2020 Book BusinessIntelligenceForEnterprises SpringerNature

Download as pdf or txt
Download as pdf or txt
You are on page 1of 290

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

net/publication/342043827

An Overall Perspective on Establishing End-to-End Security in Enterprise IoT


(E-IoT)

Chapter · June 2020


DOI: 10.1007/978-3-030-44407-5_4

CITATIONS READS

2 1,586

3 authors, including:

Shreyas Rao
Sahyadri College of Engineering and Management
12 PUBLICATIONS 70 CITATIONS

SEE PROFILE

All content following this page was uploaded by Shreyas Rao on 03 December 2020.

The user has requested enhancement of the downloaded file.


EAI/Springer Innovations in Communication and Computing

Anandakumar Haldorai
Arulmurugan Ramu
Syed Abdul Rehman Khan Editors

Business
Intelligence
for Enterprise
Internet
of Things
EAI/Springer Innovations in Communication
and Computing

Series Editor
Imrich Chlamtac, European Alliance for Innovation, Ghent, Belgium
Editor’s Note
The impact of information technologies is creating a new world yet not fully
understood. The extent and speed of economic, life style and social changes already
perceived in everyday life is hard to estimate without understanding the technological
driving forces behind it. This series presents contributed volumes featuring the
latest research and development in the various information engineering technologies
that play a key role in this process.
The range of topics, focusing primarily on communications and computing
engineering include, but are not limited to, wireless networks; mobile communication;
design and learning; gaming; interaction; e-health and pervasive healthcare; energy
management; smart grids; internet of things; cognitive radio networks; computation;
cloud computing; ubiquitous connectivity, and in mode general smart living, smart
cities, Internet of Things and more. The series publishes a combination of expanded
papers selected from hosted and sponsored European Alliance for Innovation (EAI)
conferences that present cutting edge, global research as well as provide new
perspectives on traditional related engineering fields. This content, complemented
with open calls for contribution of book titles and individual chapters, together
maintain Springer’s and EAI’s high standards of academic excellence. The audience
for the books consists of researchers, industry professionals, advanced level students
as well as practitioners in related fields of activity include information and
communication specialists, security experts, economists, urban planners, doctors,
and in general representatives in all those walks of life affected ad contributing to
the information revolution.
About EAI
EAI is a grassroots member organization initiated through cooperation between
businesses, public, private and government organizations to address the global
challenges of Europe’s future competitiveness and link the European Research
community with its counterparts around the globe. EAI reaches out to hundreds of
thousands of individual subscribers on all continents and collaborates with an
institutional member base including Fortune 500 companies, government
organizations, and educational institutions, provide a free research and innovation
platform.
Through its open free membership model EAI promotes a new research and
innovation culture based on collaboration, connectivity and recognition of excellence
by community.

More information about this series at http://www.springer.com/series/15427


Anandakumar Haldorai
Arulmurugan Ramu • Syed Abdul Rehman Khan
Editors

Business Intelligence
for Enterprise Internet
of Things
Editors
Anandakumar Haldorai Arulmurugan Ramu
Computer Science and Engineering Presidency University
Sri Eshwar College Engineering Rajanakunte, Yelahanka, India
Coimbatore, Tamil Nadu, India

Syed Abdul Rehman Khan


Tsinghua University
Beijing, China

ISSN 2522-8595     ISSN 2522-8609 (electronic)


EAI/Springer Innovations in Communication and Computing
ISBN 978-3-030-44406-8    ISBN 978-3-030-44407-5 (eBook)
https://doi.org/10.1007/978-3-030-44407-5

© Springer Nature Switzerland AG 2020


This work is subject to copyright. All rights are reserved 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 Switzerland AG
The registered company address is: Gewerbestrasse 11, 6330 Cham, Switzerland
Preface

The Internet of Things (IoT) has become an important research domain as enterprise
applications, systems, infrastructures, and their applications have shown their
potential in recent years. IoT is an enterprise bound for significant growth, and it
will have a major impact on the lives of consumers and professionals around the
world. It will enable enterprise industry to be a multitrillion dollar industry by 2025,
including enterprise manufacturing, enterprise transportation, enterprise smart mar-
ket, enterprise utilities, enterprise healthcare, etc. It will also change the way we
think about producer and consumer networks. The expectations of IoT and its rele-
vant products in this new era are quite high. Instead of smartness alone, consumers
of IoT products and services would like to see IoT technologies bring about more
intelligent systems and environments.
This book, Business Intelligence for Enterprise Internet of Things, presents the
most recent challenges and developments in enterprise intelligence with the objec-
tive of promoting awareness and best practices for the real world. It aims to present
new directions for further research and technology improvements in this important
area. Its chapters include IoT enterprise system architecture, IoT-enabling enter-
prise technologies, and IoT enterprise services and applications, for example, enter-
prise on demands, market impacts, and its implications on smart technologies, big
data enterprise management, and future enterprise Internet design for various IoT
use cases, such as share markets, healthcare, smart cities, smart environments, smart
communications, and smart homes.
This book also covers ideas, methods, algorithms, and tools for the in-depth
study of performance and reliability of business intelligence for enterprise Internet
of Things. The scope of business intelligence is to explore and present numerous
research contributions relating to the field of neural network computing, business
specifications, evolutionary computation, enterprise modeling and simulation, web
intelligence, healthcare informatics, social relationship, energy, and end-to-end
security in enterprise-aware management system in enterprise Internet of Things.
In this book, we present techniques and detailed perspectives of business intel-
ligence for enterprise Internet of Things that can be used in overcoming and solving

v
vi Preface

complex tasks in enterprise system. This book is based on various research horizons
and contributions focusing on IoT enterprise system challenges over:
• Development of innovative enterprise architecture for the Internet of Things
• Enterprise IoT modeling: supervised, unsupervised, and reinforcement learning
• The Internet of Things evolutionary computation, enterprise modeling, and
simulation
• Development of new IoT technologies for business intelligence and large-scale
enterprise analysis
• Uncertainty modeling in big data analytics for IoT
• Providing solutions to pressing problems across areas including connected and
autonomous vehicles, automation, healthcare, and enterprise security using the
Internet of Things
• The management of enterprise in mobile transparent computing for the Internet
of Things
• Bridge developments in artificial intelligence to real enterprise applications in
collaboration with IoT partners
• New generation of scientists to address the skills shortage in these areas and
increase competitiveness
• Applications and services for enterprise systems such as complex systems, multi-­
agent systems, game theory, and statistics
• Advanced future perspective in enterprise for the Internet of Things
This book opens the door for authors toward current research in enterprise
Internet of Things systems for business intelligence.
We would like to thank Ms. Mary E. James, Senior Editor, Applied Sciences,
Springer, and Ms. Eliska Vlckova, Managing Editor, European Alliance for
Innovation (EAI), for their great support.
We anticipate that this book will open new entrance for further research and
technology improvements. All the chapters provide a complete overview of busi-
ness intelligence for enterprise Internet of Things. This book will be handy for aca-
demicians, research scholars, and graduate students in engineering discipline.

Coimbatore, Tamil Nadu, India  Anandakumar Haldorai


Rajanakunte, Yelahanka, India  Arulmurugan Ramu
Beijing, China  Syed Abdul Rehman Khan
Contents

  1 Internet of Things (IoTs) Evolutionary Computation,


Enterprise Modelling and Simulation����������������������������������������������������    1
A. Haldorai, A. Ramu, and M. Suriya
  2 Organization Internet of Things (IoTs): Supervised,
Unsupervised, and Reinforcement Learning����������������������������������������   27
A. Haldorai, A. Ramu, and M. Suriya
  3 Enterprise IoT Modeling: Supervised, Unsupervised,
and Reinforcement Learning������������������������������������������������������������������   55
Rajesh Kumar Dhanaraj, K. Rajkumar, and U. Hariharan
  4 An Overall Perspective on Establishing End-to-End Security
in Enterprise IoT (E-IoT)������������������������������������������������������������������������   81
Vidya Rao, K. V. Prema, and Shreyas Suresh Rao
  5 Advanced Machine Learning for Enterprise IoT Modeling����������������   99
N. Deepa and B. Prabadevi
  6 Enterprise Architecture for IoT: Challenges and Business
Trends�������������������������������������������������������������������������������������������������������� 123
A. Haldorai, A. Ramu, and M. Suriya
  7 Semi-Supervised Machine Learning Algorithm for Predicting
Diabetes Using Big Data Analytics �������������������������������������������������������� 139
Senthilkumar Subramaniyan, R. Regan, Thiyagarajan Perumal,
and K. Venkatachalam
  8 On-the-Go Network Establishment of IoT Devices to Meet
the Need of Processing Big Data Using Machine Learning
Algorithms������������������������������������������������������������������������������������������������ 151
S. Sountharrajan, E. Suganya, M. Karthiga, S. S. Nandhini,
B. Vishnupriya, and B. Sathiskumar

vii
viii Contents

  9 Analysis of Virtual Machine Placement and Optimization Using


Swarm Intelligence Algorithms�������������������������������������������������������������� 169
R. B. Madhumala and Harshvardhan Tiwari
10 Performance Evaluation of Different Neural Network
Classifiers for Sanskrit Character Recognition������������������������������������ 185
R. Dinesh Kumar, C. Sridhathan, and M. Senthil Kumar
11 GA with Repeated Crossover for Rectifying Optimization
Problems �������������������������������������������������������������������������������������������������� 195
Mayank Jha and Sunita Singhal
12 An Algorithmic Approach to System Identification in the Delta
Domain Using FAdFPA Algorithm �������������������������������������������������������� 203
Souvik Ganguli, Gagandeep Kaur, Prasanta Sarkar,
and S. Suman Rajest
13 An IoT-Based Controller Realization for PV System
Monitoring and Control����������������������������������������������������������������������    213
Jyoti Gupta, Manish Kumar Singla, Parag Nijhawan,
Souvik Ganguli, and S. Suman Rajest
14 Development of an Efficient, Cheap, and Flexible IoT-Based
Wind Turbine Emulator�������������������������������������������������������������������������� 225
Manish Kumar Singla, Jyoti Gupta, Parag Nijhawan,
Souvik Ganguli, and S. Suman Rajest
15 An Application of IoT to Develop Concept of Smart Remote
Monitoring System���������������������������������������������������������������������������������� 233
Meera Sharma, Manish Kumar Singla, Parag Nijhawan,
Souvik Ganguli, and S. Suman Rajest
16 Heat Maps for Human Group Activity in Academic Blocks���������������� 241
Rajkumar Rajasekaran, Fiza Rasool, Sparsh Srivastava,
Jolly Masih, and S. Suman Rajest
17 Emphasizing on Space Complexity in Enterprise Social
Networks for the Investigation of Link Prediction Using
Hybrid Approach ������������������������������������������������������������������������������������ 253
J. Gowri Thangam and A. Sankar
18 Overview on Deep Neural Networks: Architecture,
Application and Rising Analysis Trends������������������������������������������������ 271
V. Niranjani and N. Saravana Selvam
Index������������������������������������������������������������������������������������������������������������������ 279
About the Authors

Anandakumar Haldorai is Associate Professor and


Research Head in the Department of Computer Science
and Engineering, Sri Eshwar College of Engineering,
Coimbatore, Tamil Nadu, India. He has received his
Master’s in Software Engineering from PSG College of
Technology, Coimbatore, and his PhD in Information
and Communication Engineering from PSG College of
Technology under Anna University, Chennai. His
research areas include Big Data, Cognitive Radio
Networks, Mobile Communications, and Networking
Protocols. He has authored more than 82 research
papers in reputed international journals and IEEE con-
ferences and 9 books and several book chapters in
reputed publishers such as Springer and IGI. He is
Editor in Chief of Keai-Elsevier IJIN and Inderscience
IJISC and Guest Editor of several journals with Elsevier,
Springer, Inderscience, etc. Also, he served as a
Reviewer for IEEE, IET, Springer, Inderscience, and
Elsevier journals and has been the General Chair,
Session Chair, and Panelist in several conferences. He
is Senior Member of IEEE, IET, and ACM and Fellow
Member of EAI research group.

ix
x About the Authors

Arulmurugan Ramu is a Professor at Presidency


University, Bangalore, India. His research focuses on
the automatic interpretation of images and related prob-
lems in machine learning and optimization. His main
research interest is in vision, particularly high-level
visual recognition. He has authored more than 35
papers in major computer vision and machine learning
conferences and journals. He c­ompleted his PhD in
Information and Communication Engineering and his
MTech and BTech in Information Technology all from
Anna University of Technology, Chennai. He has
guided several PhD research scholars under the area of
Image Processing using machine learning. He is an
Associate Editor of Inderscience IJISC journal. He was
awarded Best Young Faculty Award 2018 and nomi-
nated for Best Young Researcher Award (Male) by
International Academic and Research Excellence
Awards (IARE-2019).

Syed Abdul Rehman Khan is an expert in Supply


Chain and Logistics Management. He achieved his
CSCP (Certified Supply Chain Professional) Certificate
from the USA and successfully completed his PhD in
China. Since 2018, he has been affiliated with Tsinghua
University as a Postdoctoral Researcher. He has more
than 12 years’ core experience of supply chain and
logistics at industry and academic levels. He has
attended several international conferences and has been
invited as Keynote Speaker in different countries. He
has published more than 155 scientific research papers
in different well-renowned international peer-reviewed
journals and conferences. He has authored four books
related to the sustainability in supply chain and busi-
ness operations. He is a Regular Contributor to confer-
ences and workshops around the world. In the last 2
years, he has won five different national-/provincial-
level research projects. In addition, he has achieved sci-
entific innovation awards three times consecutively by
the Education Department of Shaanxi Provincial
Government, China. Also, he holds memberships in the
following well-renowned institutions and supply chain
bodies/associations: APCIS, USA; Production and
Operations Management Society, India; Council of
Supply Chain Management Professionals, USA;
Supply Chain Association of Pakistan; and Global
Supply Chain Council, China.
Chapter 1
Internet of Things (IoTs) Evolutionary
Computation, Enterprise Modelling
and Simulation

A. Haldorai , A. Ramu , and M. Suriya

1.1 Introduction

The Internet of Things (IoTs), for a few decades, has constituted a lot of vital topics
concerning the future state of industries. Information and Communication
Technology (ICT) is applicable in small telecommunication devices, which are
affordable and regarded more effectively in terms of processing to accessing the
internet. Moreover, Big Data technology was founded to allow businesses to store
massive amounts of information, and evaluate the incoming streams of data with
refined algorithms in actual time. As such, the evolution of the Internet of Things
(IoTs) has allowed companies to create useful remedies for various case scenarios
in various domains. In that regard, the IoTs can be regarded as the shelter terminol-
ogy for various disciplines, which have already been considered in organizational
automations. IoTs also enhances the integration of technological disciplines such as
the automation of sensing information with enterprise resource organization data.
Despite the fact that IoTs technologies are being promoted extensively, just a few
of the disciplines utilize cases which have to be implemented in organizational
cases. Contrary to that, there are about a hundred of IoTs technologies and plat-
forms, which have not been exploited yet. Moreover, there are various initiatives
that explain various technological standards and their respective developments from
establishments. The oversupply of the technological vendors enhances the imple-
mentation of integrated tech solutions. For instance, the research on development of

A. Haldorai (*)
Sri Eshwar College of Engineering, Coimbatore, Tamil Nadu, India
A. Ramu
Presidency University, Bangalore, India
M. Suriya
KPR Institute of Engineering and Technology, Coimbatore, Tamil Nadu, India

© Springer Nature Switzerland AG 2020 1


A. Haldorai et al. (eds.), Business Intelligence for Enterprise Internet of Things,
EAI/Springer Innovations in Communication and Computing,
https://doi.org/10.1007/978-3-030-44407-5_1
2 A. Haldorai et al.

IoTs suggest that interoperability, security and internet connectivity represent


approximately three vital concerns in implementing various Internet Technology
(IT) solutions. For the industries and the developers that have implemented the IoTs
remedies, the performances initiated by them have now been considered based on
connectivity. As such, this reflects our evaluation, which means that the rate of per-
formance is considered based on architecture and developments created by users
within a network cycle.

1.2  ackground Analysis of Internet of Things (IoTs)


B
for Modern Manufacturing

In this part, the connection of novel manufacturing paradigms and Internet


Technology (IT) are discussed. The necessities of the organization models for effec-
tive adaption of IoTs are also considered. Many novel infrastructures for organiza-
tions are considered first. The discussions indicate that IoTs is planned effectively
based on the architecture of the manufacturing industries. Consistently, every sys-
tem element in the ES requires a data unit, which enhances the implementation of
decisions on the elementary behaviours of the obtained dataset. A part from that, the
acquisition of information, information transfer and the process of decision-making
are fundamental functions that have to be considered in each model. In reference to
the Reliability Theory, the Internet of Things is capable of proposing novel reme-
dies that enhance scheduling, planning and the control of manufacturing organiza-
tion at any level.

1.2.1 I oTs Robustness Evaluated Based on the Reliability


Theory

Based on the aspect of scalability, the idea of reliability and robustness is consid-
ered. Thus, the IoTs shall be composed on a lot of electronic appliances whereby a
lot of them are difficult to reconfigure. The idea of replacement is different from
reconfiguration on tablet or desktop computers, which necessitate regular imple-
mentation of software updates to enhance modern upgrades. These updates require
CPU power, memory and disk space. As mentioned by the Metcalfe’s Laws, more
novel interconnectivity between systems and subsystems enhances the degree of
system probability that is concerned with the evaluation of system failures. The
capability of IoTs to function for a long time irrespective of the condition of soft-
ware and hardware is crucial to achieving user trust and acceptance. The bits errors
can possibly lead to uncontrollable issues in massive networks, which might possi-
bly characterize the IoTs. OR methods like the Reliability Theory will potentially
be used to forecast on IoTs reliability and robustness. As such, there was need to
1 Internet of Things (IoTs) Evolutionary Computation, Enterprise Modelling… 3

formulate an approach based on the Reliability theory for the present subsystem
networks, which are dependent on applications and perception layers. The total reli-
ability RT is calculated using the below formula.


(1 − Rk ) 
4
RT = R1 × R5 ×  1 − π (1.1)
 k = 2 

In the Eq. (1.1), R1 and R5 represent the relative forms of reliabilities of application
and perception layers, whereas R2, R3 and R4 show the reliability dependent on the
internet, satellite networks and mobile networks. This framework can therefore be
updated and expanded in accommodating the remaining IoTs features and charac-
teristics of novel IoTs devices in operation. As such, the RT can potentially be
applied onto the remaining IoTs features like data reliability. This application initi-
ated the introduction of an application known as Fog Computing onto the IoTs
whereas the researcher in [1] evaluated reliability of the methodology through the
combination necessities of cloud and grid based on actuators and sensors in the
IoTs. Resultantly, it was evident that reliability subsystems can be controlled. The
IoTs software reliability in the IoTs can potentially be another important concern
because software reliability may be termed as a special concern in RT. The reliabil-
ity of hardware can be considered based on material or component failure, prevent-
ing a subsystem to perform as intended earlier. The reliability of software in the
IoTs is considerably difficult to evaluate because software can potentially provide
unanticipated findings for a lot of reasons like unutilized information retrieved from
another networking appliance, which has not been noticed in the designing phase.
The obsolescence of the entrenched software in the IoTs subsystems might not be
maintained, altering reliability.
The researchers in [2] recommend a vital technique in which IoTs can be applied
in business to enhance productivity. Nonetheless, the application of IoTs may differ
for various data subsystems when it comes to providing high-resolution information
in actual time. As such, this diminishes the overall costs of moving information
from the actual world into the virtual world; for instance, the RFID tags can poten-
tially eliminate costly manual stocks in the process of computing information into
the computer. Based on the research, there are various means of applying IoTs in
business to boost productivity. These include proximity triggers such as self-­
checking in libraries, automated sensor triggers like the networking smoke deter-
miners and automatic product securities. As such, this indicates that IoTs represents
the quantum leap-forward retrieved through the internet with the capability of act-
ing as an element that governs the entire discipline for managing complex systems
and organizations. Moreover, System Dynamics (SD) has also been implemented in
organizational investment based on new technologies, particularly the incorporation
of internet in China and India. Principally, organizational investment in the IoTs can
potentially be modelled in the same manner. As such the researchers in [3] acknowl-
edge that SD can potentially be represented to quantify the ‘soft’ variables that are
fundamental in the process of incorporation of both technical and social aspects.
4 A. Haldorai et al.

The major segment in SD defines the results evident in the behaviour of subsystems
from interactions in feedback loops. Actually, different standardized feedback
archetypes and looks have been noticed, including the ‘fails and fixes’ and the ‘bur-
dens and shifts’. In that manner, the dominant element of the internet incorporation
was, over the years, argued to be the Contagion factor. With this effect or factor,
innovators begin the process of adoption before developing the process of commu-
nication. This contagion effect can potentially spread the message to the users, mak-
ing the process of adoption to be assumed by the imitators. Eventually, there will be
no new clients left, hence tampering with the whole adoption process, which leads
to market saturation. Moreover, there will also be a negative feedback factor related
to adoption like the concerns realized from security problems, internet dynamics
and IoTs. As such, internet incorporation will be dependent on the balancing aspect
on negative feedback loop organization.

1.3 Literature Analysis of Next-Generation of Enterprises

1.3.1  haracteristic of the Upcoming Generation


C
of Enterprises

In this part, we evaluate the characteristics of next-generation enterprise, which are


crucial in the process of analysing the IoTs centred on ES that can mitigate prob-
lematic requirements. The characteristics include the following:
• Decentralized Decision-Making Process: The various levels and domain of orga-
nizational activities are progressively becoming more diversified compared to
how they were years back. As such, hierarchical planning is applicable in effec-
tive enterprise planning for subsystem incorporation. Nonetheless, system com-
plexities can therefore be enhanced critically based on system dynamics and
scale. The centralized subsystem can potentially lead to fundamental inflexibility
and time delay, which enhance the response transition quickly. In that case,
decentralized and distributed architecture could be fundamental in dealing with
system dynamics and complexities.
• Dynamic and Flat Structures: Timely feedback to urgent issues necessitates more
decentralized and distributed organizational architecture. In that regard, the
obtained information can be utilized in the process of decision-making in a
timely manner. Based on the association between various networking compo-
nents, one can witness various problems such as delivering data to the concerned
components, particularly under central organization. The collected data is thus
gathered and transferred to a centralized database before being sent to the object
the moment the subsystems receive the requests from various objects.
Nonetheless, the centralized structure has its own problems, especially when
handling Big Data in a heterogeneity surrounding.
1 Internet of Things (IoTs) Evolutionary Computation, Enterprise Modelling… 5

Based on the perspective of Big Data, the management of massive streams of


data faces two challenges altering the incorporation of information systems to the
upcoming generation of enterprises. These issues include the costs incurred during
the process of decision-making. These costs normally increase when subsystem
complexity advances. As such, this leads to resource redundancy in maintaining
information locally, which is the second issue. The redundancy level wastes more
resources and time for information transfer when Big Data is transferred to other
decision-making structures.
Based on the heterogeneity surrounding, there is diversified and increased manu-
facturing resources, which have advanced the heterogeneous condition of manufac-
turing surroundings. These varieties exist at the facet of customized products,
location distributions, regulations, cultures, suppliers, standards and optional
organizations.

1.3.2 Characteristics on IoTs for Industrial Applications

The features of IoTs in industrial application include the following:


• System Dynamics: The organization of IoTs is never static; hence, it permits
various system elements to be configured any moment whenever there is need to
do so. Thus, this allows the integration of data over various industrial boundaries.
The cornerstone industrial segment can potentially assimilate with the virtual
companies, which potentially establish dynamic connections with a certain proj-
ect. The structure can therefore be dismissed whenever the projects are com-
pleted, and especially when the industry should proceed to another project.
Enterprises should have the capability to control the restructuring of virtual
industrial alley.
• Assimilated WSNs and RFIDs Networks: These features represent one of the
fundamental elements on IoTs as an information transfer protocol for data shar-
ing and acquisition. An industrial system includes a lot of sensors to obtain
machine appliances, actual-time information actuators, conveyors and features.
The ancient assimilated communications are linked to peer-to-peer or point-to-­
point, which makes it possible to make any changes. WSN and RFID assure
effective methods of supporting the decentralized and distribution of industrial
resources.
• Cloud Computing: Modern industrial operation engages various activities of
decision-making that necessitates intensive data and high capacities of comput-
ing. The manufacturing industries used to necessitate a lot of computing resources
acting as servers in decision-making units and databases. As such, this leads to
wasted investments, minimal productivity, information exchange and unbal-
anced utility of manufacturing elements among others. Cloud computing is
advantageous since it is recommended to remedy networking issues. All the
information can therefore be stored in both public and private cloud services,
6 A. Haldorai et al.

which make sophisticated decision-making process to be supported by vital


cloud computing techniques.
• IoTs and Humans: Associations normally take place between human, things and
things, and things and humans. Various forms of these interactions are composed
of various mechanisms used to support various forms of interactions. Based on
the advancement of IoTs, various forms of interactions can therefore be consid-
ered under a single umbrella. In that regard, users can possibly concentrate on the
various tasks without worrying about the associations. These associations are
responsible for making operations and designs of manufacturing subsystems to
be productive. As for the machine and human association, behaviour of humans
in a virtual surrounding can be determined. Therefore, it is possible to recognize
the behaviours of humans in WSN.
The developments in the IoTs have significantly led to transformative implica-
tion on the community and the environment via massive application segments,
including smart agriculture, smart homes, healthcare and manufacturing. In order to
completely accomplish this, a significant portion of heterogeneous IoTs appliances
is networked to effectively support the actual actuation and monitoring over various
domains, As such, analysts in [4] propose about 50 billion IoTs devices have to be
connected by the end of 2020. Earlier on, the massive portions of data that are gen-
erally referred to as the Big Data were transferred to the cloud by the IoTs devices
to enhance further analysis and processing. Nonetheless, centralized processing in
the cloud cannot be relied when it comes to the analysis of massive IoTs application
due to a number of reasons. One of the reasons is that applications necessitate close
coupling defining the feedback and requests. The second reason is that the delays
witnessed by the centralized cloud-centred deployment are not acceptable for a
number of latency-sensitive applications. Third, there are extreme chances of net-
works failing or data being lost. Lastly, this has resulted in the transition of edge-­
computing remedy.
Irrespective of the fact that the present sources of literature differentiate between
fog and edge computing, this research abstracts all of these cases. Moreover, novel
computing paradigms are also considered as a vital concern of edge layering. The
initiation of edge computing presents these major concerns based on the provision
of computational capability in close proximity to information-generating devices.
The smart edge network devices like the smartphones, Pi and UDOO boards pro-
vide fundamental support to local processing and information storage extensively
but to a smaller segment. Nonetheless, constituent networking devices in edge com-
puting is termed as heterogeneous as they are composed of certain architecture fol-
lowing a given protocol in communication. Different from the cloud where locations
of data centres are fixed, IoTs and edge devices are controlled by batteries, and solar
linked to an eternal power supply compared a cloud storage data centre linked to a
stabilized power supply. To completely take advantage of the goodness of edge
computing, it is vital to comprehend the capabilities and features of IoTs and edge
devices alongside the elements of IoTs data evaluation systems. The diversification
of the underlying edge and IoTs devices, formats communication mediums, f­ ormats,
1 Internet of Things (IoTs) Evolutionary Computation, Enterprise Modelling… 7

functional complexities and programming frameworks enhance the process of net-


working evaluation, which also makes it more time consuming and challenging.
Analysing the systems in an actual surrounding assures the best performance
behaviour; but it is not normally easy to evaluate different frameworks in advance.
Despite the availability of infrastructure, it is fundamental to perform various exper-
imental settings since it is necessary to apply skills based on associated edge and
IoTs devices that are not intuitive. Executing a lot of experiments meant to deter-
mine the correct framework necessities reconfiguration of a number of devices and
transitions to necessitated parameters, which promptly become untenable appli-
ances as a result of volume transitions required. Moreover, executing these experi-
ments in an actual environment is significantly costly because of the maintenance
and setup. Thus, the actual surrounding is considered dynamic, which means that it
is difficult to reproduce much results representing various iterations leading to mis-
interpretation based on evaluation. These issues possibly hinder the application of
actual environments used to benchmark edge computing surroundings. In order to
effectively overcome this problem, a feasible alternative relates to the application of
simulators. These simulators provide novel chances of enhancing the evaluation of
the recommended policies and frameworks in a basic, repeatable and controlled sur-
rounding. A simulated surrounding has to mimic the major heterogeneity and com-
plexity of actual networks that support a lot of multiple cases affecting the
deployment of IoTs.
Contrary to that, the research in [5] gives a performance analysis of the various
protocols governing the application layers of IoTs structure. In that regard, these
researchers provide a comparison of the Hypertext Transfer Protocol (HTTP) and
the Constrained Application Protocol (CoAP) that enhances the process of format-
ting messages, transferring information and requesting users to evaluate various test
beds. There are various deployments of evaluating various performance benchmarks
for IoTs; nonetheless, this is done on the levels of platforms.
To effectively address the initial research concern, various levels and developer
duties in IoTs architecture have to be considered. Figure 1.1 indicates the basic IoTs
architecture (Device, Gateway and Cloud Platform), which has been diminished to
three fundamental layers. In the first layer, we have the constrained controllers and
devices that stand in for the IoTs [6]. Second, there are gateways, smart devices and
routers that enable fog computing in each edge and can also associate as a pre-­
evaluated dataset from networking devices. The third layer represents the platform
store, aggregate and process data retrieved from various sources, allowing enter-
prise application to report and analyse data to the end users. Communication and
connectivity in different levels is thus not designated to a single direction. Moreover,
non-functional necessities, for example security, performance and interoperability,
are considered the key factors influencing the levels of performance. There are key
questions that pop-up for performance engineers. These include the following:
• How can computing devices, such as disk, CPU, network and memory, be
adjusted at every level?
8 A. Haldorai et al.

