon I, Luxburg UV, Bengio S, Wallach H, Fergus R, Vishwanathan S et al (eds)
31st Conference on Neural Information Processing Systems (NIPS 2017). NeurIPS Proceed-
ings, Long Beach, CA, USA
Vijayan RSK, Kihlberg J, Cross JB, Poongavanam V (2022) Enhancing preclinical drug discovery
with artificial intelligence. Drug Discov Today 27(4):967–984. https://doi.org/10.1016/j.drudis.
2021.11.023
Wang J, Wolf RM, Caldwell JW, Kollman PA, Case DA (2004) Development and testing of a
general amber force field. J Comput Chem 25(9):1157–1174. https://doi.org/10.1002/jcc.20035
Wang S, Peng J, Ma J, Xu J (2016a) Protein secondary structure prediction using deep
convolutional neural fields. Sci Rep 6(1):18962. https://doi.org/10.1038/srep18962
Wang S, Li W, Liu S, Xu J (2016b) RaptorX-property: a web server for protein structure property
prediction. Nucleic Acids Res 44(W1):W430–W435. https://doi.org/10.1093/nar/gkw306
Waterhouse A, Bertoni M, Bienert S, Studer G, Tauriello G, Gumienny R et al (2018) SWISS-
MODEL: homology modelling of protein structures and complexes. Nucleic Acids Res 46(W1):
W296–W303. https://doi.org/10.1093/nar/gky427
Webb B, Sali A (2016) Comparative protein structure modeling using MODELLER. Curr Protoc
Bioinformatics 54:5.6.1–5.6.37. https://doi.org/10.1002/cpbi.3
Wick CD, Siepmann JI (2000) Self-adapting fixed-end-point configurational-bias Monte Carlo
method for the regrowth of interior segments of chain molecules with strong intramolecular
interactions. Macromolecules 33(19):7207–7218. https://doi.org/10.1021/ma000172g
Wong S, Liu J, Kou S (2018) Exploring the conformational space for protein folding with
sequential Monte Carlo. Ann Appl Stat 12:1628–1654. https://doi.org/10.1214/17-AOAS1124
Wong F, Krishnan A, Zheng EJ, Sträk H, Manson AL, Earl AM et al (2022) Benchmarking
AlphaFold-enabled molecular docking predictions for antibiotic discovery. Mol Syst Biol
18(9):e11081. https://doi.org/10.15252/msb.202211081
 22 AI-Assisted Methods for Protein Structure Prediction and Analysis 391

Yang J, Zhang Y (2015) I-TASSER server: new development for protein structure and function
predictions. Nucleic Acids Res 43(W1):W174–W181. https://doi.org/10.1093/nar/gkv342
Yang Y, Heffernan R, Paliwal K, Lyons J, Dehzangi I, Sharma A et al (2017) SPIDER2: a package
to predict secondary structure, accessible surface area, and main-chain torsional angles by deep
neural networks. Methods Mol Biol 1484, 55:–63
Yang S, Wang Y, Cruz-Gutierrez K, Wu F, Ding C-F (2021) Localnet: a simple recurrent neural
network model for protein secondary structure prediction using local amino acid sequences
only. Research Square
Yang Z, Zeng X, Zhao Y, Chen R (2023) AlphaFold2 and its applications in the fields of biology
and medicine. Signal Transduct Target Ther 8(1):115. https://doi.org/10.1038/s41392-023-
01381-z
Yu C-H, Chen W, Chiang Y-H, Guo K, Martin Moldes Z, Kaplan DL et al (2022) End-to-end deep
learning model to predict and design secondary structure content of structural proteins. ACS
Biomater Sci Eng 8(3):1156–1165. https://doi.org/10.1021/acsbiomaterials.1c01343
Zaslavsky L, Ciufo S, Fedorov B, Tatusova T (2016) Clustering analysis of proteins from microbial
genomes at multiple levels of resolution. BMC Bioinformatics 17(Suppl 8):276. https://doi.org/
10.1186/s12859-016-1112-8
Zhang Y, Kihara D, Skolnick J (2002) Local energy landscape flattening: parallel hyperbolic Monte
Carlo sampling of protein folding. Proteins 48(2):192–201. https://doi.org/10.1002/prot.10141
Zhang J, Kou SC, Liu JS (2007) Biopolymer structure simulation and optimization via fragment
regrowth Monte Carlo. J Chem Phys 126(22):225101. https://doi.org/10.1063/1.2736681
Zhang G, Tang Q, Feng P, Chen W (2023) IPs-GRUAtt: an attention-based bidirectional gated
recurrent unit network for predicting phosphorylation sites of SARS-CoV-2 infection. Mol Ther
Nucleic Acids 32:28–35. https://doi.org/10.1016/j.omtn.2023.02.027
Zhao Y, Liu Y (2021) OCLSTM: optimized convolutional and long short-term memory neural
network model for protein secondary structure prediction. PLoS One 16:e0245982. https://doi.
org/10.1371/journal.pone.0245982
Zhu Y, Ting KM, Carman MJ (2016) Density-ratio based clustering for discovering clusters with
varying densities. Pattern Recogn 60:983–997. https://doi.org/10.1016/j.patcog.2016.07.007
Zhu L, Davari MD, Li W (2021) Recent advances in the prediction of protein structural classes:
feature descriptors and machine learning algorithms. Crystals 11. https://doi.org/10.3390/
cryst11040324
Zimmermann L, Stephens A, Nam SZ, Rau D, Kubler J, Lozajic M et al (2018) A completely
reimplemented MPI bioinformatics toolkit with a new HHpred server at its core. J Mol Biol
430(15):2237–2243. https://doi.org/10.1016/j.jmb.2017.12.007

View publication stats