Fig. 1.1 IoTs Architecture Abstract

• What are the implications of protocols, interoperability and security choices in


reference to general networking performance?
• Do the architecture scale effectively based on the increased number of gateways
and devices?
Considering the fact that IoTs stack renders more diverse, various engineers and
developers are engaged in the process of implementation. Systems and embed-
ment are concerned with the device levels, which are a bit concerned with the
gateway levels. Since gateway levels continue to expand technologically, they con-
tinue to run more advanced operating systems. Moreover, application engineers
and developers have continued to be a major segment of gateway levels. On this
platform level, a mixture of information experts, application analysers and web
developers implement the visualization and integration of information [7]. Based
on the mixed interests on every level and role, it is possible to comprehend the
influencing conditions in a holistic viewpoint, which has to be added to our pro-
spective model of approach.

1.4 Challenges and Contributions of IoTs Technologies

As mentioned earlier, IoTs increases and assures the chances of linking various
approaches, for instance, the approaches requiring capacity architecture for gate-
ways and devices based on a formal model that is acknowledged in the domains of
enterprise data systems. Due to the fact that there are a lot of modelling approaches,
there is need to review the challenges and contributions of technologies used in the
various levels.
1 Internet of Things (IoTs) Evolutionary Computation, Enterprise Modelling… 9

In the device level, for example, the AutoFOCUS31 stands for the associated
model-centred tools used to develop various processes in an embedded framework.
This therefore includes those activities involved in evaluating modelling necessities,
hardware platforms, software structure and deployment, including the generation of
codes. Software architecture is created based on various software components lined
to each other to enhance the interactions broken down into a lot of hierarchical com-
ponents [8]. The architecture defining hardware includes different resources such as
memory and processors that are connected. This also involves the platform structure
applied in the runtime and execution environments like the operating architecture
and Java Virtual Machines. The combination and integration of various models
facilitate developers to effectively apply various synthesis and analysis techniques
like model-evaluation, testing, automatic scheduling and deployment. An Eclipse
Framework representing the distributed enterprise control and automation is a major
segment of the Eclipse IoTs environment that represents a case for gateway level
modelling. This therefore assures an open-source infrastructure defining the distrib-
uted industrial process evaluation and the control framework based on IEC 61499
protocols. To effectively model the software architecture, the 4diac incorporates an
application editor, which is used to represent the functional block network that con-
sists of a single functional block, including its various interactions based on its
events. In the same manner, the separator editor is added to the model’s specifica-
tions of hardware through modelling resources and devices. Through various edi-
tors and their runtime ecosystems, the 4diac permits the use of enterprise IoTs
application, hence allowing interoperability, portability, scalability and configu-
rability based on the promotion of IEC 61499 [9]. In the platform level, various
associated approaches for supporting, integrating and automating engineering ele-
ments in software’s lifetime are controlled using the Performance Management
Work Tool (PMWT). This incorporates an automated generation of models for orga-
nization application with reference to the performance measures, modelling com-
plex individual application behaviours and simulating the overall status of Big Data
applications.
With reference to performance elements, the current remedies based on net-
working devices and gateway levels normally concentrate on assuring safety and
functionality. Contrary to that, there are novel performance models that are used to
analyse and predict the platform level behaviour. The present framework on the
level of architecture assures users of the benefits this research seeks; nonetheless,
we have based on enterprise application involving various necessities. For exam-
ple, the workloads of enterprise applications are categorically user-initiated like
the parallel user accessibilities where IoTs factors such as velocity, volume and
incoming datasets are initiated by data. Moreover, parallel and massively distrib-
uted resource clusters include properties that are normally found in enterprise
applications. In that regard, there is need to combine the present model techniques
and the missing functionalities to enhance the performance of various models of
architecture. Therefore, solvers and simulators can be implemented in deriving
various metrics like throughput, response time and utility of resources. To effec-
tively evaluate our recommended approach, we propose to model an application
10 A. Haldorai et al.

based on IoTs benchmarks to adapt and suit for models in different cases such as
enhancing the IoTs and enhancing resource capacities. The challenges and contri-
butions witnessed include the following.

1.4.1 Challenges

Modelling and simulating a real-life IoTs case is significantly problematic for a


number of reasons. These are as follows:
• Various IoTs devices require combination of the cloud and edge systems to effec-
tively satisfy various application requirements.
• Second, the process of modelling network graphs in various diverse forms of
IoTs and edge devices is challenging.
• Third, control flows and modelling data dependencies over the edge and IoTs
layer to support massing information evaluation and work flow structures are
considerably non-trivial.
• Capacity evaluation over the edge computing segment depends on different con-
figuration parameters such as data volume, upstream, downstream, data speed
and bandwidth.
• Information transfer between IoTs and edge networking device is varied from
cloud data centre communication that is centred on wireless or wired protocols.
Connection between edge and IoTs computing layer, as evaluated in this research,
is diverse. In that case, it is challenging to create an abstract without going
against expressiveness.
• Mobility levels remain to be a fundamental element of IoTs devices since the
sensor embedded in various physical is somewhat mobile. Due to the fact that the
edge devices are considerably limited, the mobility of sensors can possibly lead
to a handoff. Moreover, the information transferred to the edge devices for the
purpose of process cannot be in the present range of IoTs devices. In that regard,
to obtain the processed information, the edge-to-edge communication is neces-
sary. Modelling a handoff or mobility for massive IoTs devices with various
velocities is considerably problematic.
• Dynamicity of the IoTs ecosystem computes to either the removal or addition of
IoTs and edge devices in most cases. This is thus caused by various factors such
as network link failure or device failure. To model the scalability of IoTs devices
using heterogeneous characteristics at a rapid rate is significantly problematic.
• Due to the fact that IoTs ecosystem is an upcoming development, novel applica-
tions should be created in the future. It is vital that simulators permit users to
personalize their frameworks with reference to their networking necessities.
Creating an overall simulator, which permits easy personalization, is consider-
ably problematic.
1 Internet of Things (IoTs) Evolutionary Computation, Enterprise Modelling… 11

Various simulators have also been proposed in past works. These include
GreenCloud, and Cloud that represents the cloud environment. Nonetheless, iFog-
Sim and EdgeCloudSim had been proposed for the purpose of implementation in
edge computing ecosystems. However, there exists no simulator that can potentially
mitigate the challenges outlined above.

1.4.2 Contributions

This research proposes the IoTSim-Edge simulators, which is meant to permit users
to analyse the edge computing cases more easily since this simulator is more cus-
tomizable and configurable in the ecosystem. The IoTSim-Edge is created based on
the simulators that have been proposed in the past. However, the IoTSim-Edge cap-
tures the general behaviour of IoTs and the edge computing planning deployment
and development. Mostly, this model deals out the challenges that have been dis-
cussed. Apart from that, this simulator can be used in analysing the present and
future IoTs applications. Particularly, the model can potentially model the
cases below:
• Novel IoTs application graph modelling abstract that permits practitioners to
explain the information analytic operations and mapping of different infrastruc-
ture segments such as edge and IoTs.
• An abstraction supports modelling of the heterogeneous IoTs protocol with
energy use profiles. This therefore permits practitioners to explain configuration
of IoTs and edge devices alongside certain protocols that support the process of
networking.
• An abstraction permits modelling of cellular IoTs devices. It critically captures
the consequences of handoff as a result of IoTs movement network devices [10].
To effectively maintain a consistent information transfer channel, the simulator
is useful for enhancing an edge communication, permitting the transfer of data to
IoTs devices, that is, edge to edge.

1.5 Edge and IoTs Computing

This part provides the background analysis of the IoTs based on modelling prob-
lems. Moreover, this part also explains the overall architecture of the edge comput-
ing used to model based on the proposed simulator.
12 A. Haldorai et al.

1.5.1 The IoTs Ecosystem

The IoTs are defined as ‘the state-of-the-earth sensors’ that are embedded in
physical things and humans that are surrounding us and linked to the internet
to control and monitor connected IoTs. Various IoTs application advance our
day-to-day lifecycle in various vertical domains such smart healthcare services
and smart homes and in disaster management. The functions of IoTs are initi-
ated by six vital elements such as communication, sensing, identification, com-
putation, semantics and services [11]. The IoTs devices have the capability to
evaluate the ecosystem while separating ubiquitously controlling both the envi-
ronmental and physical surrounding information. Thus, IoTs devices are con-
sidered more identical based on various application techniques and requirements
used to implement various applications. The process of computation is distrib-
uted over various IoTs edges, devices and cloud data centre considering the
desired and functional Quality of Services (QoS) and application parameters.
To attain this, data is transferred from the IoTs devices to the network edge
using various communication protocols. The results of computation can be uti-
lized to initiate more decisive operations to attain the desired application
processes.
Based on a simple sample of the smart homes, which control a lot of devices in
urban environments, it is easier to potentially ease the lives of inhabitants. IoTs
devices include the sensors added to devices such as heaters, refrigerators, cars and
light bulbs, including mobile phones and gateways. The intelligent home systems
utilize the private cloud information centre resources. Home appliances are linked
to the gateways with light weight protocols based on the CoAP protocols enhancing
the transfer of information. Mobile phones are linked to 4G whereas gateways are
linked to Wi-Fi for information transfer to the private cloud [12]. In case the resident
individual leaves for office, the intelligent home systems possibly switch off the
heaters and light bulbs. Moreover, the systems have the capability to determine if
there are things such as foods, in the refrigerator, before sending message to some-
one to collect them.
Modelling actual IoTs applications necessitates the linking of different
actuators, sensors and edge devices on a massive scale with various operating
ecosystems. This is a difficult task because of the heterogeneous features of
IoTs and the edge devices requiring progressive optimization for resource allo-
cation, migration, provision and fault tolerance in processing various applica-
tions [13]. Moreover, the process of implementation is onefold, providing a
more generic framework for IoTs application in case the application necessi-
tates level abstraction. The key problems based on modelling application, IoTs
mobility, networking protocols and the consumption of energy are evalu-
ated below.
1 Internet of Things (IoTs) Evolutionary Computation, Enterprise Modelling… 13

1.6 Application Composition

Generally, the IoTs is made up of a series of tasks executed using some detected
data. This can be shown using a number of techniques, although the chapter
adheres to IoTs applications in a clear and (Dag) direct acyclic graph for the
MELs (microelements). Every MEL is regarded to be an abstract component
for each application, which stands for every assets, information and services
exclusively creating the microdata, micro-service, micro-computing and micro-
actuation. Therefore the modelling of every application so that they can be like
the DAG for the MEL can be very difficult; hence, there is need for encapsulat-
ing a number of features altogether. Additionally, the MEL sequences have a
crucial role since they are used for representing the flow of data within an
abstract phase.

1.6.1 Communication Protocols

Within the IoTs surroundings, various messages and network connection protocols
normally play crucial part during the communication process. Considering the
genetic elements within the IoTs environment, there are various complex network
linkage amidst various IoTs environment features. Centred on the range, certain
limitations, device type and other protocols to be used within the IoTs network, the
edged gadgets can be used for transferring information. There is need for mobile
devices to leverage certain protocols unlike the static ones.
Modelling protocols using the application graph can be difficult. Moreover,
some messaging processes can be accessible in order to ease the process of transfer-
ring the data from the sensors to every edged gadget and later to cloud servers.
There are certain protocols that have been set for this objective. Distributing infor-
mation with the aid such protocols may influence the execution processes of the
systems in dissimilar ways. Therefore a single message protocol being not in a posi-
tion of satisfying various requirements of the IoTs complex can also be used. For
these reasons, it is necessary for every device to link up every protocol for every
gadget and dissimilar layers to be used [14]. The modelling of such events using a
handshaking amid these protocols can be difficult since the movement of the IoTs
devices using the IoTs devices can be embedded using cars or rather smart phones
that aid the various users in a flexible manner. Knowing the fact that the edged
devices are normally transfixed, every IoTs device is allowed to go from one range
edged device to a different resulting in a handoff. In some cases the handoff may be
audible as soft or hard based on the velocity of the IoTs gadgets and the signal
strength of each edge device. Therefore simulating the mobility in an adverse man-
ner, there is need for incorporation of various features such as the IoTs devices
14 A. Haldorai et al.

velocity and acceleration, the pathway of the motion, topological maps, edge range
intersection, etc. Taking into account all these factor, the real mobility of the simula-
tion can be a hard role since there will be huge data number points that comprise of
extensively reliant data one the relationship and characteristics [15]. Additionally,
the transfer of information can be unsuccessful at any moment if the IoTs devices
may be moving from a certain location to a different one basically due to poor signal.

1.7 Battery Drainage

Much IoTs drainage is driven by the battery, which is normally a restriction and
hence should be recharged at various intervals. Hence it is crucial that the devices
should be in a position of holding off the battery for an increased time mostly when
the applications cannot be very easy recharging, for example, the sensors within
rivers or places of havoc. Holding the battery off for an increased period of time will
highly save much time, and this is crucial almost for every application. Therefore
transferring information at dissimilar intervals and utilizing various communica-
tions protocol.

1.7.1 Architectural Computing for IoTs Edged Devices

As shown in Fig. 1.1, the architecture of the IoTs edge has been shown in detail.
Moreover the IoTs technology comprises of a number of components such as the
sensing nodes and also actuator nodes. The sensing nodes are responsible for the
collection of data within the sensors and transmit the data whereby it will be required
to be processed and stored. The actuators will therefore be centred on the data anal-
ysis [16]. Thus the subsequent layer will be for the edged infrastructure that com-
prises of different forms of edged devices, for example, the arduino and raspberry
Pi. Such gadgets may be easily accessed openly with the aid of various forms of
containerization and virtualization techniques. These mechanisms provide suffi-
cient infrastructure that is used for the deployment of various raw data created
through various sensor nodes. For many instances, if the edge can be in a position
of creating more data, then there is no need of transmitting the information via the
cloud for processing. Lastly, its outcome is later transmitted to its actuator so that a
certain action can be executed. Therefore the services and the application layers
comprise of dissimilar services which can be accessed to various clients. Later the
applications can be acquired via a subscription channel. Some of the examples for
these services include smart city and smart home. Every MEL is normally distrib-
uted within an edge or rather the cloud information centre. Additionally, the pro-
gram helps in managing every QoS necessity within the prospective load and any
error handling. Moreover, it delivers services, for example management of resources,
device management and storage management.
1 Internet of Things (IoTs) Evolutionary Computation, Enterprise Modelling… 15

Service issued by such a layer normally ascertains that the QoS needs have been
successfully met. The present application distribution and scaling methods created
for the distributed environmental algorithms, for example, the cloud or grid, cannot
be effective when employing the new IoTs environment. The main reason is there
are extensive features and characteristics used for the smart gadgets aside the mobil-
ity elements and the latest applications technology which is made of a limited reli-
ance and hence needs process distribution. Based on the type of application, there
has been collaboration amidst the IoTs, cloud and the edge that are required for
attaining a prospective QoS desires. Therefore, creating new applications deploy-
ment alongside its scaling methods is very crucial. Moreover, it is important to
analyse and test such methods prior implementing the actual deployment.
Nevertheless, analysing these kinds of techniques require a real environment incor-
porating a set of conditions at every moment which is expensive and time consum-
ing. Additionally, based on the alignment of ownership of various gadgets,
assessment needs a number of mechanisms that may make something become much
difficult. Hence, various simulation frameworks, for example, the IoTSim simula-
tion that aids the distribution of these applications that assess the execution pro-
cesses various methods, certain events within certain conditions may be necessitated
[17]. Additionally, assessing these methods in certain circumstances may be exe-
cuted at low cost within a simulation environment.

1.7.2 The IoTs Sim Edge Architecture

Based on the architecture for the prospected simulator, it shows a series of layers. A
concise analysis of its distinctive components has been highlighted in this section.
IoTs Sim edge is structured within clouds simulation equipment. The cloudsim
delivers an underlying system that is used for assisting some basic communication
processes within various subscribed components that use the event management
mechanisms. Hence the primary components used for the cloudsim have been
extended so that they can stand in for the edge infrastructure within a line compris-
ing edged characteristics and features. The IoTs asset layers comprise dissimilar
forms of IoTs gadgets in which they have their own elements and characters along-
side execution processes used for actuation and sensing [18].

1.7.3 Implementation and Design

When the IoTSim edge is enacted, the missing gap amidst the cloudSim has to be
extended alongside vast numbers of classes so that the model can be shown in an
edged and IoTs environment. A certain entity that increases the SimEntity class may
seamlessly transmit and acquire various events for its partners via the event man-
agement engine. Moreover, this research exhibits a basic trait for the simulators, a
16 A. Haldorai et al.

cycle that comprises of various actions and procedures. Therefore, to model the
edge infrastructure, there was need for the designation of vast number of classes
within the network. Major classes include the edgebroker, edgedevice, edgedata-
centre, microelement, edgedatacentrecharacteristics and the edgelet. For instance,
the edgedatacenter’s role is used for creating a link between the IoTs devices and
edge centred on some provided IoTs protocols, and the process is accompanied by
asset edge provisioning scheduling rules and assessment of edge processes.
Moreover, the system is also responsible in the creation of the edgedevice, submis-
sion of the edgelet, establishing of a network amidst the IoTs infrastructure, assess-
ing network accessibility, etc. These particular characteristics for the edgedatacentre
are always fed using the edgedatacenter characteristics class [19].
For this reason, the edgebroker’s role is acting in place of the clients based on
establishing a connection from the IoTs devices and the edge devices, submission of
the edge and IoTs requests, and acquiring various outcomes. This class outlines a
model for the edged gadgets specifically when acquiring and processing informa-
tion from the IoTs in which the data processing is executed following a specific
edgelet rules and policies. On the other hand, the MEL class is used for modelling
the abstract processes that later performs within the IoTs data for each edged side,
or every cloud information centre. Considering the present implementation for each
MEL, only the running of the edged devices will be prioritized. Hence the edgelet
model class is required to be performed within the MEL. In some situations the edge
device could consist a battery such as a mobile phone, hence going around, thus, the
mobility and battery class will be designated to activate the edged device so that it
can extract its constituent characteristics. Therefore, the Moving Policy class pre-
scribes the conditions and characters for each edge device [20]. At the same IoTs
infrastructure can be modelled, vast number of new classes have been designed.
Regarding speed rate such as Wi-Fi, the Network Protocol is a recognized com-
munication system where it can transfer network packets at a velocity of 200 Mbps,
and the 4G LTE will be at a speed of 150 Mbps. Hence the frequency for the edgelet
in different terms, the delay period that will be required to transmit the edgelet to the
datacentre that is normally accessed from the network policy class. On the other
hand, the class models for every feature within the IoTs protocol in relation to the
battery and QoS consumption rate has been mentioned [21]. Since each IoTs pro-
cess comprises of dissimilar processing methods, everything is structured in a man-
ner that they may have dissimilar energy consumption rate. Much more detailed
explanation based on the classes has been given in the following sections.
1. EdgeDataCenter Class
This class is responsible for controlling various core edge infrastructures. At
every stage counter-intercept various incoming calls and processes and execute dis-
similar protocols centred on the payload of a particular event, for example asset
delivering and handing over various edglet requirements to its desired MELs. When
the battery of every edged device is discharged, the edgedatacenter will detach auto-
matically and send various unserved requests to its subsequent edged devices.
1 Internet of Things (IoTs) Evolutionary Computation, Enterprise Modelling… 17

2. EdgeBroker class
In this class, there are some clients’ proxy, whereby various requests are created
focusing on the prescribed necessities. This class comprises of a range of roles for
executing, for example, the submission of edge and provisioning of the MEL
requests to every datacentre, getting the IoTs devices to create and transmit data to
its prospective edged devices and acquiring its final processing outcome. The class
aids various power aware models for every IoTs gadgets. Since the edge broker
consistently assesses the consumption of every battery for the IoTs gadgets, imme-
diately there will be a disconnection from the nearby IoTs devices.
3. EdgeDevice class
The class is assumed to be the same as actual edge devices. Its main role is host-
ing a number of MEL and simplifying the protocols of the CPU sharing appliances
through a specified CPU sharing rule. Moreover, they are further linked to have a
certain number for each IoTs device that transmits its data for processing purposes.
Focusing on the present version of the simulator, a set of four classes has been used
to extend the IoTs devices class which include lightsensor, voicesensors, carsensors
and temperature sensors. Therefore, an improvised form of IoTs device may easily
extend the IoTs device class and hence enact a new set of features. Within an actual
IoTs environment, each IoTs gadget harbours an IoTs and network protocol; hence,
the class comprises of same features when utilizing the networkprotocol and
IoTProtocol classes as analysed in subsequent phases in this chapter.
4. MEL class
The role of this class is representing a single component using the IoTs applica-
tion graph. It basically represents the core processing necessities used by the appli-
cation components. Centred on the application requirements, setedgeoperation
techniques may be configured. Hence the reliance amidst certain components may
be shown by using an uplink and downlink that can be easily structured in order to
portray some possible difficult applications [22]. This can help in configuring which
stands in for any difficult IoTs applications centred on the clients necessities.
5. The EdgeLet class
The class models every IoTs created data and its subsequent MEL processing
information. At any moment that the IoTs gadget created a unique link using its
respective edged gadgets, there will be an immediate response that will generate
IoTs sensed data that will be required in the form of the EdgeLet which comprises
of a payload. The role of a payload is encapsulating the correct data so that it may
direct the MEL at the processing podium, for example, the size of each data set cre-
ated, the route followed by the MEL ID, and the IoTs device ID [23]. When using
the size of the EdgeLet data size, the delay of every network will be needed to trans-
mit the EdgeLet to its required MEL and hence can be calculated considering the
network transmission frequency.
18 A. Haldorai et al.

6. Mobility class
The class is made up of a model of IoTs and alongside various edge devices. Its
mobility has a crucial role within the IoTs edge system in which upholding actual
time consistency is crucial so that the performance processes can be executed. Its
prospective attributes may comprise of velocity, location, range and the time inter-
val. Every attribute taking off time comprises of some distinctive values so that they
may represent the vertical and horizontal direction for every edged IoTs devices.
Moreover, this chapter portrays some possible examples of the way various attri-
butes have been utilized using the edge datacentre so that the location can be
reviewed by the edge and IoTs devices. On the other hand, algorithm 1 portrays
pseudo-code used for simple tracking methods enacted within the EdgeDataCenter
class. Various clients may easily increase every class so that they can implement
their mobility models.
7. The Battery class
This class is used for modelling the battery features for the portable IoTs and the
edge devices. Through using the transmission rate modelling, the time used for
transmitting every data may be achieved considering the size of each EdgeLet.
8. Policies classes
The main functions of these classes are modelling the policies in three distinct
groups, which include: battery consumption, device movements and network trans-
mission. Therefore, the device movement rules may instruct edge datacenters so
that there can be tracking of the location and the movement of both the edge and
IoTs devices. The consumption of the battery rules normally calculates and tracks
the subsequent power capabilities for each edged and IoTs devices. The transmis-
sion of the network policy calculates the duration of time taken so that the data can
be distributed from one IoTs device to another edged device [24]. These kinds of
policies may be lengthened so that a new IoTs edged solution and strategy can be
derived. Both the classes may be extended so that various client policies can be
extended and implemented.
9. User Interface Class
This class normally provides the required techniques that can aid effective con-
figuration for each IoTs applications not knowing other data based on the simula-
tors. Moreover, it permits the user explaining every parameter while using associative
interfaces that can be converted to a required configuration file. The content below
shows the data in the user interface.
ioTDeviceEntities" : [
"

{
"
mobilityEntity" : {
"
movable" : false,
"
location" : {
x : 0.0,
" "
1 Internet of Things (IoTs) Evolutionary Computation, Enterprise Modelling… 19

y : 0.0,
" "

z : 0.0
" "

}
},
"
assignmentId" : 1,
"
ioTClassName" : "org. edge. core. iot. TemperatureSensor",
"
iotType" : "environmental",
"
name" : "temperature",
"
data _ frequency" : 1.0,
"
dataGenerationTime" : 1.0,
"
complexityOfDataPackage" : 1,
"
networkModelEntity" : {
"
networkType" : "wifi",
"
communicationProtocol" : "xmpp"
},
"
max _ battery _ capacity" : 100.0,
"
battery _ drainage _ rate" : 1.0,
"
processingAbility" : 1.0,
"
numberofEntity" : 5
}]

1.7.4 Computation and Event Processing

Various simulation processes occur on time for initializing the proper IoTs edge
structure that has been extracted for a certain file configuration. Moreover this
research exhibits the configuration test for every file within the IkT gadgets. After
the creation of the MELs and the IoTs gadgets, every edge broker will request data
from its centre so that effective connections can be established amidst the IoTs
devices and hence its respective MELs.
At any moment that the edge broker gets the new established connection (ACK),
there will be an immediate notification from the IoTs devices based on the connec-
tion. Moreover, the IoTs devices can commence creating data as edgelets, distribut-
ing the edgelets to its final MELs and reducing the battery power based on the rate
of drainage. Hence the IoTs devices can maintain the execution processes (Fig. 1.2).

1.8 Case Studies: Sim Assessment for the IoT

In order to assess the usefulness of the IoTs sim edge, three scholarly studies have
been carried out. Explanations of each study has been identified in the below case.
20 A. Haldorai et al.

1 1.. 1..
EdgeBroker EdgeDataCenter
AMQP MQTT XMPP CoAP 1

1.. 1..
1 1 1 1
1 1 Battery
loTProtocol
1 1 1 1
loTDevice Mobility EdgeDevice
1 1
NetworkProtocol
1 1 1 1
1 MovingPolicy
1
1 1
NetworkPolicy 1..
1.. 1.. 1
EdgeLet MEL

VoiceSensor LightSensor TemperatureSensor CarSensor

Fig. 1.2 Class diagrams of the IoTSim-Edge model

1.8.1 Case One: Healthcare Sector

Considering various works that involve activity information has been segmented
into various micro-applications, there were also some realizations about the similar
circumstances. So that the raw information utilizing the phase counts calculation,
every micro-operation is represented to be as the MEL which can be undertaken by
the unforeseen IoTs gadgets. There is a basic MEL graph portraying the one neces-
sary for deploying the edge within computing vicinity. For finding out the pareto-
customized deployment answers within a deployed IoTs infrastructure, it is required
to evaluating dissimilar deployment strategies for the matter to be effective. Due to
the fact that there has been limited reliance within the MEL, it is crucial considering
such a factor during the deployment process. In order to enact the main simulator,
two-edged gadgets can be embedded within the IoTs devices. For these reasons, the
devices create information centred on some elaborated data frequency.
Scientific evidence is evidence that serves to support the rate at which batteries
drain that the rate of transfer is always greater compared to its algorithm. For the
fact the edged devices powered by a battery, the process that happens within the
device edge, then the rate at which every battery drains is handled in this chapter,
although executing much more performance processes close to the IoTs gadgets can
augment the performance time that is crucial in many instances. Additionally, vari-
ous outcomes have shown and analysed overall performance duration in dissimilar
scenarios. Centred on actual scenarios, the MLs have been allocated dissimilar
MELs. Moreover, this research evaluates a quick rational about battery hours that
comprise of edged processes which demand a shrinking element. Hence various
shrinking features have occurred within the edged gadget.
Hence the result performance rate for every edged devices was 90% that was
embedded which results to saving much more battery power unlike conveying raw
information to the rest of the devices. Therefore the outcome for every processing
time is 90% for each execution of the device which maximizes the utilization of the
1 Internet of Things (IoTs) Evolutionary Computation, Enterprise Modelling… 21

battery by 266% unlike executing 10% operation. For the fact that the edge E1 com-
prises of a minimal processing and harbouring power, every process may not be
explained in this section. Dissimilar forms of evaluation may be executed by using
various scenarios. Clients may additionally suggest dissimilar computations so that
an effective deployment strategy can be found to focus on quite a number of objec-
tives found.

1.8.2 Case Two: Smart Building

Previously, smart building systems automatically caused the heating, lighting, air-­
quality, air-conditioning, etc., which have gained much focus. Various forms of
sensors that transmit at certain intervals and specific sites have the capacity of
sending information to various linked devices that may process and analyse vari-
ous data sets. Therefore edged gadgets conduct common processes and transmit
information to the cloud and when this happens there is increased storage or com-
plex analysis necessitated. Various IoTs devices transmit its data to edged devices
complying with a common communication process. Various elements such as
latency imminently rely on the rate of the data and size of the packet [25].

1.8.3  ase Three: Capacity Planning Units for Roadside Self


C
Driving Cars

Some of the incoming advancement whereby every vehicle can be embedded sen-
sors that transmit data to be RSU controllers which consider the consistency of the
traffic and maintain the safety of the road. Moreover, the RSUs may provide an
excellent platform for the edged devices to transmit the data acquired from a spe-
cific vehicle and may create a runtime decision. The car ranges alongside the RSUs
are restricted, and hence every vehicle is directed at a certain speed to a certain point
of the road. Therefore the area covered by the RSU solely relies on the protocol
transmission. Probably, the car’s connection diminishes alongside the linked RSU at
a point of time. At some moment the handoff can be difficult following the previ-
ously mentioned RSU so that various decisions can be made, while an RSU-RSU
information delivery will also be created.
Based on the point of focus of the car, an RSU may convey a message to a differ-
ent RSU that can be later directed to the final car. Because the Gap and the processor
capability of the RSU are always restricted, evaluating the count of vehicle data to
be processed is possible by making sure that there is no data loss. Additionally, it is
essential to assure that the requirements for the QkS such as the application time
response are available. The scenario was effectively structured using a simulator. At
first there is an establishment of a brand new link for the RSU referred to as RSU1,
22 A. Haldorai et al.

which commences transferring information. On the other hand, at time t3, a certain
point is reached whereby the range of both RSUs is attained. Relying on its mobil-
ity, a connection can be created by RSU2 but is still transmitted to RSU1. Regarding
its motion, the connection can be made, hence data will be sent to RSU2. Moreover,
RSU1 finds out that every car has its range and can be centred on the data being
transferred as exhibited by time t4. Such an event stipulated that these mobility
features are cooperative when accessing days pertaining every simulator.
On the other hand, precise configuration of each simulation has been considered
in this research specifically. These edged devices are unique except the location
which includes E1: 0, 0, 0 and E2: 50, 0, 0. Based on this fact, the outcome high-
lighted can precisely prove the idea [26]. Because the edged devices number are
always constant and the processing manner is timely, various requests may congest
prior processing, resulting in a greater performance processing time. Furthermore,
when a certain vehicle goes further away from the RSU, therefore the RSU need to
transmit the data to the other RSU whereby the data is later transmitted to the other
cars. Because such comes among the processing time, hence performance period
will be great. Therefore the greater the number of various requests for a certain care,
then the edge can take up great energy as assessed by these simulation that portrays
the averagely energy utilization rate for the edged devices as they increase from one
number to the other within the IoTs network.

1.9 Related Sources of Literature

If there is much interest in the computing edge of the IoTs, vast simulation equip-
ment would have been created for previous years. Various tools can be extended
considering the present network and the loud simulator, although there exist some
gap within the simulator and actual modelling edge and the vicinity of the IoTs.
This phase analyses the present simulation equipment used within the network, IoTs
environment, cloud and the way they cannot model the present IoTs edged environ-
ment. Moreover, it is shown that the simulation structure is in a position of meeting
the expectation of the accessible limitations using holistic methods.

1.9.1 Network Simulators

Some tools used for the simulation process have been recommended and hence
preferred to be used within the computer networking system for the last years.
Considering all this, this section analyses some common recognized network simu-
lation equipment. These tools include the C++ and the OMNeT++ focusing on the
discretion of various events of the environment alongside the communication net-
work. The system aids a parallel simulation network and a real-life implementation
protocol for the simulation models. Additionally, various network processes can be
1 Internet of Things (IoTs) Evolutionary Computation, Enterprise Modelling… 23

supported although cannot harbour the edge communication protocols [29].


According to scientific analysis the OMNeT++ system was created so that there can
be consistent simulation body area network and some accompanying lower powered
gadgets. Furthermore, the simulation can be endured within the network with an
extensive number being dynamically driven. By increasing the energy support for
every model different protocols for the communication process can be used. An
example of a source-driven simulation includes the TOSSIM, which is used within
the WSN (wireless sensor network). Therefore, the energy usage may not use any
mobility modelling and its consumption. In order to aid energy consumption rate
within the WSN, a different simulator has been suggested, referred to as the pow-
erTOSSIM, which is achieved by extending the TOSSIM. Therefore, the pow-
erTOSSIM takes into account various nonlinear characters for the battery
simulation model.
Furthermore, the mobility elements have not been mentioned. Whereas the NS-3
will a different C++ model for a simulation whereby it is made up of the python
interface. Such equipment is used for the simulation that normally distributes visual
aid to the neighbouring environment. The NS-3 will not be an effective IoTs simula-
tion within an edged level as they do not provide support for the application and
scheduling of various features. For the cloud simulation, majority of them have been
recommended to be used as the best cloud computing tool which provides excellent
outcomes [27]. Amidst them, the cloud sim has been regarded to be prominent sim-
ulator so far within the research sector. Various clouds’ sim that are event driven and
that aid the system and behavioural modelling within the cloud environment have
been mentioned. Nevertheless, the IoTsim edge has been created to aid extensive
scale simulations whereby there can be support for various communication systems
and also physical modelling. Such simulators comprise of customized international
hypervisor that can aid cloud brokering rules. Generally it may comprise of the
Amazon public clouds scenarios and aid the MPI. Additionally the network cloud-
sim can be a different simulation that permits such models within the network that
comprises of cloud information centres [28].
In these cases, such simulators may not be in a position of aiding the IoTs and the
edge simulators. A different cloud simulator includes the green cloud that can be
increased using the NS2 simulation equipment. On the other hand, greencloud is
made up of packet-level equipment which is used for establishing the amount of
power usage for the components within the data centre. The main functions of the
simulator are computing the energy used in order to ascertain that there is energy
awareness placement. Moreover, there is no support of the edge and the IoTs simu-
lation. The DCSim is a different cloud simulation that entirely reflects on the IaaS
cloud vicinity simulation. The main role is aiding the simulation and modelling of
various data centres, the VM and the host using restricted application number and
management of asset policies. Conversely, the IoTs simulation and the edged envi-
ronment cannot be aided using this type of simulator. Much simulation equipment
has been recommended for the simulation within an edged environment. The
SimIoT is known to be simulation equipment that effectively models the
24 A. Haldorai et al.

c­ ommunication amidst the cloud data centres and the IoTs devices, although they do
not take into account various edged device within the simulation [30].
Generally, not all simulators support the edge information transfer protocols and
the battery power. Moreover, most of these simulators are not capable of defining
the application composition in the IoTs environment. These characteristics are vital
for use in any IoTs application. The composite simulation ecosystem of the Edge
and IoTs is vital to researchers and enterprises since it helps them to gain actual
control of edge processing.

1.10 Conclusion and Future Work

In conclusion, this chapter pursues and recommends a model-centred approach used


to evaluate and predict the performances of IoTs systems and architectures. The
IoTSim-Edge model shall potentially allow engineers and developers to evaluate
various designing choices in systems, in advance. Moreover, the model will effec-
tively locate possible bottlenecks, size the resources and predict any potential feed-
backs to visualized results. The future prospects are bases of works that integrate
and combine various model approaches used to embed systems with Big Data
frameworks and enterprise data systems. In that regard, there is need to formulate a
prototype that will be used to intergrade modelling ecosystems. The present manu-
facturing ecosystems have been evaluated to effectively analyse the vital necessities
of ESs based on modern organizations.

References

1. Anandakumar, H., & Umamaheswari, K. (March 2017). Supervised machine learning tech-
niques in cognitive radio networks during cooperative spectrum handovers. Cluster Computing,
20(2), 1505–1515.
2. Abbas, N., Nasser, Y., & Ahmad, K. (2015). Recent advances on artificial intelligence and learn-
ing techniques in cognitive radio networks. EURASIP Journal on Wireless Communications
and Networking, 2015(1), Available: https://doi.org/10.1186/s13638-015-0381-7.
3. Belkin, M., & Niyogi, P. (2004). Semi-supervised learning on Riemannian manifolds. Machine
Learning, 56(1–3), 209–239. Available: https://doi.org/10.1023/b:mach.0000033120.25363
.1e.
4. Benhamou, E. (2019). Similarities between policy gradient methods (PGM) in reinforcement
learning (RL) and supervised learning (SL). SSRN Electronic Journal. Available: https://doi.
org/10.2139/ssrn.3391216.
5. Cecin, F., Barbosa J., & Geyer, C. (2006). A communication optimization for conservative
interactive simulators. IEEE Communications Letters, 10(9), 686–688, Available: https://doi.
org/10.1109/lcomm.2006.1714546.
6. Chang, X., Mi, X., & Muppala, J. (2013). Performance evaluation of artificial intelligence
algorithms for virtual network embedding. Engineering Applications of Artificial Intelligence,
26(10), 2540–2550, Available: https://doi.org/10.1016/j.engappai.2013.07.007.
1 Internet of Things (IoTs) Evolutionary Computation, Enterprise Modelling… 25

7. Anandakumar, H., & Umamaheswari, K. (October 2018). A bio-inspired swarm intelligence


technique for social aware cognitive radio handovers. Computers & Electrical Engineering,
71, 925–937. https://doi.org/10.1016/j.compeleceng.2017.09.016.
8. Farhat, S. (2018). Resource sharing in 5G multi-operator wireless network. International
Journal of Digital Information and Wireless Communications, 8(3), 156–161, Available:
https://doi.org/10.17781/p002431.
9. Fu, Y., Wang, S., Wang, C., Hong, X., & McLaughlin, S. (2018). Artificial intelligence to man-
age network traffic of 5G wireless networks. IEEE Network, 32(6), 58–64, Available: https://
doi.org/10.1109/mnet.2018.1800115.
10. Gueguen, C., Ezzaouia, M., & Yassin, M. (2016). Inter-cellular scheduler for 5G wire-
less networks. Physical Communication, 18, 113–212. Available: https://doi.org/10.1016/j.
phycom.2015.10.005.
11. Gueguen, C., Ezzaouia, M., & Yassin, M. (2016). Inter-cellular scheduler for 5G wire-
less networks, Physical Communication, 18, 113–124, Available: https://doi.org/10.1016/j.
phycom.2015.10.005.
12. Haldorai, & Kandaswamy, U. (2019). Supervised machine learning techniques in intelligent
network handovers. EAI/Springer Innovations in Communication and Computing, 135–154.
https://doi.org/10.1007/978-3-030-15416-5_7.
13. Huang, R. P., Zhou, & Zhang, L. (2014). A LDA-based approach for semi-supervised docu-
ment clustering. International Journal of Machine Learning and Computing, 313–318.
Available: https://doi.org/10.7763/ijmlc.2014.v4.430.
14. Iosifidis, A. (2015). Extreme learning machine based supervised subspace learning.
Neurocomputing, 167, 158–164. Available: https://doi.org/10.1016/j.neucom.2015.04.083.
15. Kansal, P., & Shankhwar, A. (2017). FBMC vs OFDM waveform contenders for 5G wire-
less communication system. Wireless Engineering and Technology, 08(04), 59–70. Available:
https://doi.org/10.4236/wet.2017.84005.
16. Kaur, K. S., Kumar, & Baliyan, A. (2018). 5G: A new era of wireless communication.
International Journal of Information Technology. Available: https://doi.org/10.1007/
s41870-018-0197-x.
17. Ke, C., & Shihai, W. (2011). Semi-supervised learning via regularized boosting working on
multiple semi-supervised assumptions. IEEE Transactions on Pattern Analysis and Machine
Intelligence, 33(1), 129–143, Available: https://doi.org/10.1109/tpami.2010.92.
18. Khanale, P., & Chitnis, S. (2011). Handwritten Devanagari character recognition using arti-
ficial neural network. Journal of Artificial Intelligence, 4(1), 55–62. Available: https://doi.
org/10.3923/jai.2011.55.62.
19. Khanale, P. (2010). Recognition of Marathi numerals using artificial neural network. Journal
of Artificial Intelligence, 3(3), 135–140. Available: https://doi.org/10.3923/jai.2010.135.140.
20. Ma, J. Y., Wen, & Yang, L. (2018). Lagrangian supervised and semi-supervised extreme
learning machine. Applied Intelligence, 49(2), 303–318, Available: https://doi.org/10.1007/
s10489-018-1273-4.
21. Manasa, H. R., & Pramila, S. (2015). Architecture and technology components for 5G mobile
and wireless communication. International Journal of Engineering Research and, 4(06),
Available: https://doi.org/10.17577/ijertv4is061112.
22. Mythili, K., & Anandakumar, H. (December 2013). Trust management approach for
secure and privacy data access in cloud computing. In 2013 International Conference on
Green Computing, Communication and Conservation of Energy (ICGCE), doi:https://doi.
org/10.1109/icgce.2013.6823567.
23. Nastase, V., & Strube, M. (2013). Transforming Wikipedia into a large scale multilingual
concept network. Artificial Intelligence, 194, 62–85. Available: https://doi.org/10.1016/j.
artint.2012.06.008.
24. Nishii, R. (2007). Supervised image classification based on statistical machine learning. SPIE
Newsroom, Available: https://doi.org/10.1117/2.1200612.0449.
26 A. Haldorai et al.

25. Sądel, B., & Śnieżyński, B. (2017). Online supervised learning approach for machine schedul-
ing. Schedae Informaticae, 12016, Available: https://doi.org/10.4467/20838476si.16.013.6194.
26. Shang, Z. (2005). Application of artificial intelligence CFD based on neural network in vapor–
water two-phase flow. Engineering Applications of Artificial Intelligence, 18(6), 663–671.
Available: https://doi.org/10.1016/j.engappai.2005.01.007.
27. Singh, P. (2016). Importance of wireless sensor network and artificial intelligence for safety
prerequisite in mines. International Journal of Engineering and Computer Science. Available:
https://doi.org/10.18535/ijecs/v5i11.35.
28. Singh, P. (2016). Safety of underground coal mine using artificial intelligence and wireless
sensor network. International Journal of Engineering and Computer Science. Available:
https://doi.org/10.18535/ijecs/v5i11.38.
29. Vishnoi, R., Gaur, S., & Verma, A. (2018). Comparison of various waveform contenders of
5G wireless communication based on OFDM. International Journal of Trend in Scientific
Research and Development, 2(4), 2468–2473, Available: https://doi.org/10.31142/ijtsrd15637.
30. Zhao, H. (2018). Semi-supervised extreme learning machine using L1-graph. International
Journal of Performability Engineering. Available: https://doi.org/10.23940/ijpe.18.04.
p2.603610.
Chapter 2
Organization Internet of Things (IoTs):
Supervised, Unsupervised,
and Reinforcement Learning

A. Haldorai , A. Ramu , and M. Suriya

2.1 Introduction

Internet of Things (IoTs) links up various data sensing appliances via the Internet,
which is meant to realize smart management and identification. Different intelligent
sensors are a vital building block used in establishing Internet of Things (IoTs)-
based business process applications. In this case, the business process management
(BPM) framework is not only relevant in enhancing the proficiency of collaboration
in cross-sensing organization procedures, but it is also relevant in assisting to invoke
effective management competencies before an emergency snowballs into disaster,
i.e., traffic jams, fire hazards, or networking failures. Due to these diversities in
sensing and the variations of their different functions, the direct models of the IoTs-­
centered organization processes and applications are significantly difficult.
Developers require using more coding practices, which bring out the aspect of
ignoring organizational logic orchestration. Perhaps, the fundamentals of the IoTs
applications before organizational procedure automations represent the creation of
organizational procedure models. These models are computed based on all forms of
organizational procedure obligations that signify novel functional units that have
been implemented by various services. The BPM frameworks shall benefit from
representing modeling when fundamental sensor devices can assume the position of
sensor resources for personal organizational procedures.

A. Haldorai (*)
Sri Eshwar College of Engineering, Coimbatore, Tamil Nadu, India
A. Ramu
Presidency University, Bangalore, India
M. Suriya
KPR Institute of Engineering and Technology, Coimbatore, Tamil Nadu, India

© Springer Nature Switzerland AG 2020 27


A. Haldorai et al. (eds.), Business Intelligence for Enterprise Internet of Things,
EAI/Springer Innovations in Communication and Computing,
https://doi.org/10.1007/978-3-030-44407-5_2
28 A. Haldorai et al.

The Internet of Things (IoTs) is confirmed as the distributed and interlinked


network of embedded frameworks that communicate via wireless or wired com-
munication advancements [1]. As such, this has been confirmed to be the network
of physical things and objects that have been empowered with limited storage,
communication, and computation capacities, which include embedded electronics
like actuators and sensors, network connectivity, and software that are vital for
objects used to gather process and transfer information. The IoTs represent the
objects realized from various life scenarios that include intelligent devices kept in
buildings. These devices include smart adapters, smart bulbs, smart refrigerators,
smart meters, and smart detectors. Other more sophisticated devices include the
radio-frequency identification devices (RFID), accelerometers, heartbeat detec-
tors, and sensors located in parking lots, among others. There exist a lot of services
and applications provided by the Internet of Things (IoTs) that range from novel
infrastructure to military, agriculture, personal healthcare, and home appliances.
Moreover, the domains dealt by the Internet of Things (IoTs) services include med-
ical, energy, retail, building management, manufacturing, and transportation,
among others.
A massive scale of the Internet of Things (IoTs) network utilizes novel
approaches like the management of various devices such as the absolute scale of
information, communication, storage, security, privacy, and computation. Up to
this moment, a significant research analysis has been conducted concerning the
various elements of IoTs such as communication, architecture, applications,
protocols, privacy, and security. Nonetheless, the foundation of IoTs commer-
cialization and its technological advancements are centered on privacy and secu-
rity guarantees, which include user satisfaction. The idea of IoTs using the
enhanced technologies like the software-­defined networks (SDNs), fog comput-
ing (FC), and cloud computing (CC) means that landscaping the threats for
attackers has advanced. The generation of information based on IoTs devices is
significant, and therefore, the ancient information collection processes and stor-
age techniques are not operated under this scale. Moreover, the absolute infor-
mation set can be utilized based on behavioral controls, patterns, assessments,
and prediction techniques. Moreover, the data heterogeneity produced by the
Internet of Things (IoTs) formulates more fronts proposing the present informa-
tion processing techniques [2]. As such, to effectively harness the worthiness of
information retrieved through the IoTs, there is a need for the implementation of
novel techniques. Resultantly, it is considerable to conclude that machine learn-
ing (ML) is a vital computation paradigm used to provide embedded smart
devices based on IoTs.
The idea of ML is fundamental for intelligent devices and machines to inject
vital competency from humans and device-generated data. Moreover, this idea
can be referred to as the competency of intelligent devices to automate and
adapt to various behaviors or situations based on knowledge, and this is viewed
as the fundamental element of implementing IoTs solutions. Thus, the ML
technologies have been utilized in obligations like regression, classification,
and estimation of density. Different applications like fraud detection, comput-
2 Organization Internet of Things (IoTs): Supervised, Unsupervised… 29

erized visions, malware detection, and speech recognition make use of ML


technologies and algorithms. Similarly, these technologies are used to leverage
the idea of IoTs in availing smart services to users.

2.2  adio-Frequency Identification Devices of Internet


R
of Things (IoTs) Features

2.2.1 Internet of Things (IoTs) Network Features

In this section, we shall evaluate the novel features of Internet of Things (IoTs)
networks. These features include the following:
Heterogeneity: The Internet of Things (IoTs) networks include a lot of devices
that are equipped with various capacities, information transfer rules, and features of
communication that relate to one another. To put this into simple language, the
devices can utilize various forms of information transfer techniques, including the
various communication paradigms like Ethernet and cellular paradigms. Moreover,
the variable constraints concerning the hardware resources are also some of the
techniques considered.
Large-scale deployment: It is considered that thousands of devices linked to one
another via the Internet shall surpass the capacities of the present Internet condi-
tions. The large-scale deployment of Internet of Things (IoTs) also proposes signifi-
cant issues. These issues include the designing of networks and their storage
architectures for intelligent devices, information transfer protocols, and effective
data sets. The effective data sets communication rules, proactive protection, and
identification of Internet of Things (IoTs) are fundamental for securing against
malicious attacks, technological standardization, and management of application
interfaces.
Inter-connection: The Internet of Things (IoTs) is purposed to be linked with
international data and communication infrastructure, which can be retrieved from
any location and at any moment. This form of connection is dependent on various
applications and services that are produced by IoTs service providers. In some
moments, the connection can be termed as local while in other cases it is considered
to be international. The local cases include the instances of interlinked motor tech-
nologies and swarm sensing [3]. On the other hand, the international cases will
include the intelligent home accessibility via the mobility infrastructure and crucial
management of infrastructure.
Information transfer in close proximities: One of the most outstanding features
of the IoTs is information transfer in close proximities that has to eliminate the
centralized standards such as base stations. The device-to-device (D2D) informa-
tion transfer influences the features of on-point information transfer, for instance,
the dedicated short-range communication (DSRC) and many other technological
advancements that can be rated the same. The architectural segment of the ancient
30 A. Haldorai et al.

Internet is considered to be inclined more on network-centric information transfer


as compared to how communication presently has been delinked to complement the
services of the IoTs spectrum.
The ultra-reliable and low-latency communication (URLLC): The URLLC is
considered as another feature of the IoTs networks necessitated in novel actual-time
applications and services such as organizational procedural automations, smart traf-
fic, remote surgery, and smart transportation frameworks. The vital performance
limitations include reliability and delay aspects.
Inexpensive and less-energy information transfer: More connectivity linking the
IoTs devices necessitates ultra-less-energy and inexpensive remedies that facilitate
smooth operation of networks in the modern age. Individual-organization and
individual-­healing features are necessary for contemporary and urgent IoTs infor-
mation transfer, which include disaster and emergency conditions [4]. In these
cases, the reliance of networking infrastructure is never considered an option,
which means that individual organization of networks has to be considered. The
vibrant transitions in networking are composed of a significant number of devices,
which required management in the most effective manner. All these devices have
the capability to act vibrantly. For example, the wakeup or sleep duration of devices
shall be dependent on various applications, including the time when these devices
utilize the Internet.
All these features have to be included in IoT networks. One of the vital features
among them considers the safety of networks. Among these features, however,
safety is a vital element that enhances the smooth operation of IoTs networks. The
safety of networks is considered to be significant not only for the devices but also
for the consumers. This is due to the fact that IoTs devices linked to the Internet can
potentially be tampered as individual information has been shared with different
devices. Moreover, the security and privacy of these devices is a fundamental factor
that has to be considered in networking.
Smart networking is also an intriguing feature of Internet of Things (IoTs), which
enhances informed and timely decisions that have to be executed in organizations.
The information generated by IoTs devices has to be created in a manner that facili-
tates the performance of actions to effectively make decisions enabling the process-
ing of information [5].
The intelligent urban environments utilize modern data and communication
technologies to analyze and integrate the information gathered from the vital
frameworks that navigate through the urban environments. At the moment, the
intelligent cities are capable of executing smart responses to different cases such
as weather forecasting, traffic controls, and organizational and economic activities.
An intelligent urban environment with its traffic-routing system entails a massive
number of cameras that are meant to monitor the road networks and the intelligent
algorithms that navigate across the cloud networks proposing the optimum routes
used for individuals. Moreover, the intelligent motor navigation framework per-
mits individuals to change and set up destinations through the inbuilt audio appli-
ances. The two frameworks in pairs are used to provide actual-time interactive
routing systems. Nonetheless, the individuals’ voice commands can be translated
2 Organization Internet of Things (IoTs): Supervised, Unsupervised… 31

into motor edges and their sides before they are transferred into cloud systems
where intelligent r­outing frameworks operate. The most vital route can be trans-
lated to the voices that guide individuals to their preferred destinations. The appli-
cations that have been mentioned above are utilized in different computing
resources such as edge, IoT devices, and cloud services, which include modern
language techniques that enhance developments of the various ML-centered IoTs
applications [6]. These advancements are challenging for modern language models
and IoTs frameworks. To deal with the available gap, this research has orchestrated
the enhancement lifecycle of the modern language-centered IoTs applications. In
the subsection outlined below, we have included an analysis of the enhancement
lifecycle alongside the detailed taxonomy, which surveys the techniques needed in
the enhancement of applications.

2.2.2 Privacy Issue of Internet of Things (IoTs) Deployment

Privacy and security issues are the vital factors that enhance commercial realiza-
tions of Internet of Things (IoTs) applications and services. The present Internet
setting is an attractive segment for privacy concerns that are relevant for a few tasks,
which includes the corporate levels and well-linked privacy breaches that have a
significant effect to various organizations, businesses, and healthcare facilities. The
confines of the Internet of Things (IoTs) environments and devices operate in a
manner that possesses more issues in terms of security for various devices and
applications. Until now, privacy and security concerns have been analyzed exten-
sively in IoTs domains from various aspects like data security, information transfer
security, architectural security, privacy, malware analyses, and identity management.

2.2.3 Existing Gaps in Privacy Resolutions in IoTs Networks

To effectively realize the fundamentals of IoTs, analysis of its privacy and security
concerns is relevant. Most significantly, IoT has been retrieved from the present
technologies, which makes it possible for users to identify its challenges. As such,
it is possible to embrace new or old technologies that have been in existence over
the past few decades. Past research analyses focused on the differences and similari-
ties of the privacy concerns in IoTs, with reference to ancient IT devices. Moreover,
these past evaluations settled on the security issues. The fundamental driving factors
that provide the basis of the differences and similarities include networking, appli-
cations, hardware, and software. With reference to these applications, there are
major similarities between privacy concerns in ancient IoTs and IT domains.
Nonetheless, the major concern of IoTs remains to be the resource constraints that
limit the adaptations of the novel privacy remedies of IoTs networks. Moreover, the
remedies to the privacy and security concerns in IoTs necessitate the cross layers,
32 A. Haldorai et al.

the designs, and the optimized algorithms [7]. For example, because of the
­computation constraints, the IoTs devices require unique samples of optimized
cryptographic and other algorithms to effectively cope up with the privacy and secu-
rity concerns. Apart from that, a number of IoT devices create significant challenges
for the privacy technologies.
A lot of privacy concerns are more complex since the remedies obtained cannot
be considered discrete. For example, as for the privacy issues like intrusions and
DDoS, there is evidence of the probability of the untrue positives that will provide
remedies that are not effective on the attacks. Moreover, these positives can poten-
tially diminish the trust of consumers hence reducing the efficiency of the potential
remedies. Nonetheless, the holistic privacy and security approaches over the IoTs
will render nomination from the available present security resolutions, including the
advancements of novel intelligence, evolution, and robust and scalable technologies
used to mitigate the privacy concerns in IoTs.

2.2.4  achine Learning (ML) – The Remedy to IoTs Privacy


M
Concerns

The concept of machine learning (ML) represents the smart methodologies utilized
to fully optimize the performance standards that make use of the present sample
data and those recorded through the present learning experience. More significantly,
ML algorithms propose the model of conditions that makes use of the mathematical
methods on massive information sets. ML is vital in enhancing the performance of
smart devices, in terms of learning, without being programmed explicitly. All these
competencies are utilized to enhance the predictions useful in the future based on
novel data input. ML has been considered interdisciplinary in nature since its roots
can be retrieved from a number of disciplines such as engineering and science con-
sidering artificial intelligence, data theory, optimization theory, and cognitive sci-
ence, among others. ML is also used when human competencies do not use or exist
based on navigation hostility where users do not have the competencies to utilize
their expertise, for example, speech recognition and robotics. These competencies
are also used in cases where the remedies to certain issues change anytime, such as
routing in PC networks or locating malicious codes in applications and software.
Moreover, this aspect is practically used in smart systems; for example, Google
makes use of ML in evaluating the risks against application and mobility endpoints
in operating Android systems. As such, this is vital for the identification and removal
of malware from all the infected devices.
In the same way, the Amazon Company introduced its service Macie, which
makes use of machine learning to categorize and sort out information that has been
stored in cloud storage services. Despite the fact that machine learning typically
performs well in most sectors, there are normally some real negatives and untrue
positives that have been recorded in the past. In that case, ML methods require
2 Organization Internet of Things (IoTs): Supervised, Unsupervised… 33

modification and guidance to effectively model the most accurate predictions.


Contrary to that, deep learning (DL), which is considered to be a novel option of
ML, is a model used to evaluate the accuracy of all the possible predictions. Due to
the condition of self-nature of the DL model, users consider it as a more precise
framework used in prediction and categorization tasks to enhance the innovation of
IoTs application based on personalized and contextual assistance. Despite the fact
that ancient approaches have been utilized extensively in various segments of IoTs
such as protocols, services, architecture, application, resource allocation, data
aggregation, analytics, and clustering, huge-scale deployment of IoTs advocates for
smart, reliable, and robust networking techniques. Until now, DL and ML are con-
sidered to be the novel techniques for IoTs networks because of a number of rea-
sons, i.e., IoTs networks propose an absolute amount of information that are
necessitated by DL and ML methodologies to enhance smart competencies into
systems [8]. Moreover, the effectiveness of information produced by IoTs is effi-
ciently used with DL and ML methods that enhance IoTs systems to enable more
smart and informed decisions. DL and ML are massively applied in privacy, secu-
rity, and malware evaluation and during the detection of potential attacks. The DL
method can also be utilized in IoTs devices to perform all the complex recognition
and sensing tasks used to enhance the realization of novel services and applications
and to determine the actual-time interactions between humans, the physical envi-
ronment, and the intelligent device. Some of the privacy-related actual-world appli-
cations of ML are considered as follows:
• Forensics face recognition: Such as posing, occlusion, lighting, make-ups, and
hairstyling.
• Feature recognition for privacy recognition: Includes the various handwriting
competencies.
• Malicious code identification techniques: These competencies are used to iden-
tify the dangerous codes in software and applications.
• Distributed denial of service (DDoS) detection: This form of detection technique
is used in identifying DDoS attacks that influence infrastructure via behavioral
evaluation. Based on the application of DL and ML methods of IoTs application,
users have realized other novel challenges. All these challenges are considered
multifaceted [9].
For example, it is a pending concern to effectively formulate the best model that
will be utilized in processing information from various IoTs applications. In the
same way, labeling inputs information in a precise way is considered to be one of
the most difficult obligations. The second challenge is evident in the utility of mini-
mal labeling information in the process of learning. The third challenge is realized
in the deployment of the models on resource-centered IoTs devices where there is a
need to diminish storage or processing overhead. In the same way, novel infrastruc-
ture and actual-time applications do not have the capability to withstand anomalies
formed due to the DL and ML algorithms. Thus, it is considerable to systematically
analyze the privacy remedies of IoTs, which leverage the DL and ML.
34 A. Haldorai et al.

2.3 The Present Literature Surveys

Presently, IoTs is extensively considered, based on existing literature sources,


since a lot of research has been done concerning the various aspects of IoTs
privacy. In that regard, this section provides a brief overview of the present sur-
veys before comparing these analyses with our research. According to this
research, surveys have minimally focused on ML methods utilized in IoTs.
Moreover, the present surveys either are based on applications or do not com-
pletely consider the complete spectrum of privacy and security concerns in IoTs
networks. The present literature surveys deal with the privacy concerns in IoTs
through the analysis of the available ancient remedies based on the emergent
technological initiatives. Nevertheless, surveys discussing the DL- or
ML-centered remedies are still not available. Moreover, DL and ML have been
considered in a number of research analyses, but the general data concerning the
usage of these two methods is still scarce. As for the present gap in research, we
have done an analysis of the detailed survey of the DL and ML methods, which
are utilized in IoTs privacy.

2.3.1 Survey Scope and Research Contributions

In this chapter, we have done a detailed systematic analysis of DL- and


ML-centered privacy remedies of IoTs. Initially, we had evaluated the privacy
requirements of applications in IoTs networks against threats and attacks.
Thereafter, we focused on the obligation of DL and ML in IoTs, which includes
the analysis of the various DL and ML methods used to leverage the IoTs services
and applications. To entirely focus on the practical element of IoTs, we have done
an extensive evaluation of DL- and ML-centered privacy remedies in IoTs. This
evaluation also includes the analysis of the present challenges in literature and
future concerns that focus on the analyses of DL and ML for IoTs networks. The
main purpose for this is filling the available gaps between the necessities of IoTs
privacy and the capacities of DL and ML that will permit the process of address-
ing the privacy concerns on IoTs networks. Thus, we have practically evaluated
the IoTs from the perspective of DL and ML [10]. This evaluation focuses on
privacy and security concerns of the IoTs networks. Typically, this chapter
includes a detailed evaluation of the security problems and risk frameworks in
IoTs. The analysis encompasses the privacy requirements and threat surfaces of
IoTs, which includes the analysis of DL- and ML-centered remedies in mitigat-
ing the possible privacy attacks in IoTs networks. As such, it is considerable to
argue out that the value of the present surveys is still considered to date. Moreover,
the research analysis includes the recently done works in the fields of DL and ML
for IoTs technologies.
2 Organization Internet of Things (IoTs): Supervised, Unsupervised… 35

2.4 Security Problems and Risk Frameworks in IoTs

Essentially, IoTs utilizes the transformational approach to effectively provide users


with a lot of services and applications. This form of pervasive deployment of an
extensive number of network devices fundamentally advances the degree of threat
surface. Moreover, the idea that IoTs devices are normally resource-based is unfea-
sible to utilize novel privacy techniques to mitigate infamous threats. Moreover, it is
critical to detail that the initial Internet had not been created for the IoTs. As such,
it is significant to avail IoTs privacy to the present security techniques. Until the
present day, IoTs makes use of various communication competencies like ZigBee,
IPv6, Bluetooth, 6LoWPAN, Wi-Fi, Z-Wave, and near-field communication (NFC)
[11]. All these communication technological advancements possess their challenges
and shortcomings that have been borrowed from IoTs domains. Moreover, other
problems include that of IP- and TCP-centered information transfer, which is prone
to issues like complexity, scalability, configuration, addressing techniques, and the
use of insufficient resources that pose limitations on how heterogeneous and diverse
networks are applied on IoTs. Until now, various alternative technological advance-
ments like data-centric networks and the software-defined networks (SDNs) are
applicable in serving the dominant information transfer infrastructure for IoTs. In
this manner, this research provides a brief evaluation of the attacks and threats that
IoTs is facing. Eliminating the losses of generality, privacy concerns of IoTs can
categorically be divided into various attacks as evaluated below. These threats
include the following:

2.4.1 Physical Threats

When evaluating the physical threats, the intruders are provided with a direct acces-
sibility to potentially manipulate various aspects of networking devices. To fully
access all the physical devices, social engineers are required to propose the most
effective methods that the attackers will utilize to access the network devices or
undertake an actual-time attack ranging from physical damages on networking
devices to side-channeling, eavesdropping, and many other potential attacks.
Irrespective of the various technologies being used at the physical segment of IoTs,
the condition of physical threats principally is similar to the requirements of social
engineering and their potential approaches. Moreover, to effectively launch physical
threat, attackers have to avail themselves in the close proximities of hardware and
other devices with various intentions such as facilitating physical damage of hard-
ware, limiting the devices’ lifetime, endangering information transfer techniques,
and influencing the energy resources.
Physical attacks can be considered as a foundation for other potential threats
such as tampering with alarms in homes, which lead to burglary and other serious
damages in intelligent home environments. In the same way, the replacement of
36 A. Haldorai et al.

sensors with any vulnerable sensors can potentially result to the leakage of sensitive
information. The incorporation of vulnerable nodes into networking devices can
potentially lead to any attacks hence opening doors for other intruders to take advan-
tage of the privileges and facilitating a dangerous attack. Moreover, this form of
tampering with the security of devices can be a major facilitator of potential attacks
proposing changes in security keys and routing tables that have an effect in com-
munication to the upper degree. Other possible physical threats include jamming
radio-frequencies that deny the aspect of information transfer in IoTs environments.
Adding to the many other consequences, jamming leads to the denial of consumer
services for the IoTs hence tampering with the operation of IoTs applications. It has
therefore been considered that the intruders make use of various social engineering
techniques to pose physical accessibility to the devices and hardware due to a num-
ber of reasons, which include the potential attacks mentioned above. With social
engineering, intruders can possibly manipulate users to embrace physical control of
the networking devices.
In IoTs environments, various issues can be created using the techniques
embraced in this research. All these applications range from networking to systems,
intelligent urban environments, and intelligent grids, among others [12]. Focusing
on modeling, it is fundamental for consumers of technologies to make use of effec-
tive learning concepts during the initial stages. The most fundamental selection
method can be divided into:
• Power-use-centered selection
• Functional-centered selection
As for the functional-centered selection method, users are given the chance to
select the best concept with reference to functional variations. For instance, RL
advantages from the iterative ecological perspective include agent interactive prop-
erties that require interaction with the ecology and can be used in different applica-
tions and smart systems such as intelligent temperature control frameworks and
cold issues. The TML algorithm is effective for the purposes of modeling structural
information, based on the highest degree of semantic character, mostly when inter-
pretability is needed [13]. The DL-based models are normally utilized to frame out
the complex unstructured information such as audios, time-series data, and images.
They are the best selection criteria since big data is considered with minimal require-
ment concerning interpretability.
On the other hand, the power-use-centered selection method focuses on selecting
the best model that proposes constraints in computation latency and power. Contrary
to TML, DL and RL are typically categorized as extreme computational expenses
since they normally compose the complex networking structures; hence, the accu-
racy determinants seem to exceed the TML based on computational overhead costs.
With reference to TML at its inference stage, ideal accuracy can be attained with the
most effective characteristics such as extreme-level attributes obtained from fea-
tures of engineering.
1. Link Layer and Physical Privacy Problems
2 Organization Internet of Things (IoTs): Supervised, Unsupervised… 37

IoTs includes different information transfer technologies at a minimal


dimension of IP and TCP protocol stacks and therefore gives more complex
heterogeneous network. These networking advancements include WSN,
ZigBee, Wi-Fi, MANET, NFC, and RFID. Moreover, these technologies pos-
sess their own privacy problems. This part focuses on the privacy problems in
data and physical link layers of IoTs. According to literature gaps identified in
this article, privacy problems in IoTs at various layers and their remedies have
been evaluated. As analyzed before, heterogeneity has been proposed at its
physical layers in IoTs, which further proposes various amendments in infor-
mation links. For example, key channel designs will depend on the prevalent
physical layer technologies [14]. Until now, security technologies of IoTs have
to incorporate the heterogeneity at its data and physical link layers. There are
various security concerns in physical layers of IoTs depending on the prevalent
technologies; for example, in the case of sensor nodes, the attacks on these
nodes have to be mitigated. Moreover, the identification of any malfunctions in
hardware is a fundamental element that has to be handled with care to eliminate
any form of anomaly in the upper network layer. Another privacy problem is
intrusion and this requires effective countermeasures made from prevention
and detection standpoints. It is significant to consider that there exist a lot of
attack vectors seen in intrusions at an upper segment, for example, during the
routing threats.
Based on fault detection standpoints, it is vital to identify any fault nodes in
IoTs since they critically influence the quality of services (QoS) of the IoTs appli-
cations. The core purpose of IoTs is to present a ground for a minimal-energy-
constrained thing (device) that co-exists at the minimal layer that assures the same
information transfer for heterogeneous devices. Until now, IEEE introduced a
guideline referred to as the IEEE 802.15, which allows a constrained device to
transfer information in the best manner. In IoTs, the high layers utilize minimal
power standards such as the Constrained Application Protocol (CoAP) and
6LoWPAN, since this is the mechanism required at a minimal layer that enables
the guidelines to operate seamlessly. In this case, IEEE 802.15 gives the required
changes at a minimal layer that represent the recommended standards. It is funda-
mental to consider that IEEE 802.15.4 includes the security concerns defining the
information link layers.
Information link or MAC layers are obliged for enhancing channel accessibil-
ity for various devices in addition to access management, framework validation,
security, and time management. In this research, we have focused on privacy
problems in the high layer given by IEEE 802.15.4. These securities given by the
standards at MAC layers ensure that the levels of the nodes and information trans-
fer are kept secure. Moreover, these securities ensure that the securities in the
upper layer are maintained. In that case, symmetric cryptographic algorithms like
the AES are recommended to be implemented efficiently and rapidly on chips;
hence, all these implementations in IEEE 802.15.4 hardware shall be considered
in the lower layer privacy. The IEEE 802.15.4 standard provides the AES algo-
rithms and various applications that are recommended to control the resource con-
38 A. Haldorai et al.

straints of the networking devices. The guideline supports various privacy nodes
at the link layers; for example, information might not be encrypted by just execut-
ing an integrity check.

2.4.2 Networking Layer Threats

In the networking level, the threats are focused on routing, traffic evaluation, man-­
in-­the-middle, and spoofing attacks. Apart from that, Sybil influences are poten-
tially seen in the layers of the network where false Sybil identities are utilized in
creating illusions in networking. Intrusions via various means propose a manner in
which the intruders of the networking system can potentially launch some possible
threats. As such, ensuring the safety of networks is fundamental to deal with the
threats at their initial stages. At the networking layers, the intruders have possibly
leveraged the vulnerable nodes to perceive them as fake-forwarding nodes facilitat-
ing the formation of sink holes.
This form of threat normally is linked with the mobile ad hoc and sensor net-
works, which pose a significant effect on IoTs environment. With all these threats,
there is a chance for launching an associated DDoS threat that will affect the entire
IoTs network. In the networking layer, the intruders can potentially affect the net-
work by bombarding it with a lot of traffic that comes from the compromised nodes
beyond what their network can possibly handle. Affecting the IoTs nodes and tam-
pering with the identities will pose a significant effect on networks since there are
fake nodes that allow these intruders to prepare Sybil threats where the Sybil nodes
provide an illusion to the major networks as though the actual nodes were transfer-
ring the data sets [15]. To draw the general assumption of the threat vectors in the
networking layer, users should focus on the information transfer aspects of IoTs and
utilize the resource constraints, authorization frameworks, and sophisticated
authentication.

2.4.3 Transportation Layer Threats

The transportation layers are tasked with the obligation to enhance step-by-step
delivery of transportation standards that enhance the procedure followed during the
exchange of information. In this case, the ancient transportation layer privacy con-
cern typically persists. The most serious threat in this layer is the denial of service,
which affects the network applications. Moreover, it is critical to note that due to the
status of IoTs, UDP and TCP standards have no scale with the resource-constrained
device, and hence, the lightweight version of transportation guidelines had been
recommended in research. Nonetheless, the privacy problem of the standards is of a
major concern that effectively enhances the DDoS and DoS threats in IoTs.
2 Organization Internet of Things (IoTs): Supervised, Unsupervised… 39

2.4.4 Application Layer Threats

The IoTs applications are considerably the best target for the intruders since attacks
on the application layer are comparatively easier to launch. The most known threats
include buffer-overflow threats, denial of service, malware threats, phishing, cryp-
tographic threats, exploitative web app threats, man-in-the-middle, and side-­channel
threats. The buffer-overflow attacks are considered to be the widely utilized threat
vector in various applications [16]. The present methods used to deal with the
buffer-­overflow attacks include dynamic nodes and static nodes analyses, which
include other complex techniques such as symbolic debugging techniques.
Nonetheless, the methods cannot be utilized with IoTs because of resource scarcity.
The IoTs applications are also affected by malicious code incorporation due to
buffer-­overflow attacks, in addition to vulnerabilities like cross-site scripting, SQL
injection, and object referencing, among others.
According to the Open Web Application Security Project (OWASP), there are a
lot of susceptibilities that lead to various threats on network applications. One of the
latest susceptibility that OWASP recorded was in 2017. The susceptibility produces
a collection of a lot of threats, which can be launched by attackers at any time. For
example, attackers can choose to inject malicious nodes, access controls, perform
phishing, and tamper with privilege escalation. Moreover, using malicious code
injection, the intruders can possibly collect sensitive data, tamper with the data sets,
and do a great deal of malicious activities. Botnets pose another fundamental threat
to the IoTs application and infrastructure. Controlling the threats produced via
smart botnets poses a significant problem for the IoTs due to the fact that these bot-
nets effectively crawl and scan the networks searching for unknown vulnerabilities
before exploiting them to enhance the launching of massive DDoS. It is significant
to mention that because of the scarcity of resources, the modern cryptographic stan-
dards are unfeasible for IoTs devices that expose them to potential intruders who
launch the cryptographic attacks. Generally, these threats are seen on the applica-
tion layers of the IoTs infrastructure, which makes it much costly to control.

2.4.5 Multiple-Layer Threats

The multiple-layer threats add to the aforementioned threats in this chapter and are
launched to the IoTs infrastructure. These threats include side-channel, traffic eval-
uation, man-in-the-middle, relay, and standard threats. A lot of these threats have
been evaluated in the section above. The traffic evaluation threat is considered as an
attack where the intruders control the traffic and make use of it. Users find it diffi-
cult to control this threat since communication parties normally have no knowledge
that their networks are being monitored. The intruders are searching for vital data
sets in the Internet traffic; the data include personal details, company logic informa-
tion, and credentials, among others. The idea of information transfer privacy is of
40 A. Haldorai et al.

great significance to IoTs. The information that is produced in the IoTs ecology is
utilized for the purpose of decision-making. Hence, it is fundamental to ensure the
health and quality of information. Tampering with the privacy of information in
IoTs will pose a significant challenge on the vulnerable applications. It is also fun-
damental to mention that SDN has been significantly leveraged to attain a wide
speculation of advantages from the IoTs security and applications [17]. The novel
functionalities given by the SDN control plans enhance companies to significantly
control a lot of things and sensors in the IoTs paradigm. Nonetheless, irrespective of
the SDN merits that have been mentioned above, the available interface in SDN
provides risk to threats on the vulnerable IoTs infrastructure, applications, and
devices. In that regard, the privacy concern of the IoTs is dependent of the privacy
of the SDNs.

2.5 Machine Learning and the IoT Safeguarding

This section discusses about different machine algorithms alongside their useful-
ness within the IoT applications. First, the section shall commence with the study of
machine learning which can be segmented into four major sections, which include
supervised, semi-supervised, unsupervised, and reinforcement machine learning.
Supervised machine learning: This type of learning is executed if accurate tar-
geted points have been prospected to be obtained from a specific input set. Therefore,
based on this system of learning, there is labeling of the data, which comes first, and
then the training of the marked data sets. These types of sets voluntarily rule from
the availed sets of data and furthermore give meaning to different types of classes
and in the end foresee the elements that fall under a prospective class. On the other
hand, supervised learning algorithms may be utilized if data X aligns itself with data
Y, which are provided for the training whose main objective is to adjust to a map-
ping function which is Y: ← f (X). There has been extensive application of super-
vised learning algorithms within the IoT sector; for these reasons, there should be
an introduction of various classifiers as shown in this chapter. Both logistic regres-
sion and perception can be regarded to be the easiest linear classifiers. Hence, for
the two frameworks, they can be understood to be the simplest linear transforma-
tions. Whereas perception may execute binary categorization centered on the signal
that the input data is being transitioned, RL can increase the rate of the transforma-
tion to a specific probable value, prior to getting a threshold function, which will be
applied when making the classification decision [18]. Moreover, the RL may also be
increased to multi-class classification events with the aid of the softmax function,
which is a scaling function that has class-related likelihood to the primary output.
The artificial neural networks (ANN) can simply be as previously mentioned linear
classifiers. In contrast to the perception or rather the RL that is a linear-related proj-
ect input data which is linked to the output, the ANN comprises extra “secretive
layers” that permits the ANN for modeling non-linearity.
2 Organization Internet of Things (IoTs): Supervised, Unsupervised… 41

However, in relation to the linear classifiers, it is the additional secretive layer


that leads it to become rarely known to see any linked association amid the output
and input data sets. However, in theoretical basis, having one hidden layer within
the ANN model will lead to the modeling of non-linear functions, in relation to the
restricted capacities during the encounter with unforeseen data sets. Moreover, the
ANN that has more layers will be known to be a deep neural network, which may
seem to have improved modeling capacities as mentioned in the following sections.
Unsupervised machine learning: Regarding unsupervised learning, the vicinity
normally issues the available inputs deprived of the targets desired. However, there
is no need for a labeled data set which can analyze the uniqueness amid the unla-
beled data and categorize the data into dissimilar sections. The supervised and unsu-
pervised learning tactics specifically aim at the evaluation of different data set
challenges, whereas reinforcement learning can be utilized for decision-making
processes and various comparisons. Therefore the grouping and choice selection of
the ML methods relies on the manner at which the data can be accessible. If the
input data and prospected output are known, supervised learning can be employed.
During these scenarios, the entire framework comprises only trainees so that the
input can be mapped appropriately and a stipulated input can be achieved. Regression
and classification are various instances of the supervised techniques whereby regres-
sion operates consistently while classification works alongside discrete outputs.
There are different regression methods such as supported vector regression (SVR),
polynomial regression, and linear regression, which are the mostly utilized meth-
ods. Subsequently, the grouping operates with various discrete output figures.
This is mostly used in instances for grouping various algorithms such as the
k-nearest neighbor, SVM, and logistic regression. Other algorithms may be utilized
for the grouping and classification, for example, the neural networks. Outputs have
not been properly defined and clustered with various classifier objects centered on
known criteria for example k-clustering [19]. The extent of accuracy for the fore-
seen analytics is based on how enough the ML methods have utilized recent data for
creating new models and the way they have been used for predicting various mod-
els. Various algorithms, for example, the SVR, the naive Bayes, and the neural net-
works, have been utilized for predictive frameworks.

2.5.1 The Main Objective

The main objective of the unsupervised learning algorithms is to enhance the


understanding of the web relationship amid information that comprises only X
data, which is only present when the class Y is not available. For instance, algo-
rithm clustering may be utilized in order to identify various capable series with
others that have not been named and acquired outcomes that can be utilized for
further evaluation. Therefore, k implies to be the core component analysis, which
falls among the recognized unsupervised algorithms. With k, it means that the main
objective is identifying a group series amid information through assigning various
42 A. Haldorai et al.

samples of clusters centered on the range existing between the centroid and the
samples based on every cluster. On the other hand, the PCA can be dimensionally
utilized as a reduction that can relate with various raw elements prior to choosing
ones that are informative.
The semi-supervised learning: Comparing the initially mentioned groups, there
will be no tags that exist for the observation of the data sets or rather the labels,
which have presented all notable observations. Considering the vast number of
practical scenarios, the expenses used in labeling can be very high; hence, it needs
equipped and talented human resource to be employed. In reinforcement learning
(RL), having no accurate results can be explained, and every agent can understand
from the response once it has interacted carefully with the neighboring surround-
ings. For this reason, it executes various actions that lead to the implementation of
various decisions based on the reward acquired [20]. Therefore, for an agent which
has been rewarded for the execution of recommendable actions or castigation of
poor actions and utilize feedback methods so that they can increase long-term rec-
ompenses. This has been immensely inspired through the learning of various char-
acter traits of animals and people. These behaviors have established a pretty
approach within the dynamic use of bots whereby different systems may learn to
accomplish various roles by not using explicit programming. It is crucial that people
should select an excellent reward so that the triumph and catastrophe of these agents
may rely on various gathered rewards.
For this segment, there will be an introduction of various techniques that can be
utilized in formulation of the previously mentioned streaming video example that
comprises RL. Considering the previously mentioned results, RL, which is an agent
that associates itself with the neighboring environment, to get a customized control
policy, can be gained through experience. Therefore, there is a need for involvement
of three basic elements, which include action, reward, and observation. Centered on
the mentioned elements, formulation of the adjustable bitrate streaming challenges
can be easily possible. Uniquely, various observations may buffer various occupan-
cies, the network throughput, and so on. For each phase, every agent may decide on
the bitrate of the incoming chunk. Hence, a reward can be acquired once an agent
receives the response. The algorithm proposes, gathers, and then simplifies the out-
comes used for executing the initial decisions; hence, it can optimize various poli-
cies based on various network coverages. The RL based algorithms may lead to
environmental based noises such as unforeseen networking situations, video assets
etc. Various DL architectural features that can be accessible in literature may include
the CNN (convolutional neural networks), BM (Boltzmann machine), DBN (deep
belief networks), RNN (recurrent neural networks), FDN (feed-forward deep net-
works), GAN (generative adversarial networks), and LSTM (long-short-term mem-
ory). For the CNN and the RNN, they are extensively utilized for deep architectural
learning [21].
Deep reinforcement learning: In this segment the DL can be a type of ML method
that is utilized for estimating the function, grouping, and foreseeing while the RL
will be a different type of ML technique which can be employed in decision-making
processes whereby every software may be educated pertaining to how various
2 Organization Internet of Things (IoTs): Supervised, Unsupervised… 43

o­ ptimal actions can be achieved though the interaction with the nearby environment
with some states. Both the RL and the DL chip in within these circumstances and
the data is dimensionally huge while the environment can be non-stationary. Hence,
conventional RL will not be sufficient. Through the combination of both the DL and
the RL, various agents may learn on their own and draft effective policies in order
to extract maximum substantial rewards. Considering this approach, the RL receives
aid from the DL in order to allocate an excellent policy while the DL executes vari-
ous action estimations so that it can get the quality of actions within a prospective
state. Moreover, the RLL and the DL gain advantages from one another [22]. The
DL is in a position of understanding the more complicated series although it is
exposed to various inaccurate groupings. For this situation, the RL comprises a
powerful capacity for automatically understanding the environment by not con-
structing and aiding the DL during effective grouping.

2.5.2 IoT Security Machine Learning Methods

In this section, there will be a discussion of some ML algorithms that are specifi-
cally aimed at underlying safety and confidential challenges within the IoT net-
working system. Accurately, there will be consideration of various authentic
invasion techniques and other mitigation practices, DDoS threats, intrusion sensing,
anomaly, and some malware evaluations. The supervised learning algorithms oper-
ate using tagged information and use an IoT network within spectrum sensing,
adaptive cleaning, channel approximation, and localization challenges. Such a sec-
tion comprises two speculated forms of methods, which include regression and
grouping. Grouping within supervised machine learning can be utilized in foresee-
ing and also in modeling accessible sets of data. On the other hand, regression can
be used in predicting consistent patterns of variables. Some few widely employed
classifications of algorithms include the naïve Bayes, decision tree, SVM, and ran-
dom forest [23]. Therefore, the SVM employs a framework known as kernel, which
is utilized in identifying various disparities between two common points that com-
prise different classes.
The SVMs are in a position of modeling the non-linear sections. The SVM can
genetically include the memory and hence be hard deciding the effective kernel,
there by modeling large sets of data becomes difficult. Hence, a random forest is
usually recommended unlike the SVM. On the other hand, the naïve Bayes (NB) has
been using Dover time in helping solve global problems such as the spam sensing
and classification of text. Considering having all naïve and other input elements,
every other random forest provides an ideal environment for the modeling of actual
problems. With random forest algorithms, it will be much easier to implement and
adjust huge-sized data sets. Such algorithms may take a much longer period of prac-
ticing compared to the supervised algorithms, which include the NB and the
SVM. However, they can attain a greater accuracy and may consume less amount of
time in foreseeing various issues. Moreover, they have been centered on the
44 A. Haldorai et al.

s­ tructuring of various graphs that segment into leafs and branches, hence showing
the class and decision that have been implemented.

2.6  n Analysis of Present Machine Learning Centered


A
on Solutions Within the IoT Security

This section provides a quick survey of the present ML-centered solutions that high-
light dissimilar safety problems within the IoT. Authentication and access within the
IoT system has been a primary requirement for users within the IoT. Every user
should be authenticated so that every application or service can be utilized effec-
tively. Normally, these applications can focus on the exchange of data within differ-
ent conditions. Such data can be accessed within the IoT devices and it is processed
and directed to the decision-aided system in order to acquire accurate meaning from
it. Such processes can range based on the architecture of the IoT although the flow
of information can be the same within these frameworks. Having no loss in general-
ity, if a certain application or any client requires some data within the IoT gadgets,
then the entity should be effectively authenticated [24]. If this happens, the request
of accessing the information can be neglected. Similar to other networks, the access
control can be of significant use within the IoT networking system and similarly can
be problematic considering that it can be restricted by network volume, heterogene-
ity, asset limitation of various devices, security of various networks, vulnerable
threats, etc. Moreover, it is crucial for giving and denying various clients access to
critical information for the applications that use the IoT services. As discussed
regarding the ML-centered mechanisms within the IoT, first it is essential to focus
on various segments that have access control.
According to scientific research, the access control can be divided into different
sections such as CWAC (context-aware access control), RBAC (role-based access
control), and PBAC (policy-based access control). Other researches have shown
that improved groupings include the ABAC (attribute-based access control), CAC
(capability-based access control), UCAC (usage control-based access control), and
OAC (organization-based access control). Some of the present works may be seen
within the mentioned sections. On the other hand, a comprehensive survey was car-
ried out and utilized the access control mechanism within the IoT network hence
pointing out various pitfalls within the IoT. Various authors may analyze the present
mechanism centered on the outstanding applications for the IoT. Such applications
are widely segmented in primary classes, the enterprise and personal. There are a lot
of applications such as healthcare, minor offices, digital homes, sensor networking
systems, smart firms, perilous infrastructure, business applications, and many more.
The access control systems are utilized within applications and within the architec-
tural levels.
For the architectural layers, every product or service comprises a collection of
decisions which include the access control languages such as XACML (extensible
2 Organization Internet of Things (IoTs): Supervised, Unsupervised… 45

access control mark-up language), UMA (user-managed access), and OAuth (open
authorization). Such architectural phase control contrivances can be utilized by
using the present protocols within the IoT network; for example, the OAuth can be
enacted across the present IoT protocols, for example, the CoAP (constrained appli-
cation protocol) and the MQTT2. Additionally, the various services provided by
Facebook, Google, Netflix, Microsoft, and many others comprise many client
accounts, which all utilize OAuth [25]. Subsequently, a discussion pertaining to the
present DL- and ML-centered authentication and various accessible mechanisms
within the IoT networking systems is presented below.
Access control for ML-based authentication processes within the IoT:
According to extensive analysis, a physical layer of authentication mechanism has
been suggested within the IoT network. Hence, the suggested mechanism utilizes
various physical properties, which include the strength of every signal. The
essence is that some revolutionary techniques have been used for physical layer
verification processes and gaming theoretical approaches, and these techniques
are machine-based, which ensures that there is effective isolation of various spoof-
ers which emanate from the benign IoT clients. The authentication systems can be
created spoofing method that can be attempt the utility while increasing the threat
frequency. On the other hand, the frequency channels get feedback, which is uti-
lized in establishing a Nash equilibrium (NE). Hence, it is important to point out
that every packet acquired via the radio channels comprises various channel states
that can leverage the test threshold focusing on the authentication of every deci-
sion, which would have been created. Due to this reason, various scholars have
attempted to use Q-learning and Dyna-Q that aid in comprehending the state of
every channel while not acquiring enough data pertaining to the channel. With the
aid of experiments, scientists have concluded that sensing precision and execution
process for the Dyna-Q model is effective compared to the Q-learning methods. A
familiar physical authentication layer project within the distributed Frank-Wolfe
(dFW) algorithm has been recommended.
The access control of the DL-centered authentication process within the IoT:
The prospected client’s authentication methods used within the IoT focusing on
the physiological events caused by human beings have been influenced via Wi-Fi
signals. Therefore, the recommended scheme has been based on various recogni-
tion events for humans. Activities may be executed using data that are coarse-
grained and have features that are smaller. Due to this reason, the information
states for every channel within the Wi-Fi signals have been created by the IoT
gadgets, and hence these characteristics are so much dissimilar considering their
features. The DNN (deep neural network) can be used in learning more about
human physiological and various behavioral qualities, which can be useful within
the authentication process. For the three layers, there will be extraction of differ-
ent forms of events, and at the second layer, there will be learning of the activity,
while the third layer will comprise highly based elements that are centered on the
authenticity of each user.
46 A. Haldorai et al.

2.6.1 Mitigation and Detection of Threats

Following the subsequent sections, there was concise explanation of various inva-
sions that were launched within several layers of the IoT network. Due to the limited
number of resources and heterogeneity for the IoT devices, they create a sufficient
platform for threats to come in. Normally, attackers may use the identified vulner-
abilities within all the networks, and the gadgets can be executed using dissimilar
types of invasions. Therefore it is recommendable to mention that the invasions may
range from low-profile hacking to a gadget that is enormous, for example, the
WannaCry, and can be much more complicated such as that of Dyn and Mirai.
Conventional invasion sensing and various mitigation systems can be centered on
cryptography, and at times there can be inadequate precision, and this may result to
inaccurate positives. For this reason the ML-centered methods, for example, the
DL, KNN, and SVM, and the unsupervised learning methods can be used. In this
section, there will be a brief outline of the ML-based techniques that are used for
attack sensing and various mitigation processes within the IoT networking system.
Moreover, in this section there will be an analysis of present DL and ML techniques.
The ML invasion sensing and its mitigation used within the IoT network: There
has been proposition of the semi-supervised attack sensing mechanism that can be
used within the IoT. In real sense, the recommended system is centered on the use
of the ELM (extreme learning machine) algorithms alongside the FCM (fuzzy
C-means) techniques, generally referred to as the ESFCM. The ESFCM can be
enacted within the fog substructure. One of the speculated features of the ESFCM
is that it deals with tagged information hence augmenting the sensing rate within the
evenly spread attacks. Although the sensing precision within the ESFCM can be
lower compared to the initially mentioned DL mechanisms, it may outwit the con-
ventional mentioned ML algorithms used for sensing various invasions [26].
Nonetheless, some semi-supervised learning systems draw various elements from
both the unsupervised and supervised learning systems, which may make them
effective compared to other systems. As stated previously, the IoT comprises a vast
number of breeds in which some of them include personal networks and may
become more complex within the infrastructural industry, for example, the
smart grid.
The DL-centered invasion sensing and mitigation techniques within the IoT: The
sensing of the DL-based invasions within the IoT through leveraging of the fog
system has been stated. For this reason, the invasion sensing system can be enacted
within the edge of the smart infrastructure. Hence, the aligned attack sensing mech-
anism comprises various account parameters that include the learning systems and
will choose the type of output that will be allocated to certain data. Hence, the the-
ory that has been employed in utilizing the fog technology is an effective asset limi-
tation and a natural application within the IoT. In an event where there is intense
infrastructure, therefore the learning mechanism can be close to the information
nodes so that there can be timely delivery and quality decisions can be made when
there is an attack.
2 Organization Internet of Things (IoTs): Supervised, Unsupervised… 47

2.6.2 Attacks Within the DoS and Distributed DoS (DDoS)

DoS and DDoS are the two most dangerous invasions whose attacks are always
difficult to alleviate within the IoT surroundings. Many claims have been stipulated
so that enough reasons can be provided to satisfy the question of why these attacks
are immense. Some of the reasons may include a sheer number of linked IoT gad-
gets being heterogeneous, the Internet, insufficient mechanisms considering the
limitation of resources for the IoT gadgets, communication within platforms,
extensive communication, etc. Such scenario illuminates various dangers for the
IoT gadgets in contradiction to the DDoS invasions. These types of attacks can be
evaded so that there can be assurance of sufficient operation of the IoT applications.
Based on this, the IoT can be known as a land of chances applicable to the DDoS
invaders. During 2016, there was an extraordinary increase in invasions in contra-
diction to the IoT substructure extensively. An example of such an attack was
Mirai4 that led to the decline of the Internet whereby devices, for example, printers,
babycams, etc., have been utilized like robots for the execution of the DDoS inva-
sions upon many firms [27]. Equally, some like Mirai bots were reported also. With
its inception in 2016, the malware led to substantial disruption within Internet ser-
vices based on its complicated contagious contrivance within the IoT networking
system. Due to the claim that only IoT devices can be hacked, they can be utilized
and be regarded as front tools for the execution of disreputable invasions in contrast
to various organizations, which advocate intense intellectual safety measures,
which can safeguard their equipment.
Until now, crucial research outcomes have been acquired via different methods
in order to alleviate the DDoS threats within the IoT, although dissimilar architec-
tures ensure that it is hard to create a combined system in order to curb the DDoS
threat within dissimilar IoT podiums. Conventional DDoS sensing and vindication
systems within the IoT networking system have been used on routers, gateways, or
accessible points with the aid of intrusion sensing and anticipation methods. Based
on the previously mentioned protocols, both the CoAP and the MQTT can be uti-
lized as protocols for the IoT. According to scientific research, the DDoS invasion
has been evaluated based on the attack on the generic IoT, which utilizes the
CoAP. Such works can be utilized to simulate an invasion in order to evaluate the
capacities within the IoT network. Additionally, in order to sense and mitigate the
DDoS attacks, some enabling technologies, for example, fog and cloud computing,
can be utilized for assisting the detection of DDoS systems within the IoT. For
example, fog computing can be used as a centered approach for safeguarding vari-
ous malicious threats. Amid several flavors for the IoT networking systems, various
critical development-assisted networks should portray rigid resilience for curbing
the DDoS invasions.
Cloud computing and fog DDoS mitigation frameworks can also be sug-
gested. Within the DDoS levels, a conventional mechanism can be employed
within several layers within the IoT industrial networking systems. Hence, the
minimum levels that utilize the computing edge along with the SDN gateways
48 A. Haldorai et al.

and the networking traffic can be accumulated within a fog computing level that
comprises the DDoS. Moreover, the honeypots can be used within this phase.
At last, the cloud computing levels can be created and evaluated within a cloud-
based platform in order to sense various vulnerable DDoS invasions. Based on
the previous studies, it is vividly clear that there will be no bullet, which is in a
position of catering for the DDoS threat within asset-restricted network cover-
age, and it is known that literature that is intellectual can be important in sens-
ing and mitigating the DDoS. Moreover, falsified values cannot be questioned
in these scenarios; hence, the request can be neglected [28]. For these reasons
the asset-limited and insufficiently effective gadgets within the IoT networking
system may bring out an alluring environment that can be good for harmful
threats. In spite of the latest developments in understanding such types of inva-
sions, still it is important to work on the intellectual contrivances that do not
only accompany the traffic amount but are also characters of the threat. In con-
sidering the subject of this issue, machine learning can be regarded as an effi-
cient candidate that can be leveraged for use in DDoS sensing within the IoT
networking systems. In order to alleviate the DDoS invasions within the IoT
networks, the research sector has been in a position of leveraging some ML and
DL experts.
1. The ML-centered methods to highlight the DDoS and the DoS invasions within
the IoT: First, there is a detailed rationale based on the present ML systems for
the DDoS sensing within the IoT networking systems. The typical elements
within the IoT systems in which involved consistent communication lacking the
back end networking servers. Moreover, there was a comparison between the
decision tree, the k-nearest, the random forest, the neural networks, and the SVM
based on whether they can sense DDoS in the IoT. On the same note, within the
SDN environment, the SVM-centered DDoS sensing systems have been used.
Apart from this, for DDoS sensing, the SVM has been proved to have high accu-
racy compared to other mechanisms. Other techniques can also be compared
with others such as the RBF (radial basis function), naïve Bayes, bagging, and
random forest.
2. The ML-based IDS within the IoT: There was a proposition of using ML-based
IDS that is lightweight, has low power, and can be used for operating the
6LoWPAN. This mechanism utilizes three methods such as the K-means, the
decision tree, and the hybrid methods which combine the previously mentioned
methods. Furthermore, the leveraged ML methods can sense the intrusions
within the IoT pathways.
3. The IDS within the IoT heterogeneous networking systems: Within a heteroge-
neous networking system for example, the RNN (recurrent neural network) has
been utilized within the LSTM infrastructure. RNN (random neural network) has
also been used that can be effective for the realization of the fast intrusion-based
sensing within minimal power of the IoT networking system.
2 Organization Internet of Things (IoTs): Supervised, Unsupervised… 49

2.6.3 Analysis of Malicious Programs Within the IoT

The depreciation in number alongside the heterogeneity of IoT gadgets has pro-
vided a cool and excellent platform for cybercrimes. The potential vulnerability can
be utilized within the system. A notorious invasion within the domains can be done
through vulnerable code injection which can affect the performance of the present
IoT gadgets. The threat, which can be performed by injection of different malware,
can be linked to the use of various authorizations, security protocols, and authentici-
ties. Aside from the mentioned techniques altering the IoT gadgets so that the soft-
ware can be physically modified can aid attackers inject vulnerable codes. Prior to
involving yourself into deep data based on the malware, it is crucial to comprehend
various groups of malicious programs that may put the IoT security at risk.
Generally, a malware can be a persistent threat based on the previously mentioned
vulnerabilities that can be performed via various invasions. Common forms of mal-
ware include, although not restricted to, ransomware, bot, Trojan, adware, etc.
This section will summarize the groupings of different malware, which influence
the operation of the IoT gadgets, and further analyze the present solutions such as
ML-centered methods, which maintain the IoT gadgets’ security. According to sci-
entific study, a vast number of smart gadgets that have been linked to the Internet are
deprived of enough security, which not only is dangerous to the devices but also
may permit the invader to gather resources meant for enormous attacks, for exam-
ple, the DDoS. For example, scientists have attempted to test different music gad-
gets for various problems and used vulnerable codes to show that these devices are
highly prone to different attacks when linked to the web.
Invaders also can utilize Internet-based webcams installed in public places, res-
taurants, and residential places for threatening intentions. Furthermore different
types of malware have been created for the disintegration of normal operation of a
business and other entities; these malware programs include, although not restricted
to, the Night Dragon, NotPetya, CryptoLocker, Stuxnet, etc. There are detailed clas-
sifications and generalized principles for various malware programs. Such may
include the generic attacks but there are also customized families for the malware
threats which are specifically aimed at the IoT gadgets. Many malware invasions
may include CryptoLocker, WannaCry, Stuxnet, etc. Such attacks have caused sub-
stantial losses for the industry and depreciation of public image for the entire firm.
Information connected to these threats is all over the entire chapter. It is crucial that
people comprehend a generalized concept for the invaders for launching the threat
using a malware.
Precisely, the invaders may accumulate ideas pertaining to capable targets, for
example, various network sensors via taking a reconnaissance. A vast number of
methods are present which can be used for executing a reconnaissance, such as
Wireshark, Nmap, and Metasploit, alongside the use of social engineering tech-
niques. Various present tools give a chance of additional data based on the different
uses that may create a more conducive platform for threats to come in. For this
reason, every attacker will first think about the type of threat he/she would use in
50 A. Haldorai et al.

relation to the uniqueness of the device class. Until now, certain application evalua-
tion techniques, for example, the OWASP (Open Web Application Security Project),
give access to the primary sources of threats by attackers making them to select an
effective exposure [29]. Generally, the OWASP gives an opportunity to various
sources so that they can exploit the resources that can be utilized by various invad-
ers, for example, the injection of SQL, misconfiguration of security measures, mis-
guided authentication, and XSS (cross-site scripting).
Moreover, based on the device type and its exploitation, the attacker can transmit
a payload to a specific target via the use of various media, for example, rootkit,
phishing, and various updates. For these reasons, the malware group can range from
a simple sole malware to a multifaceted, intellectual, inactive, and versatile emanat-
ing malware. Currently, intelligent malware can be very adaptive based on the vicin-
ity of the IoT whereby it can be used for getting access to the prospected network
and adjusting the performance processes based on the network. For example, vari-
ous malware may go astray when detected and can be latent for a certain period;
hence, they do not perform malevolent coding during that period so that their main
intention would not be compromised. Based on this concept, various measures that
can be used for evading the detection method are as follows:
1. Malware elusion methods: Different techniques that are used for evading these
malware can be used. Encoded malware implies to the encrypted harmful codes
in relation to the subsequent decryption so that it may surpass the signature-­
linked antivirus. Although the decryptor tends to be the same, for dissimilar ver-
sion in relation to the same malware, it still remains to be detected. In order to
overwhelm all of these, a decryptor can be transformed and hence will avoid the
detection contrivance. Such a form of malware is known as oligomorphic mal-
ware. Correspondingly polymorphic malware creates diverse decryptors that
will make it hard for sensing engines to identify the malware [30]. The metamor-
phic malware can be indeed the complicated malware among other groups since
it emanates to a new generation whereby it can be dissimilar to any other for
detection.
2. Malware ML-centered evaluation within the IoT: Moreover there has been
accumulation of various supervised learning methods that use random forest
groupings which are employed for sensing malware that is Android-based.
Therefore every classifier’s sensing precision can be evaluated based on the
malware checklist with the application of Androids. There has been an investi-
gation on the techniques for sensing various malware and their propagation in
the WMS (wireless multimedia system) in IoT. Research has shown that cloud-
based techniques within the SVM have been leveraged in order to sense vari-
ous capable malware and their propagation and can be utilized in dissimilar
gaming so that they can minimize the malware threats. Just after acquiring the
Nash equilibrium, there has been an attempt of looking out for a conducive
plan for the WMS so that it can protect itself against malware. Correspondingly,
2 Organization Internet of Things (IoTs): Supervised, Unsupervised… 51

the prospected SVM linear methods for the classification of various malware
that are Android-­based are used.
The detection system is centered on ANN before being evaluated by malware
and benign application in Android devices. Essentially, MalDozer is centered on a
series like API technique calls in resource permission, Android, and fresh method
calls, among others. Moreover, the recommended methods include the automated
engineering characteristics during the process of training. Moreover, the research
proposes an image recognition-centered DDoS malware-detecting method of IoTs
network. The research recommended a solution, whereby the researchers collected
and categorized two significant segments of malware known as Linux and Mirai.
Thereafter, research practices convert the segment binaries of the IoTs application
to the grayscale image. Afterward, CNN is conditioned to categorize the images to
malware and good-ware. Other researches considered using the deep auto-encoders
to show the botnet threats in IoTs. In the solution identification, the researchers
evaluated the behavior of networks before using the deep auto-encoders to separate
any form of anomaly in the network. Thus, the discussion concerning DL and ML
approaches for the detection of malware in IoTs networks indicated that the tech-
niques used in data training are used to identify indefinite malware.

2.7 Conclusion and Future Contributions

In conclusion, the emergent advancements recorded over the past few years
show that Internet standards and computer systems have enhanced the process
of communication between various network devices. Approximately 25 billion
devices are speculated to link up with the Internet by 2020. This speculation
gives rise to the novel development idea of the IoTs, which is a combination of
the embedded advancements such as wireless or wired communication devices,
actuator and sensor devices, and physical objects linked to the Internet. The
most vital objective of computing is to enrich and simplify human experiences
and activities which is also one of the visions linked to twenty-first-century
computing. The IoTs necessitates information to provide effective services to
individuals or improve the IoTs system evaluation to attain a smart environ-
ment. In that way, frameworks have to be able to access raw information from
various resources over the networking devices and evaluate data to extract user
competencies. Big data is explained as a high-velocity, high-volume, and high-
variety data set, which is expensive and innovative to enhance user automation,
decision, and insight. In consideration to the challenges caused by big data, the
future relies on the launching of novel concepts of intelligent data to enhance
productivity, effectiveness, and efficiency.
52 A. Haldorai et al.

References

1. Anandakumar, H., & Umamaheswari, K. (March 2017). Supervised machine learning tech-
niques in cognitive radio networks during cooperative spectrum handovers. Cluster Computing,
20(2), 1505–1515.
2. Anandakumar, H., & Umamaheswari, K. (October 2018). A bio-inspired swarm intelligence
technique for social aware cognitive radio handovers. Computers & Electrical Engineering,
71, 925–937. https://doi.org/10.1016/j.compeleceng.2017.09.016.
3. Adli Mehr,K., &Musevi,N. J. (2016).Security bootstrapping of mobile ad hoc networks using
identity-based cryptography. Security and Communication Networks, 9(11),1374–1383,
Available: https://doi.org/10.1002/sec.1423
4. Bredt, S. (2019). Artificial Intelligence (AI) in the financial sector—Potential and pub-
lic strategies.Frontiers in Artificial Intelligence, 2, Available: https://doi.org/10.3389/
frai.2019.00016
5. Chittaro,L.,&Ranon, R. (2004). Hierarchical model-based diagnosis based on structural
abstraction. Artificial Intelligence, 155(1–2), 147–182. Available: https://doi.org/10.1016/j.
artint.2003.06.003
6. Chatzigiannakis,I., &Nikoletseas, S. (2004) Design and analysis of an efficient communication
strategy for hierarchical and highly changing ad-hoc mobile networks. Mobile Networks and
Applications, 9(4), 319–332. Available: https://doi.org/10.1023/b:mone.0000031591.74793.52
7. Haldorai, & Kandaswamy, U. (2019). Supervised machine learning techniques in intelligent
network handovers. In EAI/Springer innovations in communication and computing (pp. 135–
154). New York: Springer. https://doi.org/10.1007/978-3-030-15416-5_7.
8. Delgadillo,G. Ó. (2016).Fishes and mobile macroinvertebrates of artificial habitats in Taganga
Bay, Colombian Caribbean. Bulletin of Marine and Coastal Research, 38(1), Available:https://
doi.org/10.25268/bimc.invemar.2009.38.1.338.
9. Dubois,D.,&Prade,H. (2001). Special issue of the journal Artificial Intelligence on “Fuzzy set
and possibility theory-based methods in artificial intelligence.”Artificial Intelligence, 127(2),
269–270, Available: https://doi.org/10.1016/s0004-3702(01)00080-7.
10. Dubois,D.,&Prade, H. (2003).Fuzzy set and possibility theory-based methods in artifi-
cial intelligence. Artificial Intelligence, 148(1–2), 1–9, Available: https://doi.org/10.1016/
s0004-3702(03)00118-8.
11. Ebendt,R.&Drechsler, R. (2009).WeightedA∗search – Unifying view and application. Artificial
Intelligence, 173(14), 1310–1342, Available: https://doi.org/10.1016/j.artint.2009.06.004.
12. Giudici, P. (2018).Fintech risk management: A research challenge for artificial intelli-
gence in finance. Frontiers in Artificial Intelligence, 1. Available: https://doi.org/10.3389/
frai.2018.00001.
13. Gong,T., &Bhargava, B. (2012).Immunizing mobile ad hoc networks against collaborative
attacks using cooperative immune model. Security and Communication Networks, 6(1),
58–68, Available: https://doi.org/10.1002/sec.530.
14. Gu,B., &Chen, I. (2005).Performance analysis of location-aware mobile service proxies for
reducing network cost in personal communication systems. Mobile Networks and Applications,
10(4), 453–463, Available: https://doi.org/10.1007/s11036-005-1557-x.
15. Guo, M. (2011).A simple and rapid authentication protocol in mobile networks. Security and
Communication Networks, 7(12), 2596–2601, Available: https://doi.org/10.1002/sec.386.
16. Jávor,A.,&Szűcs, G. (1998).Simulation and optimization of urban traffic using AI.
Mathematics and Computers in Simulation, 46(1), 13–21. Available: https://doi.org/10.1016/
s0378-4754(97)00154-7.
17. Kharin, Y. (1994).Artificial intelligence frontiers in statistics AI and statistics III. Knowledge-­
Based Systems, 7(1), 57–58, Available: https://doi.org/10.1016/0950-7051(94)90017-5.
18. Lasi, H. (2013).Industrial intelligence – A business intelligence-based approach to enhance
manufacturing engineering in industrial companies.Procedia CIRP, 12, 384–389, Available:
https://doi.org/10.1016/j.procir.2013.09.066.
2 Organization Internet of Things (IoTs): Supervised, Unsupervised… 53

19. Leo Kumar, S. (2017).State of the art-intense review on artificial intelligence systems applica-
tion in process planning and manufacturing. Engineering Applications of Artificial Intelligence,
65, 294–329. Available: https://doi.org/10.1016/j.engappai.2017.08.005.
20. Marquis,P.&Schwind,N. (2014).Lost in translation: Language independence in propositional
logic – Application to belief change. Artificial Intelligence, 206, 1–24, Available: https://doi.
org/10.1016/j.artint.2013.09.005.
21. Mihăiţă, A.,Dupont,L.,&Camargo,M. (2018).Multi-objective traffic signal optimization using
3D mesoscopic simulation and evolutionary algorithms. Simulation Modelling Practice and
Theory, 86, 120–138, Available: https://doi.org/10.1016/j.simpat.2018.05.005.
22. O’Halloran,S.,&Nowaczyk, N. (2019).An artificial intelligence approach to regulating systemic
risk. Frontiers in Artificial Intelligence, 2, Available: https://doi.org/10.3389/frai.2019.00007.
23. Panneerselvam, S. (2019).Survey of artificial intelligence used in industries a review. Industrial
Engineering Journal, 12(5), Available:https://doi.org/10.26488/iej.12.5.1188.
24. Park,S.,&Moo,Y. S. (2010).Updating route information in mobile ad hoc networks. Journal of
Mobile Communication, 4(1), 27–32, Available: https://doi.org/10.3923/jmcomm.2010.27.32.
25. Suganya, M.,&Anandakumar, H. (December 2013).Handover based spectrum allocation in cog-
nitive radio networks. In 2013 International Conference on Green Computing, Communication
and Conservation of Energy (ICGCE). doi:https://doi.org/10.1109/icgce.2013.6823431.
doi:https://doi.org/10.4018/978-1-5225-5246-8.ch012
26. Rao, M. (1992).Frontiers and challenges of intelligent process control. Engineering
Applications of Artificial Intelligence, 5(6), 475–481. Available: https://doi.
org/10.1016/0952-1976(92)90024-e.
27. Tünay, O. (2010).Chemical oxidation applications for industrial wastewaters. Water
Intelligence Online, 9, Available: https://doi.org/10.2166/9781780401416.
28. Varaprasad, G. (2012).Stable routing algorithm for mobile ad hoc networks using mobile
agent. International Journal of Communication Systems, 27(1), 163–170, Available: https://
doi.org/10.1002/dac.2354.
29. Vôhandu, L. (1994).Artificial intelligence frontiers in statistics. Engineering Applications of
Artificial Intelligence, 7(1), 87, Available: https://doi.org/10.1016/0952-1976(94)90049-3.
30. Wu, Y.,Deng,S.,&Huang, H. (2012).Information propagation through opportunistic com-
munication in mobile social networks. Mobile Networks and Applications, 17(6), 773–781,
Available: https://doi.org/10.1007/s11036-012-0401-3.
Chapter 3
Enterprise IoT Modeling: Supervised,
Unsupervised, and Reinforcement
Learning

Rajesh Kumar Dhanaraj, K. Rajkumar, and U. Hariharan

3.1 Introduction

This chapter introduces the Internet of Things (IoT) and machine learning (ML),
including algorithms [1]. According to Statista [26], the worldwide artificial intel-
ligence marketplace totaled $9.51 billion (U.S. dollars) in 2018 and was predicted
to reach $14.69 billion in 2019 (Fig. 3.1). Statista also estimated that the number of
connected devices in 2019 would total 26.66 billion (Fig. 3.2).
Currently, IoT and ML are fast-growing digital systems. Contemporary systems
often intersect, as progress in any kind of technology is no longer achievable in
isolation. IoT and ML systems have robust intersections that allow for a variety of
strategies, as discussed in the following sections.

3.1.1 Internet of Things

In recent years, the Internet has been extended to nearly all connected products
(called “Things”) and their virtual representations. The Internet of Things (IoT) has
provides for many possible uses, services, and products in numerous areas, includ-
ing smart homes, healthcare, and smart transportation [2, 3]. Research in this area
has received plenty of interest, and naturally plenty of funding. It is supported
through the efforts of academia and business, along with standardized systems for a

R. K. Dhanaraj (*)
Galgotias University, Greater Noida, India
K. Rajkumar · U. Hariharan
Galgotias College of Engineering and Technology, Greater Noida, Uttar Pradesh, India

© Springer Nature Switzerland AG 2020 55


A. Haldorai et al. (eds.), Business Intelligence for Enterprise Internet of Things,
EAI/Springer Innovations in Communication and Computing,
https://doi.org/10.1007/978-3-030-44407-5_3
56 R. K. Dhanaraj et al.

Market revenue in billions U.S dollars


140
118.6
120

100 94.41

80 70.94

60 51.27

34.89
40
22.6
20 14.69
9.52

0
2018 2019 2020 2021 2022 2023 2024 2025

Fig. 3.1 Growth of the artificial intelligence industry

Connected devices in Billions


80 75.44

70
62.12
60
51.11
50
42.62
40 35.82
30.73
30 26.66
23.14
20

10

0
2018 2019 2020 2021 2022 2023 2024 2025

Fig. 3.2 Estimated number of IoT-connected devices worldwide from 2015 to 2025

number of towns in telecommunication, health insurers, the semantic Web, and


informatics.
For many years, typical methods were created only for specific uses with mini-
mal versatility. Therefore, when a device was working, it could not be transformed
flexibly and dynamically. The first step in introducing the IoT (or more generally,
the potential future of the Internet) calls for service platforms, products, and
­applications that can record, communicate, shop, share, and gain access to informa-
3 Enterprise IoT Modeling: Supervised, Unsupervised, and Reinforcement Learning 57

tion through the actual physical planet; in fact, they are able to speak together
around the world. This will likely result in brand-new possibilities for many differ-
ent domains, including health and fitness, eco-friendly power, manufacturing, smart
residences, and personalized end-user apps.
In this way, IoT plays a much more crucial role in daily life. The volume of infor-
mation on the Web and the Internet is already too much to handle and continues to
grow at an incredible pace: Each day, approximately 2.5 quintillion bytes of infor-
mation are produced. In addition, 90% of the current information was produced in
the last few years [4]. Sensory specifics, the info originating from receptors, may be
analyzed by means of algorithms and switched into straight in machine info that
designs possess a definite understanding about the legitimate data. By doing this, the
device is able to act human in some way (some call this type of technique “artificial
intelligence”). In addition, we are able to innovate much more useful programs,
services, and products that alter our lives dramatically and automatically. For exam-
ple, meter readings can be used to predict and balance energy usage within smart
grids; examining a combination of traffic, pollution, water, and congestion sensory
details can provide much better information to site visitors as well as community
management; processing and checking sensory products linked to individuals can
improve remote health care [5]. Such an information transformation procedure is
better illustrated using a “knowledge hierarchy”. We have adapted definitions for the
concepts of serotonin and semantics (see Fig. 3.1).

3.1.2 Machine Learning

Machine studying (ML) has revolutionized how we do business. A disruptive cut-


ting edge technology which differentiates ML out of various other strategies to
hands free operation is actually a level far from the rules-based programming. ML
algorithms have made it possible for engineers to use information without having to
explicitly program models to examine specific parts of a problem. Rather, the
devices themselves return the proper responses, depending on the information they
have received [5]. This ability has caused manufacturers to reconsider the methods
they normally use to generate choices from the information.
Machine learning is used to generate predictions from new information by using
historical details as an instructive case in point. For example, you might wish to
determine a consumer lifetime valuation within an eCommerce retailer computing
website for an upcoming client meeting. If you currently have historical details on
customer interactions and net income related to the buyers, you might wish to make
use of machine learning. ML can determine the buyers who are prone to have the
greatest net benefit, helping you to concentrate your efforts on them.
Among the strategies used to provide algorithms with information, the most
popular design is known as supervised learning. This chapter discusses this area of
information science and the reasons why it is regarded as low-hanging fruit for
companies that intend to venture into ML.
58 R. K. Dhanaraj et al.

3.1.3 Machine Learning Plus IoT

ML is being increasing explored by manufacturers because of the hoopla surround-


ing IoT. Many companies have recognized it as an important strategic area, whereas
others are launching pilot programs to determine the possibilities for IoT in their
business activities. Consequently, virtually every IT seller is suddenly announcing
IoT and consultation services expertise. However, increasing profits by means of IoT
is not simple. First, a possible lack of concrete goals is quite disconcerting.
Improvements in digitization and IoT locations are brand-new prerequisites with two
customers as well as sellers. Many commercial enterprises have neglected to deter-
mine how places are going to change together from the setup of an IoT technique.
Quite simply, well-defined, concrete intermediary goals are often absent. For
example, manufacturing businesses generate a tremendous quantity of information
every day. Nevertheless, businesses do not systematically gather and analyze these
data to enhance a procedure’s effectiveness or even create additional objectives. In
addition, very few vendors actually know how to explain, using concrete terminol-
ogy, to a prospect the best way to implement beneficial IoT strategies. Just a prom-
ise associated with a cloud-based IoT wedge is not sufficient [6, 7].
According to Gartner [27], companies in Finland have advanced to a chapter in
which discussions on IoT involve specialized terminology rather than internet busi-
ness objectives. Consumers are offered revolutionary proposals and are also empow-
ered to take control of the project. However, vendors have to boost their abilities to
describe, using a lot more concrete terminology, how businesses can leverage the
use of IoT, as well as be inclined to help companies recognize the potential and
build practical blueprints. When a seller provides a response on an individual analy-
sis that is too general, alarm bells ought to be ringing.

3.1.3.1 Finding Solutions in IoT Data

A business’s brand-new facility was encountering serious production issues resulting


from a crucial turbine disaster. A third party was employed to resolve the situation.
After approximately 6 weeks, the team of four specialists working with just a field
evaluation had made very little progress. The facility required a lot of equipment
maintenance and time-stamped environmental data. Even with all of this specific,
readily available information, no one was able to transform it into usable data [8, 9].
We assisted the company with an analysis. When all the information was
enhanced through analytics, underlying issues were discovered, primarily with
much-needed oxygen feeds throughout the manufacturing process. Since this dis-
covery, the facility did not have any major problems. This is a key example of
exactly how machine learning can be used to attain much greater effectiveness. With
the correct algorithms, a device can be gradually trained to perform internal and
external production-related elements, improve the use of consumables, and enhance
the effectiveness of the whole output procedure.
3 Enterprise IoT Modeling: Supervised, Unsupervised, and Reinforcement Learning 59

3.1.3.2 Machine Learning and IoT in a Business Domain Model

The whole planet is going crazy with information, both artificial intelligence and
IoT. Many publications have discussed the quantity of information we produce each
and every working day, and countless statistics have demonstrated just how much
detail we will produce in the coming years. Thus, we want to discuss how ideas or
algorithms coming from other systems can be put into IoT information for optimi-
zation. In our previous publications, we discussed data science algorithms with IoT
information; here, we discuss machine learning.
On an extremely basic level, Machine learning techniques minimize the man power
using training. ML reads patterns in information as a stand-alone device and makes
autonomous choices without requiring a developer to create a brand-new group of
codes. When you use Siri on your iPhone, for example, you may notice that its replies
become more sophisticated and accurate as you use it more often. That is among the
fundamental uses of machine learning [10–12].
But exactly how would machine learning improve the IoT? Each time IoT recep-
tors collect information, someone has to be working on the backend to classify the
information, process it, and make sure the information is delivered back to the unit
for choice generation. When the information generated is huge, exactly how can an
analyst take care of the influx? Driverless automobiles, as an example, need to make
fast choices when on autopilot, so depending on people is not possible. That is
where machine learning becomes useful. To determine which algorithm needs to be
used for a specific group of projects, first need determine the job. Several of the
activities involve discovering uncommon details, system findings, predicting values
and groups, function extraction, and others [7, 13].
Classifying the information sets for various jobs make it much easier for a novice
to recognize the proper algorithm program [6]. For example, to focus on informa-
tion system findings, clustering algorithms, including K means, can be used.
K-means was created to deal with substantial chunks of information that include
several detail types. In another example, one-class support vector machines and
Principal Component Analysis (PCA)-based anomaly detection algorithms are ideal
for instruction information coming from uncommon details.
Using IoT and machine learning in the same sentence is like playing buzzword
bingo. However, the two principles make good sense as a pair. By 2025, it is believed
that the IoT will produce more than 180 zettabytes of information yearly—that is
180 trillion gigabytes. Machine learning is best when used with sizable datasets. To
understand why the IoT requires ML to dominate the planet, we begin by examing
the two principles individually in Fig. 3.3. Then, we analyze four unique ML and
IoT real-life applications. The IoT is developing at an unprecedented speed. In fact,
there are 127 brand's -new gadget are connected to the web everyday. Improved
connectivity is only going to speed up the progression. McKinsey expects that, by
2022, 100% of the worldwide public will have access to low-power wide area net-
works (LPWANs) [3, 14, 15]. This can allow long-range marketing communica-
tions with attached products while optimizing each expense as well as power
consumption demands.
60 R. K. Dhanaraj et al.

Fig. 3.3 Data generation through an IoT system

3.2 Machine Learning Algorithms

Machine learning algorithms have specific and unique uses. The following list of
machine learning algorithms includes AI and developing machine learning sys-
tems [8]:
1. Linear Regression
2. Logistic Regression
3. Support Vector Machines
4. Random Forest
5. Naïve Bayes Classification
6. Ordinary Least Square Regression
7. K-means
8. Ensemble Methods
9. Apriori Algorithm
10. Principal Component Analysis
11. Singular Value Decomposition
12. Reinforcement or Semi-Supervised Machine Learning
13. Independent Component Analysis
An understanding of the algorithms will allow you to use them effectively in almost
any data problem and a variety of functional machine learning projects.
The following three categories apply to machine learning algorithms [16]:
1. Supervised Learning
2. Unsupervised Learning
3. Reinforcement Learning
It is important to understand each algorithm in order to select the correct one to
meet your problem and learning requirements, as shown in Fig. 3.4.

3.2.1 Supervised Learning

Supervised learning techniques are used by optimum machine learning users. In


supervised learning, an algorithm’s learning process is completed with an instruc-
tion dataset. Even by making use of a training dataset, the task could be regarded as
3 Enterprise IoT Modeling: Supervised, Unsupervised, and Reinforcement Learning 61

Fig. 3.4 Types of machine learning

a mentor that is supervising the learning process. The algorithm can help within
generating predictions regarding the data in the training process and obtain the mod-
ifications carried out by the teacher itself. A conclusion may indicate if the algo-
rithm has accomplished an appropriate amount or a certain degree of performance.
There are two kinds of supervised learning problems, which can be more grouped
generally as regression or classification issues:
• Classification Problems: A problem which generates adaptable, which moves
lower merely especially such as the “red” or perhaps “blue” or perhaps it may be
“disease” and “no disease”.
• Regression Problems: A regression concern is whether the variables are really
authentic, such as “dollars” or perhaps it might be “weight” [16, 17].
Several issues can be noted within the data type, including things like time series
prediction and recommendations. Supervised machine learning algorithms includ-
ing the following:
1. Decision Trees
2. Naive Bayes Classification
3. Support vector machines for classification problems
4. Random forest for classification and regression problems
5. Linear regression for regression problems
6. Ordinary Least Squares Regression
7. Logistic Regression
8. Ensemble Methods
62 R. K. Dhanaraj et al.

3.2.2 Decision Trees

A great deal of information is apparent in the name decision tree; using easy phrases,
a decision tree can help you make a choice regarding the information at hand. For
example, a banker can make a decision on providing a mortgage to an individual
based on age, profession, and amount using a decision tree. When creating a deci-
sion tree, the task is launched in root node, responds to certain questions within each
node, and considers the area that corresponds to the specific solution. Adhering to
the procedure, we traverse to the root node, then to a leaf, then type conclusions
within the context of the information. An example is provided in Fig. 3.5.

3.2.3 Unsupervised Learning

Unsupervised learning uses an algorithm in which you insert the information (X)
along with hardly, no any corresponding variables are getting established. The main
objective for unsupervised learning is to help model the underlying framework or in
an effort to assist the learners in understanding the information. Unsupervised learn-
ing is different from supervised learning because there are no any proper responses
are not mentioned the numbers to the instructor. Algorithms remain as their very
own products to help you find the intriguing framework that is contained in the
information. Unsupervised learning may be grouped as clustering or connection
issues [7]:

Fig. 3.5 A tree showing


the survival of passengers
on the Titanic
3 Enterprise IoT Modeling: Supervised, Unsupervised, and Reinforcement Learning 63

1. Clustering: A clustering shows what you would like to find out and also assists
in determining the natural groups belonging to the information, such grouping
clients according to their buying behavior.
2. Association: A connection becomes the learning issue. It is exactly where you
will find the guidelines that will explain the larger areas of your information. For
example, individuals who purchase X are also individuals who have a tendency
to purchase Y.
Some common unsupervised learning algorithms include the following:
• K means for clustering problems
• Apriori algorithm for connection guideline learning problems
• Principal Component Analysis
• Singular Value Decomposition
• Independent Component Analysis

3.2.4 Reinforcement or Semi-supervised Machine Learning

For some problems, you will need to enter a huge amount of information. In semi-­
supervised machine learning, the information is first labelled as (X), then later some
is labelled as (Y). This method is somewhere between supervised learning and unsu-
pervised learning. A great example is an image archive in which the subjects of only
some photos are labelled (e.g., dog, cat, most people, but a school is unlabelled) [4].
A great deal of practical society items learning related problems falls straight
into this specific group because they might be costly or time consuming. To label the
information, access may need to be obtained through the domain name profession-
als. The unlabeled information is low-cost also comparatively simple to gather as
well as find. Unsupervised learning strategies can be used to accomplish great
things. They can help you learn as well as find out the different legitimate compo-
nents that occur in the type of variable. Supervised learning methods help make the
very best of estimate predictions with unlabeled information. This ­information can
then be used in supervised learning algorithms as instructive information to make
predictions based on different details, as shown in Fig. 3.6.

3.3 Machine Learning Applications in IoT

3.3.1 Price Savings Come to Industrial Applications

Predictive abilities are incredibly beneficial in a manufacturing environment. By


taking in information from several sources within and on devices, machine learning
algorithms can “learn” what is common for any device then identify when some-
thing abnormal starts to happen. “The collected information is delivered to the serv-
64

Fig. 3.6 Overall classification of machine learning


R. K. Dhanaraj et al.
3 Enterprise IoT Modeling: Supervised, Unsupervised, and Reinforcement Learning 65

ers, exactly where it’s contrast to earlier details collected out of this printer, and
even information collected by using the same devices very much. The wedge of ours
is able to identify the smallest alterations as well as alert you of acquiring malfunc-
tions. This particular evaluation is actually carried out within real time and also the
outcomes are actually shown on the technician’s smart phone in just seconds.” [7]
Predicting if a machine requires upkeep can be quite important, converting
directly into money by saved expenses. An excellent case is Goldcorp, a mining
business that uses surprisingly impressive cars are dragged to the bay. When these
transporting automobiles decompose, Goldcorp is losing 2 billion dollars each day
from inefficiency [13]. Goldcorp is currently making use of machine learning to
predict with more than 90% reliability when devices will require upkeep, which
results in a large cost savings.

3.3.2 Shaping Experiences to Individuals

Almost everyone is familiar with machine learning in their daily lives. Netflix and
Amazon use machine learning to understand our tastes and provide a much better
experience for their end users. Thus, they suggest products that you might want or
provide suggestions for films and television shows to watch.
Likewise, IoT machine learning can be very important in shaping our environ-
ments to our personal preferences. The Nest Thermostat is a terrific illustration, as
it uses machine learning to study your tastes for cooling and heating, ensuring that
the home is the perfect temperature whenever you return from work or awaken each
morning.

3.3.3 The New Innovative IoT Emerging Business Models

Like each main technological shift, the growth of the IoT has been both successful
and disastrous. The passion with which business owners, investors, and govern-
ments have adopted IoT is a testament to its deep cultural and economic
possibilities.
Nonetheless, passion has outpaced comprehension. Consequently, the IoT busi-
ness is rife with all sorts of errors and misapplications that typically plague early
adoption. As IoT moves beyond its infancy stage, the economic realities are becom-
ing more clear. Probably the most considerable of the revelations is the fact that
conventional hardware does not conveniently affect IoT systems [3, 15] because IoT
products offer recurring, infrastructure-related expenses unlike conventional hard-
ware. Consequently, IoT systems are only confirmed to be beneficial when the soft-
ware leads to recurring, constant value for a client. This value is usually accomplished
by a number of methods that are distinct, including elevated advantages, decreased
operating expenses, and facilitation of compliance. With this section, we examine the
66 R. K. Dhanaraj et al.

Internet of Things (IoT)

4.0
Retail Healthcare Insurance Healthcare Buildings &
Smart Cities & Connected
Main Category Living
Energy Industry
∑ Energy efficiency &
∑ Assisted & HVAC ∑ Construction
∑ Digital signage
∑ Adherence & ∑ Adherence & autonomous driving ∑ Home equipment & ∑ Education ∑ Connected field
∑ In-store offering &
Support Support ∑ Fleet management appliances ∑ Energy ∑ Digital factory
promotions
∑ Clinical ∑ Clinical ∑ In-vehicle ∑ Living assistance ∑ Environmental ∑ Product design &
∑ Supply Chain
Industries / ∑ Smart ordering & ∑ Virtual care ∑ Virtual care infotainment ∑ Safety & security ∑ Roads, traffic & engineering
∑ Wellness & ∑ Wellness & ∑ Shared mobility transport
Applications payment
∑ Smart navigation
∑ Vehicle-to-
∑ Social & security
∑ Smart maintenance
∑ Vending machines Prevention Prevention infrastructure ∑ Supply Chain
∑ Vehicle assistance solutions ∑ Water & Waste management
∑ Workplace
operations

Fig. 3.7 Applications of the IoT

special economics of IoT and its most promising industrial uses. We also incorporate
particular use cases to help you understand the applications, as shown in Fig. 3.7.

3.3.4 The Economics of IoT Using Machine Learning

The standard hardware industry is based upon the traditional, product-at-a-margin


principal. A business creates a portion of hardware for X and also sells it to custom-
ers for X + Y. What complicates the unit inside IoT is the fact that the X value is
usually snowballing as well as recurring. That is, IoT items incur constant
infrastructure-­related expenses that do not affect conventional hardware. Therefore,
to stay operational, IoT products need a network where to run, along with a system
wedge where to gather as well as control information. Consequently, conventional
device economics easily fail on a company’s scale or as equipment ages. IoT busi-
nesses are actually switching to business to shield their margins from erosion. These
designs frequently depend on recurring payments near the buyer, such as to the
“hardware as a service” industry version. Hayesis placed a unit whereby a hardware
company leased the item to clients for a monthly or annual fee. IoT wedge suppli-
ers, meanwhile, charge monthly fees because of the compilation, business, and cre-
ation of device-driven details. Due to these recurring expenses, IoT systems are
economically beneficial whenever they provide recurring, constant worth to a client.
Thus, there may be a typical group of problems preventing a connected unit from
becoming successful:
• Model has an annuity (recurring earnings stream)
• Consistent or even improved sales through association with a consumable
• Direct and indirect operational effectiveness gains
• Exponential progress of perpetuity
A product has an incredibly scant, non-variable life cycle. Opening of various
earnings by the way of info put together. In other circumstances, an online business
offering linked treatments with an ordinary design will probably be unsuccessful.
That is because, with time, the recurring backend expenses will erode with the busi-
ness’s margins. Within these sections, we examine the most promising industrial
uses of IoT systems, as well as describe the apps with distinct use case examples [18].
3 Enterprise IoT Modeling: Supervised, Unsupervised, and Reinforcement Learning 67

3.3.5 Compliance Monitoring

Every year, American manufacturers invest an estimated $192 billion to comply


with economic, environmental, and work safety regulations. IoT systems have
shown strong possibilities in this realm. By remotely overseeing very sensitive
assets, IoT products can enable companies to considerably reduce expenses related
to regulatory compliance. A good example is software used in the oil and gas
industry. Gas and oil extraction and processing are governed by strict compliance
requirements. Environmental and safety laws require constant vigilance to uphold.
However, internet-connected equipment makes compliance considerably less
complicated and cheaper. In the past, an area agent would have to actually exam-
ine an extraction or processing site to confirm regulatory compliance. Today, an
IoT unit can be used in the area to remotely check vital compliance metrics, such
as oil leaks and gasoline pollutants. Not only does this decrease the expenses for
onsite tracking, but it is also a lot more responsive. Although an area agent might
be in a position to go to a certain site occasionally, an IoT unit is able to offer regu-
lar, up-to-the-minute details in real time. This responsiveness also helps to limit
the scope of penalties when noncompliance occurs. For pollutants and leakages,
regulatory penalties are usually proportional to the quantity of materials produced.
IoT products might limit a business’s risk for penalties by allowing to to react
faster to leaks.
IoT systems also are proving to be helpful with facilitating compliance within
the insurance industry. Insurance companies are using IoT products to monitor
­elements of policy adherence. A particularly intriguing illustration of this use
can be found in the homeowner’s insurance sector, where insurers are using IoT
products to monitor fluids amounts within basements. When water starts to build
up in a basement, the IoT unit sends notifications to the insurer and property
owner, so that they can correct the problem before substantial deterioration
occurs. This allows the insurer to decrease the price of policies and protects
homeowners by preventing damage to their homes. For these reasons, IoT sys-
tems may decrease the overhead associated with compliance and considerably
decrease a business’s costs associated with fines or policy payouts, for example
[19], as shown in Fig. 3.8.

3.3.6 Preventative Maintenance

The expenses related to equipment malfunction have a tendency to cascade. Just one
malfunctioning device can impede production, interrupt a whole supply chain, or, as
discussed previously, lead to regulatory noncompliance. Fortunately, the same fea-
tures that allow IoT systems to facilitate compliance also help companies protect
invaluable assets. Returning to the use case of gasoline and petroleum extraction,
we can see how IoT systems help to revolutionize equipment upkeep. Petroleum
extraction sites are full of expensive and highly complex products. Typically, these
68 R. K. Dhanaraj et al.

Fig. 3.8 IoT compliance monitoring

products were safeguarded exclusively by occasional and possibly imperfect field


inspections. Because of their intricacy and scope, even small technological mal-
functions could to lead to decreased production and substantial monetary losses.
Catastrophic disasters, meanwhile, could cost companies countless dollars. Today,
a low-cost IoT unit can be used to remotely keep track of gear, monitor upkeep
schedules, and protect against malfunctions. When a machine part starts to under-
perform or maybe malfunction, the IoT unit will immediately issue an automatic
alert. Doing this will restrict losses caused by inefficiency, as well as prevent large-­
scale problems and injuries [9].

3.3.7 Remote Diagnostics

The remote monitoring abilities of IoT technology can be effectively used outside
of the area of physical upkeep. IoT products are now being used to obtain compre-
hensive and regular analytic details in several areas, ranging from pharmaceuticals
to agriculture. The compilation of data is allowing companies to automate some
procedures and remotely manage manufacturing. A good illustration can be found
in the agricultural sector. Forward-thinking farmers have used IoT products to mon-
itor crops. A connected unit can constantly monitor environmental conditions, such
as soil conditions, sunlight, humidity, and temperature. This data is then compiled
and analyzed to help farmers assess long-term environmental conditions. Additional
IoT products may be used to immediately react to this information. For example,
when a device registers very low moisture in the soil, it can instantly apply water to
fix it. These features increase efficiency and decrease overhead expenses. Just like
other IoT programs, the outcome is better, more effective, and less costly [20].
3 Enterprise IoT Modeling: Supervised, Unsupervised, and Reinforcement Learning 69

3.3.8 Asset Tracking

Visibility is easily the most useful commodity within supply chain control. However,
more than 60% of businesses do not have complete supply chain visibility; the typi-
cal inventory precision threshold of U.S. merchants is only 63%. Fortunately,
low-­
­ cost IoT systems now allow companies to overcome this issue. A basic
3G-connected microcontroller can find and monitor inventory in real time, from
anywhere on Earth. This unprecedented level of visibility will help businesses to
prevent theft and loss, enhance fleet advantages, and improve forecasting. Monitoring
information can be shared immediately with other entities in a supply chain to addi-
tionally boost visibility and decrease inefficiencies. Cisco and DHL, for example,
expect that IoT systems can have a $1.9 trillion effect on their strategies and the
supply chain management industry. To help illustrate the possibilities, Fig. 3.9
shows IoT-­based advantage monitoring that benefits an simple, linear supply chain.

3.3.9 Automatic Fulfillment

Improved visibility on the customer end of this resource chain also allows small busi-
nesses to carry out brand new, much more profitable programs. Up-to-the-­minute
inventory as well as usage information produces automated satisfaction and price
agreements for new buyers. This performance can help companies by improving need
forecasting, generating product sales, and informing marketing and advertising
efforts. In doing so, customers can benefit by experiencing less merchandise short-
ages and more effective customer support. A fascinating example of this IoT program
is the Amazon Dash button, which is a small IoT unit that can be set up to reorder
certain products when pressed. For instance, a consumer can easily use their button to

Fig. 3.9 Asset tracking


70 R. K. Dhanaraj et al.

Fig. 3.10 Example of automated fulfillment

purchase a 12-pack of Bounty paper towels anytime it is pressed. Each switch is pro-
grammed for just one item and amount at any given time. The buttons can be mounted
with a user’s home, near each item’s location. Not only do the designs help to improve
product sales, but they supply priceless details on client consumption practices. These
system designs are still emerging, but their possibilities are clear. At the end of 2016,
Amazon announced that Dash gross sales had increased by more than 400% com-
pared with 2015, with 60 new brands signing on to participate in the program. Even
though the Dash design depends on client ­feedback to trigger replenishment, genu-
inely automated devices have started to gain traction, as shown in Fig. 3.10.

3.3.10 Adding Value to the IoT with AI and ML

Artificial intelligence may be used hand in hand with a different technology,


machine learning. Often used interchangeably, the terms AI and ML represent the
basic principle of acquiring applications that have intelligence. This particular intel-
ligence enables them to evaluate information and generate choices much like the
human mind. The essence of IoT products is to gather information and generate uses
from it; information from physical products via AI and ML allows us to expand
these procedures. The Internet of Intelligent Things (IoIT) uses artificial intelli-
gence takes importance over the IoT domain for higher interpreting information
about products that are connected [13].
The products within an IoT network are connected through receptors, actuators,
hardware, and software programs to provide people with rational inputs. The basis
3 Enterprise IoT Modeling: Supervised, Unsupervised, and Reinforcement Learning 71

of the IoT is AI and ML because it enables the products to interpret the information
collected. Whenever connected devices gather and compile raw details, the applica-
tions empowered with ML abilities can merge and assess the information [20].

3.3.11 Advantages of Using IoT and ML Together

The IoIT produces IoT programs that recognize their potential. AI and ML allow for
much more comprehensive insight at a quicker speed. Businesses are implementing
IoIT to the benefits discussed in the following sections.

3.3.12 Improved Accuracy Rate

If you have ever attempted to evaluate information in several different spreadsheets,


you know it can be a tiresome task. Human brains are restricted to performing par-
ticular duties at a particular speed, so when our brains are tired, we are likely to
make mistakes. The IoT has the capacity to process large amounts of information
and create an analysis. The entire process is machine and software driven; it can be
performed with no human intervention, which tends to make it error free and also
increases reliability [26].
E-commerce transactions, online payments, and ATM withdrawals, for example,
are very susceptible to fraud. Using the consolidated strength of human comprehen-
sion and IoT machine learning as well as Robotic Process Automation (RPA) meth-
ods of manmade intelligence, fraud may be prevented, therefore stopping some
monetary losses.

3.3.13 Predictive Analysis and Maintenance

Predictive analytics is a type of evaluation that examines pre-existing information


and predicts probable succeeding outcomes. It would not be an exaggeration to say
that IoT and AI are the basis of predictive maintenance. Presently, IoT products are
being used by businesses to monitors for malfunctions or concerns in an automatic
way with no human intervention. By using AI, this process can allow models to
perform a predictive evaluation, giving businesses the ability to identify possible
mishaps, malfunctions, and advance maintenance. For this reason, the risk of losses
is extremely reduced because problems are now being recognized in advance.
For example, delivery businesses can make use of predictive evaluation to exam-
ine and analyze their information to avoid unexpected downtime for shipping and
maintain shipping by way of frequent servicing [20].
72 R. K. Dhanaraj et al.

3.3.14 Improved Customer Satisfaction

The center of each company is client satisfaction. Presently, organizations such as


Amazon have gained recognition for being the best customer-centric businesses by
keeping the goals of their clients above all else. Nevertheless, human-based con-
sumer encounter may not be successful for a number of reasons, including language
barriers and time restrictions [20].
Businesses are realizing the advantages of AI by using chatbots to interact with
clients. Great quantities of client information can be used to supply them with a far
more personalized experience, as well as answering their queries appropriately.

3.3.15 Increased Operational Efficiency

Predictions produced via manmade intelligence are extremely beneficial for increas-
ing the operational effectiveness of a company. In-depth insights received via man-­
made intelligence are often used to enhance a company’s processes, which may lead
to enhanced operational effectiveness plus reduced expenses.
With exact predictions, insights can be obtained concerning price consuming
matters as well as the period of the housing industry also automates these to enhance
the effective numbers. Additionally, for shipping businesses, the insights received
via man-made intelligence can help to improve procedures, equipment options, and
inventory to reduce needless expenditures [25].
• Improve the basic safety of equipment – The IoT can keep track of all machin-
ery, from compressors to high-speed engines, to quickly recognized potential
problems. This can result in greater safety for staff members and even the
environment.
• Decrease lost revenue related to downtime – Downtime is costly. By using ML
and the IoT, problems can be recognized in advance to considerably decrease
downtime, thus minimizing lost revenue.
• Allow for correct record keeping – The IIoT makes use of the information to
keep track of equipment, flag questionable components, and determine what
maintenance is needed.
• Predict equipment failure before it occurs – Real-time data allows a company
to assess what is occurring, but the IIoT is able to take action on your behalf.
Using man-made intelligence, algorithms, and much more, information analyz-
ers are now able to predict and alert you to potential failures.
• Provide real-time data – With ML analytics, a company can obtain a data evalu-
ation in real time to ensure that things are operating properly and that the busi-
ness is flourishing.
3 Enterprise IoT Modeling: Supervised, Unsupervised, and Reinforcement Learning 73

3.4 Challenges in IoT Implementation

Problems in IoT implementation are related to network security expenses and infor-
mation evaluation. Most businesses may be ready to deal with sensors and data ana-
lytics, but they neglect to focus the same attention on network investments, system
integration, and security. This gap highlights a rift between technologies in addition
to excellent data transfer, where IT division is really sharp on buying the technical
feature needed having a prosperous IoT undertaking, it is the lack of really worth
sent, which is likely to create business experts look at the unsuccessful endeavor.
Capturing invaluable info that leads to phony attention could also work as the pri-
mary reason a great deal of business experts have a look at their IoT projects unsuc-
cessful. Problems experienced by businesses that create quality problems, resulting
in the venture disaster, which includes night conclusion as well as finances overruns
occasions [20].

3.4.1 I nteroperability and Compatibility of Various IoT


Systems

According to the marketplace analysts with McKinsey, 40–60% of overall value is


based on the ability to attain interoperability between various IoT methods. With
many vendors, service providers, and OEMs, it becomes very difficult to preserve
interoperability between various IoT methods. Receptors and networking are the
essential parts of the IoT. However, not every single machine comes with complex
receptors and social networking abilities. Also, receptors with different energy con-
sumption and security requirements, built within history devices, might not be able
to offer the exact same outcomes.
A fast workaround might be to add external receptors. However, this is also dif-
ficult because it must be determined what components and data will be communi-
cated to the network.

3.4.2 Authentication and Identification of Technologies

Currently, there are approximately 20 billion linked products in existence, as well as


in order to feel foundation of all the gear requires a good offer of safeguarding
implications without having just intricacy. Using many devices that are connected
on a single wedge requires formalization or a method to authenticate the gad-
gets [26].
74 R. K. Dhanaraj et al.

3.4.3 Integration of IoT Products with IoT Platforms

For a profitable setup of an IoT program, businesses have to incorporate a variety of


IoT connected goods with correct IoT platforms. Inappropriate integration could
result in irregularities in productivity along with executing to move really worth
towards the customers. A vice president with Gartner, Benoit Lheureux [27],
claimed that “Through 2018, 50 % of the expense of applying IoT remedies will
probably be invested combining different IoT parts with every back-end and other
methods. It’s essential to be aware of integration is actually an important IoT
competency.”
However, there may be too many IoT endpoints to aggregate the sensor informa-
tion or transmit it to an IoT wedge. With rich integration, businesses are able to
mine the massive information by way of Big Data methods in order to produce
awareness and predict results [19].

3.4.4 Connectivity

Connectivity is a common marketing difficulty because the Internet is not available


everywhere with the same speed. A worldwide satellite business, Inmarsat, dis-
closed that 24% of users reported connectivity problems as one of the greatest prob-
lems within an IoT deployment. Strong networks are needed to gather information,
but they may experience difficulties in transmitting data.
The caliber of indicators collected by the receptors and transmitted to the net-
works mostly depend on the routers—MAN, LAN, and also WAN. These networks
have to be well connected via various systems to facilitate fast and high-quality
reception. Though the variety of gadgets that are connected to the networking cov-
erage, and that generates keeping track of as well as keeping track of issues.

3.4.5 Handling Unstructured Data

An increasing number of connected products also increase the difficulties of dealing


with unstructured details on the volume, velocity, and type. However, an actual
struggle for businesses is determining what information is valuable, as only quality
information is actionable. Unstructured details cannot be kept in an SQL structure.
Unstructured information kept in a NoSQL structure makes the retrieval of informa-
tion a little complicated. With the launch of Big Data frameworks, such as Cassandra
and Hadoop, the problems with unstructured details have decreased. However,
merging the fundamental data with IoT causes problems. In addition, at this time,
there are not any common recommendations for the use and retention of informa-
tion or metadata.
3 Enterprise IoT Modeling: Supervised, Unsupervised, and Reinforcement Learning 75

3.4.6 Data Capturing Capabilities

A challenge of recording information is to transform the information collected from dif-


ferent sources into a common structure that can be analyzed and automated. According
to a HubSpot article, sponsored by Par Stream [25], 86% of company stakeholders assert
that information is essential to their IoT tasks, but just 8% are actually able to shoot as
well as evaluate IoTin formation wearing a regular fashion, as shown in Fig. 3.11.
The IoT depends on receptors for networks and indicators for that division, so
particular anomalies in deep runtime, like a power outage, could cause incorrect
data to be captured.

3.4.7 Intelligent Analytics

The intent behind IoT is to convert data into substantial information. Flaws in the
data or information design could result in false conclusions. We have to understand
the info inside itself is not a comprehension, type of right problems have to become
guided from the actual info reach the understanding.
According to an article by Hubspot [25], 42% of IoT stakeholders experience
issues with recorded information, and 30% stated that their analytic abilities were
not effective or adaptable, as shown in Fig. 3.12.
Historical methods, including standard analytics programs in which only limited
information can be examined at any given time, can certainly restrict the ability to
control real-time details. The following are some difficulties that prevent smart
analytics:
• Unpredictable activity of a computer throughout an incident.
• Traditional analytics software.

Fig. 3.11 Data capturing


76 R. K. Dhanaraj et al.

Difficult to capture useful data 36%

Data is not captured reliably 25%

Data is captured too slowly to be actionable 19%

Too much data to analyze effectively 44%

Analysis capabilities are not flexible enough to ask the questions we want 30%

Data is analyzed too slowly to be actionable 26%

We’re not sure what questions to ask 27%

Business processes are too rigid to allow us to act on the analysis 24%
Other 7%
We have no challenges 6%

0% 5% 10% 15% 20% 25% 30% 35% 40% 45% 50%


Data capture Data analysis Acting on analysis

Fig. 3.12 Challenges with collecting and analyzing data in an IoT project

• Slow adoption of current engineering as a result of the excessive cost.


• Lack of competent workers in information mining, algorithms, machine learn-
ing, and complicated processing [26].

3.4.8 Data Security and Privacy Issues

Even prominent companies such as Apple and visionaries such as Elon Musk have
not been spared by online hackers. Recent occurrences of ransomware strikes have
inhibited the confidence of many companies. One analysis reported that by 2020,
25% of cyber-attacks will focus on IoT products [24]:
• Malware infiltration: 24%
• Phishing attacks: 24%
• Social engineering attacks: 18%
• Device misconfiguration issues: 11%
• Privilege escalation: 9%
• Credential theft: 6%
Lapses in cyber-safety could occur in any organization and affect customers.
Therefore, it is crucial for every business to take measures to enhance security. A
report disclosed that 45% of IoT unit proprietors do not use any kind of third-party
protection application. In addition, 35% of these individuals do not change the
default password on their gadgets. Thus, buyers and businesses must collaborate to
implement effective security policies in IoT setup [20].
3 Enterprise IoT Modeling: Supervised, Unsupervised, and Reinforcement Learning 77

3.4.9 Consumer Awareness

Many individuals are not aware of the IoT, although they fully understand the
dependency on smart apps for information, stocks, and entertainment. It is not really
essential for customers to understand how elements function commercially.
However, a lack of fundamental knowledge may cause anxiety about price and
security, which may result in the slow adoption of technologies.
As outlined in a survey of 3000 Canadian and U.S. customers carried out by
Cisco, 53% of customers would not choose to individual information collected,
notwithstanding of this unit. This indicates that subscribers are reluctant to disclose
their information, which may serve as a deterrent to IoT implementation [20].

3.4.10 Delivering Value

Based on a Forbes Insights Survey[23], 29% of professionals are feeling a signifi-


cant struggle in creating IoT abilities stands out as the quality of IoT. This informa-
tion indicates the challenge of IoT program growth businesses with obtaining
valuation for their customers. Thus, prior to plunging straight into the improvement
of IoT programs, a business needs to determine what value they are likely to provide
via what abilities, as well as how the solution will enhance effectiveness and effi-
ciency, while simultaneously producing customer satisfaction.
Because the IoT is about connected things, the IoT tasks also need a great degree
of help in this manner. Approximately 50% of businesses with IoT initiatives are
clearly associated with IT services, suppliers, or consultation services companies,
depending upon them to assist with shipping, delivery, and supply company recom-
mendations. Feel foundation, creating an IoT Development Company, that thinks
engineering away from labor as well as like function by merging all the components
of IoT with in a manner that is based on connectivity, improving recognition and
holding accuracy of all of the phases [20]. Nevertheless, continue a range of
enhancements of produce abilities to enhance the effectiveness of the item or maybe
system depending on the most recent engineering.

3.5 Conclusion

The modern era of the IoT and AI will greatly improve existing tasks. With hands-­
free operation and in-depth evaluations, companies can enjoy the advantages of
progress while maximizing profits. The challenges of this era are to make better use
of AI and the IoT for the new generation. The distinctive, infrastructure-related
expenses of IoT systems is required for predicting the consistent hardware models.
78 R. K. Dhanaraj et al.

With these brand-new versions, IoT companies need to provide recurring value to
their clients in order to achieve success. Effective IoT businesses are delivering
recurring value to their customers by applying service-­based internet business ver-
sions and monetizing consumer information. To ensure that the designs catch the
attention of clients, successful IoT companies should also supply recurring value in
exchange. The value can be made available in a number of ways. As the IoT busi-
ness matures, distinct products have started to appear. IoT products can supply
recurring value by improving operational efficiencies, facilitating compliance,
enhancing sales, and opening new sales channels.

References

1. Kumar, D. R., Krishna, T. A., & Wahi, A. (2018). Health monitoring framework for in time
recognition of pulmonary embolism using internet of things. Journal of Computational and
Theoretical Nanoscience, 15(5), 1598–1602.
2. Anandhalli, M., & Baligar, V. P. (2017). A novel approach in real-time vehicle detection and
tracking using raspberry pi. Alexandria Engineering Journal, 57(3), 1597–1607.
3. Atzori, L., Iera, A., & Morabito, G. (2010). The internet of things: A survey. Computer
Networks, 54(15), 2787–2805.
4. Kamath, G., Agnihotri, P., Valero, M., Sarker, K., & Song, W. -Z. (2016). Pushing analytics to
the edge. In GLOBECOM, IEEE, pp 1–6.
5. Deepak, S., & Anandakumar, H. (2019). AODV route discovery and route maintenance in
MANETs. In 2019 5th International Conference on Advanced Computing & Communication
Systems (ICACCS). https://doi.org/10.1109/icaccs.2019.8728456.
6. Nishiguchi Y, Yano A., Ohtani, T., Matsukura, R., & Kakuta, J. (2018). Iotfault management
platform with device virtualization. In 2018 IEEE4th World Forum on Internet of Things
(WF-IoT), pp. 257–262.
7. Qiu, J., Wu, Q., Ding, G., Xu, Y., & Feng, S. (2016). A survey of machine learning for big data
processing. EURASIP Journal on Advances in Signal Processing, 2016, 67.
8. Redmon, J., Divvala, S., Girshick, R., & Farhadi, A. (2016). You only look once: Unified, real-­
time object detection. In The IEEE conference on Computer Vision and Pattern Recognition
(CVPR), pp. 779–788.
9. Kumar, R. N., Karthick, S., Valarmathi, R. S., & Kumar, D. R. (2018). Design and analysis
of multiply and accumulation units using low power adders. Journal of Computational and
Theoretical Nanoscience, 15(5), 1712–1718.
10. Anandakumar, H., & Umamaheswari, K. (2017). Supervised machine learning techniques in
cognitive radio networks during cooperative spectrum handovers. Cluster Computing, 20(2),
1505–1515.
11. Howard, A. G., Zhu, M., Chen, B., Kalenichenko, D., Wang, W., Weyand, T., Andreetto, M.,
& Adam, H. (2017). Mobilenets: Efficient convolutional neural networks for mobile vision
applications. Computer Vision and Pattern Recognition (cs.CV).
12. Iandola, F. N., Han, S., Moskewicz, M. W., Ashraf, K., Dally, W. J., & Keutzer, K. (2016).
Squeezenet: Alexnet-level accuracy with 50x fewer parameters and <05mbmodel size.
arXiv:1602.07360
13. Rajesh Kumar, D., & Shanmugam, A. (2017). A hyper heuristic localization based cloned node
detection technique using GSA based simulated annealing in sensor networks. In Cognitive
computing for big data systems over IoT, pp. 307–335.
3 Enterprise IoT Modeling: Supervised, Unsupervised, and Reinforcement Learning 79

14. Bansod, G., Raval, N., & Pisharoty, N. (2015). Implementation of a new light weight encryp-
tion design for embedded security. IEEE Transactions on Information Forensics and Security,
10, 142–151.
15. Cecchinel, C., Jimenez, M., Mosser, S., & Riveill, M. (2014). An architecture to support the
collection of big data in the internet of things. In: 2014 IEEE World congress on services,
IEEE, pp. 442–449.
16. Anandakumar, H., & Umamaheswari, K. (2018). A bio-inspired swarm intelligence technique
for social aware cognitive radio handovers. Computers and Electrical Engineering, 71, 925–
937. https://doi.org/10.1016/j.compeleceng.2017.09.016.
17. Varghese, B., Wang, N., Barbhuiya, S., Kilpatrick, P., & Nikolopoulos, D. S. (2016, November
18–20). Challenges and opportunities in edge computing. In 2016 IEEE international confer-
ence on smart cloud, Smart Cloud 2016, New York.
18. Zhang, D., et al. (2018). Review on the research and practice of deep learning and reinforce-
ment learning in smart grids. CSEE Journal of Power and Energy Systems, 4, 362–370.
19. Chen, G., Parada, C., & Heigold, G. (2014). Small-footprint keyword spotting using deep
neural networks. In ICASSP, IEEE, pp 4087–4091.
20. Doshi, R., Apthorpe, N., & Feamster, N. (2018, May). Machine learning DDos detection for
consumer internet of things devices. In 2018 IEEE Security and Privacy Workshops (SPW),
pp. 29–35.
21. Dastjerdi, A. V., & Buyya, R. (2016). Fog computing: Helping the internet of things realize its
potential. IEEE Computer, 49(8), 112–116.
22. Dhiviya, S., Malathy, S., & Kumar, D. R. (2018). Internet of things (IoT) elements, trends and
applications. Journal of Computational and Theoretical Nanoscience, 15(5), 1639–1643.
23. https://www.forbes.com/forbes-insights/our-work/internet-of-things/
24. Ioannis Stellios, Panayiotis Kotzanikolaou, Mihalis Psarakis, Cristina Alcaraz, Javier Lopez,
A Survey of IoT-Enabled Cyberattacks: Assessing Attack Paths to Critical Infrastructures and
Services. IEEE Communications Surveys & Tutorials 20 (4):3453-3495
25. https://blog.hubspot.com/marketing/iot-retail
26. https://www.statista.com/statistics/607716/worldwide-artificial-intelligence-market-revenues/
27. https://www.gartner.com/smarterwithgartner/five-steps-to-address-iot-integration-challenges/
Chapter 4
An Overall Perspective on Establishing
End-to-End Security in Enterprise IoT
(E-IoT)

Vidya Rao, K. V. Prema, and Shreyas Suresh Rao

4.1 Introduction

IoT is a vast network of networks consisting of physical and virtual interconnected


entities. These entities have unique addressing schemes and interact with each other
to provide certain customized or generic services. In 2012, the International
Telecommunication Union (ITU-T) recommended a standard definition of IoT as “a
global infrastructure for the information society, enabling advanced services by
interconnecting virtual and physical things based on existing and evolving interop-
erable information and communication technologies [1]”. Technically speaking,
IoT has its applications in diverse areas like healthcare, surveillance, transport,
security, manufacturing, environmental monitoring, and food processing, and it is
integrated with technologies like autonomic networking, decision making, machine-­
to-­machine communication, cloud computing, big data analytics, confidentiality
protection, and security [2].
Enterprise Internet of Things (E-IoT) is the next level of sensor technology that
connects every physical object to form a vast network of embedded computing devices.
These devices are generally made up of tiny components. They have constrained pro-
cessing capabilities, low memory, and limited power resources. This emerging tech-
nology has reduced manual intervention and has increased business efficacy.
Gartner Press released an article in August 2019 showcasing that by 2020, there
would be about 5.8 billion IoT endpoints, as compared to 4.8 billion endpoints d­ uring
2019. That means there is almost a 21% increase in the addition of new endpoints.

V. Rao (*) · K. V. Prema


Manipal Institute of Technology, Manipal Academy of Higher Education,
Manipal, Karnataka, India
e-mail: prema.kv@manipal.edu
S. S. Rao
Sahyadri College of Engineering and Management, Mangalore, Karnataka, India

© Springer Nature Switzerland AG 2020 81


A. Haldorai et al. (eds.), Business Intelligence for Enterprise Internet of Things,
EAI/Springer Innovations in Communication and Computing,
https://doi.org/10.1007/978-3-030-44407-5_4
82 V. Rao et al.

These endpoints are categorized under various use cases like utilities, government
buildings, automation, physical security, healthcare providers, manufacturing and
natural resources, information and transportation, retail, and wholesale. Among these
use cases, utilities have taken a major share of 17% with 1.33 billion endpoints with
applications like electric smart grid, smart metering, and smart electricity supply.
Apart from this, physical security application surveillances, intruder systems, as well
as CCTVs have taken about 0.70 billion endpoints.
These endpoints have generated a total revenue of about $389 billion in countries
like North America (NA), Greater China (GC), and Western Europe (WE). Statistics
of Gartner’s study have shown that about 75% of the revenue would be generated by
electronic endpoints in the world, i.e., about $120 billion revenue from NA and
$91 billion and $82 billion revenue from GC and WE, respectively, by 2020. It is
expected that the two main use cases that shall take a good share in the electronic
revenue are connected to consumer cars and networkable printing and photocopying
with $71 billion and $38 billion revenue, respectively. Then comes the government
indoor and outdoor surveillances that add on to the revenue as the government is
considering civilian security as its top priority.
These endpoints are enabled with various sensors like cameras, proximity sen-
sors, temperature sensors, air quality sensors, flow sensors, and many more sensors
that are unprotected. The reason is that the agent-based technologies do not protect
them from various attacks like distributed denial of service (DDoS), ransomware
attack, stealing of sensitive intellectual properties, cryptojacking attacks, etc. [3].
This is a major cause of concern as the data produced by these devices consists of
user health information, bank details, passwords, location information, and many
more. Hence these devices are subjected to security threats due to (a) malicious or
compromised node in the network, (b) defective manufacturing, and (c) presence of
an external adversary. There may also be threats to security initiated by nature.
These natural threats include earthquakes, floods, fire, and hurricanes that cause
severe damage to the computer systems. As it is hard to safeguard against natural
calamities, it is advisable to reduce the damage by collecting backup of data through
a contingency plan. Similarly, there could be human threats that can be classified
under information-level attacks, adversary location attacks, access-level attacks,
and host-based attacks. To enable the security of the devices, it is essential to select
the hardware components that have the following properties: default authentication
capabilities, end-to-end traffic encryption, secure boot loading process, enforce-
ment of digital signatures during firmware update, and transparent transactions.
Also, it has been identified that there are almost 1.1 billion data points created
every week, with 2.5 billion GB of data being generated across the world. Likewise,
about 500 GB of data is generated by offshore oil rigs and 100 GB of data from oil
refineries per week. Also about 10,000 GB of data is generated by jet engines every
30 min. Overall, it is said that about 90% of the world’s data has been generated in
the last 2 years. Thus, when such a huge amount of data points and data are available
on the public network like the Internet, they are susceptible to various attacks.
Hence, it is essential to identify the possible security safeguards at the earliest.
4 An Overall Perspective on Establishing End-to-End Security in Enterprise IoT (E-IoT) 83

With the growth of connected devices under IoT, there is an increase in the poten-
tial vulnerability on security, privacy, and governance. Though IoT can make people’s
life convenient, it might fail to ensure security and privacy of the user data leading to
a number of undesirable consequences. For example, in 2015, IoT baby monitors
were hacked through which the hackers were able to monitor the live feeds of the
baby, change the camera settings, and authorize other users remotely to view and
control the baby monitor [4]. During 2017, intruders could over-write the part of
Ukraine’s power grid that caused the first cyber attack [2]. Even the Internet-­connected
cars and wearable devices can also become a threat to the user’s security and privacy.
In [5] Atmali et al. have analyzed the impact of the above attacks on IoT applica-
tions like power management, smart car, and the smart healthcare system. Through
their study, they have projected that there is a need for security and privacy consid-
erations at the level of (a) actuators, (b) sensors, (c) RFID tags, and (d) the Internet/
network. Attack on actuators in power management applications can lead to finan-
cial loss due to excessive power consumption. Similarly, in smart cars, these com-
promised actuators may control the broken system costing a driver’s life. Also, in
the healthcare system, these compromised actuators can inject the wrong dosage of
medicine to a patient who is remotely monitored by the doctor. Likewise, a compro-
mised sensor can fake the data that may lead to the wrong diagnosis of a patient. At
the same time, these compromised nodes can reveal the personal information of the
patient or the data related to a user’s home through power management system.
Section 4.2 of this chapter explains the various security threats and attacks, fol-
lowed by elements of security in Sect. 4.3, some of the lightweight existing solu-
tions in Sect. 4.4, threat modeling tools in Sect. 4.5, Kali Linux-based ethical
hacking in Sect. 4.5.4, major IoT security practices of E-IoT in Sect. 4.6, and lastly,
conclusion in Sect. 4.7.

4.2 Security Threats and Attacks

Devices within the IoT communicate personalized data of many users. This data
consists of user health information, bank details, passwords, location information,
and many more. These devices are subjected to security threats like (a) malicious
node in the network, (b) defective manufacturer, and (c) external adversary [5].
These threats lead to security attacks that can be initiated either by nature or human.
The natural threats may include earthquake, floods, fire, and hurricane that cause
severe damage to the computer system. Although it is hard to safeguard against
natural calamities, it is advisable to reduce the damage by collecting backup of data
through contingency plan. Accordingly, the security attacks caused by the humans
affect the node privacy [6, 7]. Such attacks can be classified as follows [5, 8]:
1. Information-level attacks: All IoT devices are enabled with sensors that record
the data from the physical environment and communicate the information over
84 V. Rao et al.

the Internet. As the Internet is an open domain, attackers can easily tamper the
information under following categories [7–9]:
(a) Denial of service (DoS): DoS is an attack over the network component that
makes it unavailable for an intended user.
(b) Masquerade: An intruder behaves as an intended user and tries to talk with
the network component.
(c) Modification of message: An intruder can alter or delete or fake a message
sent by a legitimate user.
(d) Man-in-the-middle (MITM): MITM is a kind of attack wherein a malicious
user takes control of the communication channel between two or more
endpoints.
(e) Message replay attack: It is a security breach in which the message is stored
by malicious node without the knowledge of intended users, and the mali-
cious node transmits an altered message that is forwarded to the receiver.
2. Adversary location attack: An intruder can be present in any part of the IoT eco-
system. He can either be within or outside the IoT environment [9]:
(a) Internal attack: An attack caused by the components within the IoT border.
It is also called as insider attack where the intruder tries to inject malicious
code toward the IoT components.
(b) External attack: An attack caused by an advisory that is located outside the
IoT environment in a remote place.
3. Access-level attack: Access-level attacks are broadly classified into active and
passive attacks [10]. In the passive attack, an attacker can read the packet that is
transmitted, but he/she cannot alter the packets like eavesdropping and traffic
analysis. On contradictory, in active attack, the attacker sees the data and then
alters the content of the data and transmits the altered data back to the network.
4. Host-based attack: Many devices in an IoT environment are made up of different
manufacturers [10]. These devices are subjected to user compromise attack, soft-
ware compromise attack, and hardware compromise attack. This is because the
manufacturer can hold the devices’ information which can be misused by him.
Hence the production of such poorly secured goods results in compromising the
user privacy. At the same time, any manufacturer can attack his competitors
through their devices.

4.2.1 IoT Four-Layered Architecture and Associated Attacks

P. P. Ray [10] has surveyed various domain-based architectures that vary from RFID
to healthcare to security to cloud services. But in general, a four-layered design of
IoT is considered for different research as in Fig. 4.1. Mainly it comprises of per-
ception layer, network layer, transport layer, and application layer. Each layer has its
own properties and protocols. Primarily, the perception layer forms the physical
4 An Overall Perspective on Establishing End-to-End Security in Enterprise IoT (E-IoT) 85

Fig. 4.1 Generic


four-­layered IoT
architecture

layer of the IoT ecosystem. It deals with sensors, devices, machines, actuators, and
movements of unprocessed raw data. In this layer, the data transmission medium
used is copper wire, coaxial cable, or radio wave. They have protocols like IEEE
802.3, Wi-Fi, LR-WPAN, 2G, 3G, 4G, and LTE networks [11].
Next is the Internet layer, which is also called the network layer. The main job of
this layer is to provide host identification and packet routing. IETF has proposed
many routing protocols that are suitable for low-powered device networks. Some of
the protocols are IPv6, IPv4, RPL, 6LoWPAN, multipath RPL (MRPL) [12],
energy-efficient probabilistic routing protocol (EEPR) [13], congestion avoidance
multipath routing protocol (CA-RPL) [14], movement-aided energy balance
(MABE) [15], least path interface beaconing protocol (LIBP) [16], and cognitive
machine-to-machine RPL (CoRPL) [17].
Then comes the transport layer which is considered for end-to-end message
transfer. The transmission can be either connection-oriented or connectionless with
protocols like transmission control protocol (TCP) and user datagram protocol
(UDP), respectively. This layer involves various processes like segmentation and
reassembly of packets, congestion control, error control, and flow control.
Lastly, the application layer interfaces with all the lower layers by establishing a
secure connection between the devices and servers. It uses standard port 80 and port 22
for most of HTTP and SSH protocols, respectively. Some of the protocols standardized
by IETF are constrained application protocol (CoAP), message queuing telemetry
transport protocol (MQTTP), extensible message and presence protocol (XMPP), data
distribution services (DSS), and advanced message queuing protocol (AMQP) [11].
Likewise there are various attacks based on layers of IoT as shown in Fig. 4.2 [3,
18]. Sensing/perception layer is generally made up of sensors, RFIDs, NFCs,
ZigBee, Bluetooth, and other intelligent hardware devices. These devices are
exposed to more external attacks like node compromise attack, fake node injection,
86 V. Rao et al.

Fig. 4.2 Attacks based on architecture

access control, and RF interference on RFIDs. The second layer is the Internet layer
and is subjected to attacks like address compromise attacks, routing information
attack, RFID spoofing, and sinkhole attack. The next layer is the transport layer that
experiences attacks like denial of service (DoS), masquerade, distributed DoS
(DDoS), man-in-the-middle (MITM) attack, and session hijacking. And finally, the
application layer experiences attacks like phishing attack, viruses, worms, mali-
cious scripting, revealing of sensitive data, user authentication attacks, software vul-
nerability, and stealing of intellectual property.

4.2.2 Attacks Based on Phases of IoT

IoT can also be defined as an interconnection of “factual and virtual” objects placed
across the globe that are attracting the attention of both “makers and hackers.” IoT
can be divided into five different phases as mentioned in [19] by Jeyenthi as shown
in Fig. 4.3. The first phase is termed as the data collecting phase: primary interface
between physical environment and sensors. There can be either static objects like
body sensors or RFIDs or dynamic objects like sensors and chips on vehicles. The
second phase is the storage phase: as many IoT devices are having low self-storage
capability, IoT provides a server or cloud-based storage. Next is the intelligent pro-
4 An Overall Perspective on Establishing End-to-End Security in Enterprise IoT (E-IoT) 87

Fig. 4.3 Phases of IoT and their possible attacks

cessing phase: it is where the analysis of stored data and later appropriate services
are provided to the users. IoT devices can be queried and controlled remotely using
the results obtained after processing of data. The fourth phase is data transmission:
it deals with processing of data communication between all of the above phases.
Last is the delivery phase: it is where the activity of delivering the processed data to
all the objects in time without being altered or hacked is performed.
Among the five phases, the data perception phase is subjected to more attacks
like data leakage, data authentication, and data loss as the devices are easily avail-
able to users and hackers. Similarly, in storage phase, we can see attack on avail-
ability, modification of message, denial of service (DoS) attack, attack on integrity,
and data fabrication. Attacks on authentication are seen at the processing phase, and
channel security attack, session hijacking, routing protocol attack, and flooding are
seen at the transmission phase. Lastly, at the delivery phase, man- and machine-­
made attacks are found as shown in Fig. 4.3.

4.3 Elements of Security

To ensure the IoT security, there are four elements of security [18]. They are device
authentication, secure connections, secure data storage, and lastly, secure code exe-
cution. The device authentication grants the access privilege of the devices to the
legitimate users. Secure connection enables the protection of the data that is travel-
ling across the network (data in motion). Secure storage provides protection for data
in rest using various lightweight encryption schemes. And lastly, the secure code
execution serves the intended host machines to use the data and process it in a
secure manner as in Fig. 4.4.
88 V. Rao et al.

Fig. 4.4 Elements of IoT security [18]

As these poorly secured IoT devices can serve as means of entry point for cyber
attackers by allowing various malicious individuals to re-program a device and cause
malfunctioning, it becomes essential to provide security and privacy at the devices
level. In order to develop a safer IoT solution, it is required to consider three major
security requirements: (i) confidentiality, (ii) integrity, and (iii) authentication [20].
• Confidentiality means keeping information secret from the unauthorized user.
For example, when transmitting certain sensitive data like location of military
camp to the base station, it must be forwarded in secrecy to avoid intruders to
understand the information that is being transmitted.
• Data integrity ensures that the messages transmitted are reached at the destina-
tion unaltered. Data integrity certifies the user that it has never been altered or
corrupted by protecting the data over a communication channel.
• Authentication is a process of determining whether the data is transmitted by
legitimate users or not. The user needs to identify the peer nodes that they need
to communicate.

4.4 Lightweight Secure Measures for IOT

Elliptic curve cryptography (ECC) was introduced in early 1985 by Neal Koblitz
and Victor Miller [21]. They stated that the hardness of ECC security depends on the
discrete logarithmic problem defined on the elliptic curve. Later, Gura et al. [22]
experimented ECC and RSA on an 8-bit CPU to compare their performance and
found that the use of ECC for a lower-bit processor provides the same level of secu-
4 An Overall Perspective on Establishing End-to-End Security in Enterprise IoT (E-IoT) 89

rity as that of RSA. Later during 2013, Wenger [23] developed an ECC-based access
control scheme over a prime field on 16-bit MSP430 micro-controller whereby the
results confirmed the feasibility of ECC for resource-constrained devices.
Basically, ECCs are often implemented by using a static public elliptic curve that
is shared among all the users in the network. In [24] the recommended elliptic curve
domain parameters are provided for the Weierstrass curve equation y2 = x3 + ax + b
that is accepted by various researchers [25]. Liu et al. [26] have proposed software
and hardware architecture for resource-constrained embedded devices. Their work
has shown the feasibility of ECC on the embedded system. But the use of a fixed
elliptic curve can be challenged on intensive cryptanalysis. Wang et al. [27] made a
study on using a fixed prime field to build a crypto-system for applications devel-
oped for different processors varying from 8 bits to 256 bits.
A lightweight multi-message and multi-receiver heterogeneous-based signcryp-
tion is proposed by Rahaman et al. [28]. They have used the hybrid elliptic curve to
generate signatures. The work is evaluated for various attacks like replay attack,
forward secrecy attack, and unforgeability using the AVISPA simulator tool. For the
heterogeneous environment, the attackers are inclined to impersonate legitimate
users. To solve such an issue, Jingwei Liu et al. [29] have proposed a novel authen-
tication scheme. They have provided a lightweight anonymous authentication and
key agreement scheme as proposed. Their scheme could toggle between the public
key infrastructure (PKI) and certificates analysis. Their method showed resistance
against replay and DoS attacks.
The combination of cloud-based services with IoT has raised the issue of limita-
tion regarding low latency and high mobility. To address such issues, Haldorai et al.
[30] have proposed the authentication and key agreement scheme for fog-based IoT
for the healthcare application. By using bilinear key agreement protocol, they have
proposed a protocol that showed resistance against MITM, replay attack, known-­
session key attack, and intractability.
Recently, based on card shuffling logic, a data confidentiality algorithm is
designed using ECC, proposed by Khan [31]. The use of random card shuffling has
shown double encryption and increased the security of the algorithm. The algorithm
can encrypt or decrypt any type of ASCII values. As the algorithm uses ECC, it is
suitable for resource-constrained devices. Li et al. [32] proposed a lightweight
mutual authentication protocol using public-key encryption schemes for smart city
applications. Their simulated work has shown a balance among ciphertext size,
usability, and efficiency. The generation of online and offline signatures created
overhead on the device storage. Diro et al. [33] have used ECC to provide ­lightweight
encryption for fog-based IoT applications. They have shown better efficiency
regarding runtime, throughput, and ciphertext expansion. But they could only han-
dle a smaller data size.
An OTP-based end-to-end authentication scheme was proposed by Shivraj et al.
[34]. Their scheme used Lamport’s OTP scheme with ECC-based authentication
algorithm. Even though the scheme performed better than existing OTP-based sig-
nature schemes, they could not justify the implementation on a real-time scenario.
A security framework for IoT and cloud computing is proposed by Daisy Premila
90 V. Rao et al.

et al. [20]. They used ECC-based message encryption and multi-factor authentica-
tion to ensure confidentiality, integrity, privacy, and authentication. They have con-
cluded that the use of ECC-based security measures is better than RSA to eliminate
the ambiguity and enhance security. But the research to collaborate IoT and cloud
computing needs to depend on infrastructure.
During 2018, to address the usage of the static curve in ECC, Jia Wang et al.
[35] proposed a dynamic elliptic curve-based Internet of Vehicles (IoV) network.
Their work showed good computational efficiency and security for a smaller key
size. But storing the elliptic curves as a plain text in embedded systems would lead
to security concern. To address the data integrity issue of Java card-based applica-
tion, Gayoso et al. [36] initiated the use of ECC-based encryption algorithm called
an elliptic curve integrated encryption scheme (ECIES) and concluded that ECIES-
based encryption is the best among encryption schemes for resource-constrained
devices.

4.5 Threat Modeling for IOT Security

Threat modeling (TM), whose lifecycle is depicted in Fig. 4.6, is a process of iden-
tifying the potential threats, enumerating and prioritizing the threats, and providing
countermeasures to mitigate the threats. TM can be applied to any platform of a
working process like software, application, networks, IoT devices, or business pro-
cesses. Shostack [37] has summarized the reasons to incorporate a threat model in
SDL which are (i) to find the bugs at the earliest, (ii) understand the security require-
ments, and (iii) engineer and deliver a better product. Basically, TM includes com-
ponents like target-of-evaluation (ToE) (a design or model of what type of platform
needs to be analyzed), a list of assumptions that can be threats on ToE, a list of
potential threats on ToE, possible countermeasures toward the identified threats,
and verification of success (VoS) that validates the threat model.
Before modeling a threat, there are four questions that need to be answered,
which are as follows:
1. What are we building? A detailed data flow diagram (DFD) is designed by speci-
fying various roles and responsibilities of each participant.
2. What can go wrong? The various possible threats are analyzed using methods
available in STRIDE, PASTA, STRIKE, or VAST.
3. What are we going to do about that? Potential mitigation strategies against the
threats are framed.
4. Did we do a good job? Once the mitigation is applied, the system is validated for
the stability and security against the threats.
4 An Overall Perspective on Establishing End-to-End Security in Enterprise IoT (E-IoT) 91

4.5.1 Microsoft Security Development Lifecycle

Microsoft SDL was introduced during 2008 to ensure security and privacy consid-
erations throughout all the phases of the development process. This helped develop-
ers to build highly secure software, address security compliance requirements, and
reduce development cost. The core of Microsoft SDL is threat modeling. Threat
modeling helps in shaping the application design and meeting the security objec-
tives of the company by reducing the risk severity. The five major steps of threat
modeling involve (Fig. 4.5) the following:
1. Defining security requirements: To understand the ecosystem of the device, i.e.,
analysis of the ToE by framing various use cases. In this process the external and
internal assets are identified.
2. Creating an application diagram: Here a detailed data flow diagram of the pro-
posed ToE is framed with appropriate trust boundaries and security requirements
for each participant.
3. Identifying the threats: Microsoft TMT follows STRIDE-based threat modeling
where the threats are identified. Potential adversaries are identified under four
categories called remote software attacker, network attacker, malicious insider
attacker, and advance hardware attacker.
4. Mitigating the threats: For the threat identified, relevant countermeasures are
established.
5. Validating that threats have been mitigated: Finally, the verification of the threat
model against the mitigation is performed to check the stability of the proposed
system.

4.5.2 STRIDE Framework Methodology

It is important to develop a secure design for any software application or system.


Failing to do so may cost about 30 times higher than estimated cost [38]. Hence
threat modeling plays a vital role in the software development lifecycle. Among
various threat modeling methods like STRIDE, PASTA, VAST, and STRIKE,

Fig. 4.5 Microsoft


Security Development
Lifecycle (SDL) using
TMT
92 V. Rao et al.

Table 4.1 STRIDE threat model with associated security properties


Security
Threats property Definition
Spoofing Authentication Unauthorized access
Using another user’s identity
Tampering Integrity Malicious modification
Unauthorized information changes
Repudiation Non-repudiation Denying to perform action
Information Confidentiality Unprivileged user gains access and compromises the
disclosure system
Denial of service Availability Denying services to valid users
Threats to system availability and reliability
Elevation of Authorization Exposure of information to individuals not supposed
privilege to access

STRIDE has taken a major share among the industrial development processes [39,
40]. STRIDE is developed by Microsoft as a part of their Security Development
Lifecycle. STRIDE is an acronym for spoofing, tampering, repudiation, information
disclosure, denial of service, and elevation of privilege [41]. The security properties
and attack types associated with STRIDE are summarized in Table 4.1 [38].

4.5.3 Overview of Threat Modeling Tool (TMT)

Microsoft TMT is used to provide assistance in analyzing the design of a system or


an application in order to check for security risks and provide solution for the threat
found. Figure 4.4 displays the initial page of TMT when launched. This page has
two partitions; the top part is used to create the threat model of the user’s choice
using the templates provided by the Microsoft, while the bottom part helps the user
to customize his own template on the default Microsoft Security Development
Lifecycle (SDL) template as in Fig. 4.6.

4.5.4 Kali Linux-Based Ethical Hacking

Kali Linux was developed by Mati Aharoni and Devon Kearns of Offensive Security
and was mainly suitable for digital forensic and penetration testing under ethical
hacking [42]. Kali Linux has approximately 300 hacking tools that are broadly cat-
egorized under information gathering, vulnerability analysis, wireless attacks, web
application, exploitation tools, forensic tools, sniffing and spoofing tools, password
attacks, maintaining access, reverse engineering, and hardware hacking tools.
Among these, the most commonly used tools are Metasploit framework, dsniff, tcp-
dump, Nmap, Wireshark, Aircrack-ng, Armitage, Burp Suite, BeEF, and so on [42].
4 An Overall Perspective on Establishing End-to-End Security in Enterprise IoT (E-IoT) 93

Fig. 4.6 Microsoft TMT initial screen

Featuring the rapid growth of smart cities, Barghuthi et al. [43] have made a
study of how the increase in the population of smart cities shall add to an increase
in the security breach and damage to businesses by 2050. Thus, they have proposed
Kali Linux-based vulnerability assessment and penetration testing solution using
low-cost Raspberry Pi 3 devices. Through their results, it has been concluded that
Raspberry Pi 3 can be used as a machine to check the vulnerability similar to any
traditional PC or laptop-based Kali Linux machine.
To replace the expensive and resource-intensive devices used for industrial vul-
nerability and assessment tests, Hu et al. [44] proposed an automated vulnerability
assessment using OpenVAS and Raspberry Pi 3 device. They have detailed methods
for analyzing the vulnerability assessment of distributed architecture. They made
the study on variables like CPU temperature, CPU usage, and CPU memory of the
device at the time of vulnerability assessment.
Visoottiviseth et al. [45] developed a GUI-based penetration testing tool called
PENTOS used for IoT devices. PENTOS runs on Kali Linux and is specifically
designed for the ethical hacking of wireless communication like Wi-Fi and Bluetooth.
PENTOS enables the analysis of password attack, web attack, and wireless attack that
ensure to gain access privilege of the various algorithms. They also have explained the
Open Web Application Security Project (OWASP) specified ten vulnerabilities of IoT
applications.
Finally, they have given the recommendations for the secure deployment of the
IoT environment. Denis et al. [46] performed various penetration tests using tools
available on Kali Linux. They were able to set up a private network and generate
94 V. Rao et al.

attack reports and visualize the reports using Kali Linux tools. The attacks they
performed were hacking phones, MITM attack, smartphone penetration testing,
spying, hacking phones’ Bluetooth, and hacking WPA-protected access, and then
they hacked the remote PC using IP and open ports.
Liang et al. [47] experimented on different methods of doing DoS attack using
Raspberry Pi-based Kali Linux. They have provided an attack framework and com-
pared various DoS attacks on their framework. They have used Hping3 with random
IP, SYN flood with spoofed IP, and TCP connection flood tools. The comparison
was made under the parameters like CPU utilization, memory utility, time for the
success of an attack, and packet loss rate. Ryan Murray [48] has proposed a forward-­
looking approach for a secure eHealth solution called HealthShare. It could share
data among various organizations that were hosting the patient’s data over the cloud.
Detailed steps as to conduction of MITM and DoS attack using tools like Ettercap,
Pexpect, manual SET, threads using the timer and Nmap timer, and Scapy have also
been provided.

4.6 Major E-IOT Security Practices

As E-IoT is deployed on a larger scale with heterogeneous business applications,


the cybersecurity space has obtained an intense research spectrum. Some of the
important security practices that should be followed by enterprise IoT are
explained below.
(a) Understand your endpoints: Every endpoint of the business network is assem-
bled by various manufacturers using different open-source operating systems.
These devices are potential entry points for cybercriminals. Thereby, it is essen-
tial to deploy devices in a tamper-proof environment using secure hardware and
software resources.
(b) Track and manage the endpoints: Business enterprise poses the responsibility of
constant check on the devices that are deployed under their network and should
be updated with frequent firmware and security patches. As it is infeasible to
monitor each device physically, Earl Perkins of Gartner Solutions has recom-
mended “rolling out an asset discovery, tracking, and management strategy” to
be implemented before the IoT project begins.
(c) Change the default passwords and other credentials: The manufacturers set
their devices with a common default password, which has to be updated by the
enterprise officials frequently. This is because, most of the time, hackers are
well aware of default passwords and sneak into your network by brute force
attacks.
(d) Execute risk-driven strategies: IoT projects need to be analyzed for risk possi-
bility using various threat modeling tools. Such tools help to identify the risks
in the network and guide the network administrator to take corrective actions.
Also, performing regular pen-testing at the hardware and software levels shall
ensure the attack resistivity of the network.
4 An Overall Perspective on Establishing End-to-End Security in Enterprise IoT (E-IoT) 95

(e) Consideration of the latest encryption protocols: Business enterprises should


encrypt the data passing from and to their network using the updated and latest
encryption schemes. If in case a single device is accessed by multiple users,
then the focus should be on user authentication, identity-level control, and pro-
viding data integrity.

4.7 Summary

IoT is a rapidly growing network that has its major contribution in making the busi-
ness enterprise smarter. E-IoT could connect to a diverse domain of applications and
devices across the globe thus leading to various levels of attacks and threats. Various
levels of hardware and software issues are studied with possible lightweight solu-
tions. A generalized layer of security architecture is discussed, followed by a brief
description on threat modeling tool. In addition, Kali Linux-based pen-testing on a
real-time E-IoT is also studied. Finally, the major E-IoT practices are generalized
that help future researchers to concentrate on the specific issues in E-IoT.

References

1. Telecommunication Standardization Sector of ITU. (2012). Series Y. Global information infra-


structure, internet protocol aspects and next-generation networks (pp. 1–6). Geneva: ITU.
2. Vermesan, O., & Friess, P. (2014). Internet of things-from research and innovation to market
deployment (Vol. 29). Aalborg: River Publishers.
3. Ahemd, M. M., Shah, M. A., & Wahid, A. (2017). IoT security: A layered approach for attacks
and defenses. In 2017 international conference on Communication Technologies (ComTech),
IEEE, pp. 104–110.
4. Khan, R., Maynard, P., McLaughlin, K., Laverty, D., & Sezer, S. (2016). Threat analysis of
blackenergy malware for synchrophasor based real-time control and monitoring in smart grid.
In ICS-CSR.
5. Atamli, A. W., & Martin, A. (2014). Threat-based security analysis for the internet of things.
In 2014 international workshop on Secure Internet of Things (SIoT), IEEE, pp. 35–43.
6. Stankovic, J. A. (2014). Research directions for the internet-of-things. IEEE Internet of Things
Journal, 1(1), 3–9.
7. AbdAllah, E. G., Hassanein, H. S., & Zulkernine, M. (2015). A survey of security attacks
in information-centric networking. IEEE Communication Surveys and Tutorials, 17(3),
1441–1454.
8. Abomhara, M., & Kien, G. M. (2015). Cyber security and the internet-of-things: Vulnerabilities,
threats, intruders and attacks. Journal of Cyber Security, 4, 65–88.
9. Nawir, M., Amir, A., Yaakob, N., & Lynn, O. B. (2016). Internet of things (IOT): Taxonomy
of security attacks. In 3rd International Conference on Electronic Design (ICED), IEEE,
pp. 321–326.
10. Ray, P. P. (2018). A survey on internet of things architectures. Journal of King Saud University-­
Computer and Information Sciences, 30(3), 291–319.
96 V. Rao et al.

11. Arulmurugan, R., Sabarmathi, K. R., & Anandakumar, H. (2017). Classification of sentence
level sentiment analysis using cloud machine learning techniques. Cluster Computing, 22(S1),
1199–1209.
12. Quynh, T. N., LeManh, N., & Nguyen, K. N. (2015). Multipath RPL protocols for greenhouse
environment monitoring system based on internet of things. In 2015 12th international confer-
ence on Electrical Engineering/Electronics, Computer, Telecommunications and Information
Technology (ECTI-CON), IEEE.
13. Park, S. -H., Cho, S., & Lee, J. -R. (2014). Energy-efficient probabilistic routing algorithm
for internet of things. Journal of Applied Mathematics, Hindawi Publishing Corporation, Vol.
2014, Article ID 213106, 7 pages.
14. Haldorai, A., & Ramu, A. (2019). Cognitive social mining applications in data analytics and
forensics (Advances in social networking and online communities). Hershey: IGI Global.
https://doi.org/10.4018/978-1-5225-7522-1.
15. Haoru, S., Wang, Z., & An, S. (2013). MAEB: Routing protocol for IoT healthcare. Advances
in Internet of Things, 3, 8–15. Scientific Research.
16. Ngqakaza, L., & Bagula, A. (2014, May 26–28). Least Path Interference Beaconing Protocol
(LIBP): A frugal routing protocol for the internet-of-things. In 12th international confer-
ence proceedings, Wired/Wireless Internet Communications- WWIC2014, Paris, France,
pp. 148–161.
17. Aijaz, A., & Aghvami, A. H. (2015). Cognitive machine-to-machine communications for
internet-­of-things: A protocol stack perspective. IEEE Internet of Things Journal, 2(2),
103–112.
18. Miller, L., & Johnson, C. A. (Eds.). (2016). IoT security for dummies. Chichester: Wiley.
19. Anandakumar, H., & Umamaheswari, K. (2017). Supervised machine learning techniques in
cognitive radio networks during cooperative spectrum handovers. Cluster Computing, 20(2),
1505–1515.
20. Bai, T. D. P., Rabara, S. A., & Jerald, A. V. (2015). Elliptic curve cryptography based security
framework for Internet of Things and cloud computing. In Conference on recent advances on
computer engineering by WSEAS, pp. 65–73.
21. Miller, V. S. (1985). Use of elliptic curves in cryptography. In Conference on the theory and
application of cryptographic techniques (pp. 417–426). New York: Springer.
22. Gura, N., Patel, A., Wander, A., Eberle, H., & Shantz, S. C. (2004). Comparing elliptic curve
cryptography and RSA on 8-bit CPUs. In International workshop on cryptographic hardware
and embedded systems (pp. 119–132). Berlin/Heidelberg: Springer.
23. Wenger, E. (2013). Hardware architectures for MSP430-based wireless sensor nodes perform-
ing elliptic curve cryptography. In International conference on applied cryptography and net-
work security (pp. 290–306). Berlin: Springer.
24. SEC, S. (2000). SEC 2: Recommended elliptic curve domain parameters. Standards for
Efficient Cryptography Group, Certicom Corp.
25. Silverman, J. H. (2009). The arithmetic of elliptic curves (Vol. 106). New York: Springer.
26. Liu, A., & Ning, P. (2008). TinyECC: A configurable library for elliptic curve cryptography in
wireless sensor networks. In Proceedings of the 7th international conference on Information
processing in sensor networks, IEEE Computer Society, pp. 245–256.
27. Wang, J., & Cheng, L. M. (2017). Dynamic scalable ECC scheme and its application to
encryption workflow design. In Proceedings of the international conference on Security and
Management (SAM), pp. 261–262.
28. Rahman, A. U., Ullah, I., Naeem, M., Anwar, R., Noor-ul-Amin, Khattak, H., & Ullah, S.
(2018). A lightweight multi-message and multi-receiver heterogeneous hybrid signcryption
scheme based on hyper elliptic curve. International Journal of Advanced Computer Science
and Applications, 9(5), 160–167.
29. Liu, J., Ren, A., Zhang, L., Sun, R., Du, X., & Guizani, M. A. (2019). Novel secure authentica-
tion scheme for heterogeneous internet of thing. arXiv preprint arXiv:1902.03562.
4 An Overall Perspective on Establishing End-to-End Security in Enterprise IoT (E-IoT) 97

30. Haldorai, A., Ramu, A., & Murugan, S. (2019). Smart sensor networking and green technolo-
gies in urban areas. In Computing and communication systems in urban development (pp. 205–
224). Cham: Springer. https://doi.org/10.1007/978-3-030-26013-2_10.
31. Khan, M. A. (2018). Multidisciplinary Journal of European University of Bangladesh. Cell,
1713(006814), 01914–098494.
32. Li, N., Liu, D., & Nepal, S. (2017). Lightweight mutual authentication for IoT and its applica-
tions. IEEE Transactions on Sustainable Computing, 2(4), 359–370.
33. Diro, A. A., Chilamkurti, N., & Nam, Y. (2018). Analysis of lightweight encryption scheme for
fog-to-things communication. IEEE Access, 6, 26,820–26,830.
34. Shivraj, V., Rajan, M., Singh, M., & Balamuralidhar, P. (2015). One time password authentica-
tion scheme based on elliptic curves for Internet-of-Things (IoT). In 5th National Symposium
on Information Technology: Towards New Smart World (NSITNSW), IEEE, pp. 1–6.
35. Wang, J., Li, J., Wang, H., Zhang, L. Y., Cheng, L. M., & Lin, Q. (2018). Dynamic scalable
elliptic curve cryptographic scheme and its application to in-vehicle security. IEEE Internet of
Things Journal, 6(4), 5892–5901.
36. Gayoso Martínez, V., Hernández Álvarez, F., Hernández Encinas, L., & Sánchez, C. (2011).
Analysis of ECIES and other cryptosystems based on elliptic curves. Journal of Information
Assurance and Security, 6(4), 285–293.
37. Shostack, A. (2014). Threat modeling: Designing for security. Indianapolis: Wiley.
38. Verheyden, L. (2018). Effectiveness of threat modelling tools. Master thesis.
39. Bodeau, D., McCollum, C., & Fox, D. (2018). Cyber threat modeling: Survey, assessment, and
representative framework. HSSEDI, The Mitre Corporation.
40. Meghanathan, N., Boumerdassi, S., Chaki, N., & Nagamalai, D. (2010, July 23–25). Recent
trends in network security and applications. In Third international conference, CNSA-2010,
Chennai, India, 2010 proceedings, Vol. 89. Springer.
41. Khan, R., McLaughlin, K., Laverty, D., & Sezer, S. (2017). Stride-based threat modeling for
cyber-physical systems. In 2017 IEEE PES Innovative Smart Grid Technologies conference
Europe (ISGT-Europe), IEEE, pp. 1–6.
42. K. Linux. (2016). Kali linux tools listing. https://www.kali.org
43. Al Barghuthi, N. B., Saleh, M., Alsuwaidi, S., & Alhammadi, S. (2017). Evaluation of portable
penetration testing on smart cities applications using raspberry pi III. In 2017 fourth HCT
Information Technology Trends (ITT), IEEE, pp. 67–72.
44. Hu, Y., Sulek, D., Carella, A., Cox, J., Frame, A., & Cipriano, K. (2016). Employing miniatur-
ized computers for distributed vulnerability assessment. In 2016 11th International Conference
for Internet Technology and Secured Transactions (ICITST), IEEE, pp. 57–61.
45. Visoottiviseth, V., Akarasiriwong, P., Chaiyasart, S., & Chotivatunyu, S. (2017). PENTOS:
penetration testing tool for internet of thing devices. In TENCON 2017–2017. IEEE region 10
conference, IEEE, pp. 2279–2284.
46. Denis, M., Zena, C., & Hayajneh, T. (2016). Penetration testing: Concepts, attack methods,
and defense strategies. In 2016 IEEE Long Island Systems, Applications and Technology
Conference (LISAT), IEEE, pp. 1–6.
47. Liang, L., Zheng, K., Sheng, Q., & Huang, X. (2016). A denial of service attack method for an
IOT system. In 2016 8th international conference on Information Technology in Medicine and
Education (ITME), IEEE, pp. 360–364.
48. Murray, R. (2017). A raspberry pi attacking guide. Birmingham: Packt Publishing.
Chapter 5
Advanced Machine Learning
for Enterprise IoT Modeling

N. Deepa and B. Prabadevi

5.1 Introduction

The Internet of Things (IoT) which is going to be the future enables every object
(any living or non-living thing) to interact through the Internet without any form of
human intervention like human-to-human or human-to-computer interactions.
Ultimately like the Internet, IoT will also become a fundamental need. It is also
termed as the Internet of Objects as it is an inter-connection of real-world entities
[1]. Any object to be a part of the IoT system must have a unique identifier and
capability to process the information. In terms of IoT, sensors and actuators allow
every object in the IoT system to communicate. IoT makes everything smart; here,
smart refers to making all the application-specific objects interact without any
human intervention. Some of the IoT applications are smart cities, smart medicines,
smart home, smart farming, and so on [2].
IoT makes activities smart by joining hands with various other technologies like
embedded systems (supporting hardware and software communication), sensors
(sensing live application-specific accurate information), big data analytics (manag-
ing colossal volume of data), cloud computing (data storage and scalability), and
machine learning (making machines learn and predict from experiences). IoT has
achieved its most significant impact on business through its variant named enter-
prise IoT (E-IoT). The term enterprise refers to a business that encompasses various
front-office and back-office activities. IoT in convergence with business automates
all the activities of an enterprise by embedding computing devices in enterprise
objects thus enabling them to communicate for accomplishing the business

N. Deepa (*) · B. Prabadevi (*)


School of Information Technology and Engineering, VIT University,
Vellore, Tamil Nadu, India
e-mail: deepa.rajesh@vit.ac.in; prabadevi.b@vit.ac.in

© Springer Nature Switzerland AG 2020 99


A. Haldorai et al. (eds.), Business Intelligence for Enterprise Internet of Things,
EAI/Springer Innovations in Communication and Computing,
https://doi.org/10.1007/978-3-030-44407-5_5
100 N. Deepa and B. Prabadevi

People
ERP Systems Enterprise IoT
(Internet Cloud)

Value Chain

Business
Applications
IoT devices and
sensors
Enterprise Assests

Fig. 5.1 Components of enterprise IoT

processes without or with less intervention of humans. This, in turn, will reduce
manual operations thereby enhancing the efficiency of the business.
Enterprise resource planning (ERP) software is already in place for automizing
all the business activities both front office and back end. Many ERP software such
as Oracle, SAP, PeopleSoft, Fedena, Odoo, and CloudERP are providing better
enterprise solutions to respective different business domains such as education,
finance, manufacturing, and so on. So this will also be a part of an E-ToT environ-
ment. The emergence of E-IoT applications in the enterprise will automize the busi-
ness processes. The various components of E-IoT are depicted in Fig. 5.1.
Though enterprise-specific business processes exist in each business, in general,
most of the business processes respond based on the signals or commands from
robust machines and other types of equipment. Also, these business processes can
react based on the predefined business rules embodied in the E-IoT application.
Fig. 5.2 depicts how the business processes interact with other entities and how the
solution is retrieved by them. To deploy IoT applications on a large scale, the enter-
prises must have sufficient capital, good reach, enormous resources, and appropriate
reasons for deploying it. Subsequently, it will be easier for businesses to attain
greater business value, as well to utilize it for further adoption and business invest-
ment from the return of investment (ROI).
The greatest challenge for E-IoT is how the heterogenous business data will be
handled. So, E-IoT bridges with technologies like mining to extract useful informa-
tion from data stored and processed in the cloud and predicts different trends and
patterns from knowledge gained through mining using machine learning and big
data analytics. Also, real-time stream analytics have a more significant role to play
with live data involved in business processes. Real-time streaming helps the enter-
prise decision makers to visualize data involved in every transaction and other busi-
ness activities before taking an operational decision [3]. These technologies help in
5 Advanced Machine Learning for Enterprise IoT Modeling 101

Fig. 5.2 Business process interaction with other entities in an enterprise

predictive maintenance of equipment, predicting market trends, determining pro-


duction styles through customer buying pattern prediction, predicting the stock
market, forecasting market demand to assist in demand-based manufacturing, pre-
dicting marketing trends, deciding prices for stock depending on different market-
ing stages of a product (i.e., from product promotion to stabilized brand product),
and so on.
Integrating IoT devices with enterprise processes requires tiring work. This
includes custom activity in various functionalities like hardware configuration, its
deployment, and middleware configuration. So it is better to understand the existing
business before proceeding for integration. Business process modeling (BPM) is the
most popular technique in developing business models and accomplishing complex
business processes in enterprises. Though several tools on BPM exist, those tools do
not resolve the challenges in IoT-based business processes. Some of the challenges
are adaptive event-driven business processes, distributed processes, and business
processes that deal with unreliable data and resources [4].

5.2 Enterprise Forecasting

Business forecast refers to predicting the probable outcome in almost all the busi-
ness processes or future situations of an enterprise and acting accordingly to attain
success. Henceforth a better business forecasting facility of an enterprise deter-
mines its success in the short term. Most of the activities in the business are uncer-
tain. So to be on the safer side and to react better in uncertain situations, forecasting
102 N. Deepa and B. Prabadevi

Enterprise Forecast

Enterprise Forecast
Types

Based on term Based on Position Based on nature Based on nature


of the people of forecast of output

Long Micro Point


term continuum Quantitative forecast

Short Macro Interval


term continuum Qualitative forecast

Density
forecast

Fig. 5.3 Types of enterprise forecasts

is a must in all the activities we do.There are various models for enterprise forecast-
ing [4] as depicted in Fig. 5.3. The models are classified based on term: long-term
forecasts are those that determine the activities in the long run, whereas short-term
forecasts determine the immediate future, i.e., short period. Therefore top manage-
ment will be focusing on long-term forecast while intermediary level or basic level
of management in an enterprise will be focusing on short-term forecast to decide
their short-term plans.
The forecasts are classified based on the position or level of people in an enter-
prise hierarchy. At each level, management will focus on different stages of the
micro- and macro-continuum. For instance, the production manager will be inter-
ested in forecasting the number of assembly parts, the number of employees, and
the number of plants to be operated to complete the current tender while the top
management will be bothered about the effective utilization of plants over a while
(for a year). The forecasting can be classified in the way it is determined as qualita-
tive or quantitative. The qualitative forecast is based on expert opinions or input
from the enterprise’s customers. Most enterprise practitioners will prefer this when
having very meager or no information about past or historical data or when the
market is facing a new scenario. In such situations, experts will help to predict the
future based on prior experiences. Some of the examples of this approach are mar-
ket research where predictions are done based on a survey with a certain group of
people and Delphi approach where the experts are forced to give forecasts. The
quantitative forecasting is just opposite to the former one discussed. It is merely
based on accurate historical data and omits the human involvement in the analysis.
5 Advanced Machine Learning for Enterprise IoT Modeling 103

Thus predictions are done based on the previous data and not considering the expert
opinions (i.e., manipulation based on human judgments). One of the commonly
used quantitative approaches is the indicator approach which considers the relation-
ship among certain indicators (the positive indicators) and determines the lagging
one’s performance by leading indicator’s previous performance results. The other
methods are econometric and time series. The econometric model is purely based
on mathematical or statistical approaches like regression techniques such as simple
linear regression, multiple regression, and regression with time-series data. Instead
of relying on the static data members relationship, it will calculate the consistency
of the variables in dataset over time and then determine the relationship strength.
The time-series models generally refer to the set of models that predict based on
past data. It differs from others by considering more weight for recent data and
omitting the outliers. The enterprises over the other models best prefer this approach.
The forecast can be classified based on the output as a point, an interval, or a density
forecast. If the forecast output is the single best value, then it is a point forecast,
whereas if a forecast output falls between a range of values, then it is an interval
forecast, or if the output is probably distributed for future value, then it is a density
forecast. Choosing one best forecast method among many available methods is
determined based on various factors such as type of product (new vs. old), goals
(short term or long term), and constraints (cost and time factors). However, it is
infrequent that one single model fits for all the enterprise forecasts. Always a hybrid
combination of forecasts will give the best results. In general, forecasting is done in
five steps, namely, problem analysis and dataset collection, data preprocessing,
building and evaluation of the model, implementation of the model, and evaluation
of the results. Specifically, the performance of the forecast depends on the type of
data and the source from which it was collected. Predicting the future can be better
assisted with machine learning algorithms.

5.3 Machine Learning Algorithms and Their Application

Machine learning is a technology in which the computer will have the capability to
learn from the available dataset and implement the learning to predict unique pat-
terns from the data. It helps in developing models for data analytics. It identifies
hidden knowledge or interesting patterns from the available dataset and predicts the
future in most of the problems. The two popularly used methods in machine learn-
ing are supervised learning and unsupervised learning.
The supervised machine learning algorithms are allowed to perform learning on
the dataset in which class labels or outcomes are known. These kinds of algorithms
are used in credit card companies where the past fraudulent dataset is used for train-
ing the model and the learning helps to predict the fraud in the future. Some of the
popularly used supervised algorithms are support vector machine (SVM), decision
tree classifier, naive Bayes algorithm, linear regression method, logistic regression
method, K-nearest neighbor algorithm, etc. The unsupervised machine learning
104 N. Deepa and B. Prabadevi

Fig. 5.4 The steps in the


development of machine
learning models

algorithms learn from the dataset in which class labels or outcomes are not known.
The main aim of these kinds of unsupervised algorithms is to analyze the data and
produce interesting hidden patterns in the data. For example, these algorithms can
be used to group customers with similar purchasing behavior in retail marketing.
Some of the unsupervised learning algorithms are hierarchical clustering algorithm,
K-means clustering algorithm, self-organizing maps, etc.
The various steps in the development of machine learning models are shown in
Fig. 5.4. The data relevant to the problem should be gathered either from real life or
from past historical data. As the raw data collected may contain missing information
or may be erroneous and inconsistent, statistical methods can be applied to clean the
data. The raw data collected may be in a different unit of measurement, and it should
be normalized to the same unit of measurement which helps to improve the accu-
racy of the developed model. If the dataset consists of high-dimensional data, fea-
ture selection/dimension reduction can be applied to reduce the dimension of the
data or to identify the useful set of features. Then the dataset needs to be divided for
training (learning) and testing purpose. The model is developed by applying appro-
priate classification or clustering algorithms. From the results, knowledge
5 Advanced Machine Learning for Enterprise IoT Modeling 105

prediction is obtained to make reliable decisions for the identified problem. The
results can be visualized with the help of graphs for further analysis and comparison
with the state-of-the-art techniques.
As most of the enterprises are working with a large amount of data nowadays,
machine learning algorithms have become popular in the business domain. They
develop models using machine learning algorithms to identify the interesting pat-
terns in their huge dataset. They make rapid decisions based on the knowledge
acquired through machine learning models on their dataset which helps them to
obtain maximum profit. Machine learning algorithms are applied to the dataset
acquired from IoT sensor devices to predict the knowledge in various domains.
As there is a rapid increase in the application of IoT devices, the amount of data
generated by them also increases. In order to process and analyze the data collected
from IoT devices, models should be developed which can process the huge amount
of data. Machine learning algorithms have proved to provide better results in pro-
cessing data obtained from IoT sensors. Machine learning algorithms have globally
become increasingly popular, and they have been providing various benefits in our
day-to-day life. The various application domains of machine learning algorithms
are shown in Fig. 5.5.

Fig. 5.5 Application domains of machine learning algorithms


106 N. Deepa and B. Prabadevi

Machine learning algorithms have been applied to develop classification and pre-
diction models for the development of agriculture such as land suitability analysis
and crop classification based on multiple parameters [5–11]. A new disease risk
prediction algorithm based on convolutional neural network was developed to pre-
dict the chronic diseases in patients. Experimentation was done using the hospital
data collected from China. Missing data in the dataset was handled by latent factor
model. The results of developed algorithm were compared with existing methods
and showed 94.8% prediction accuracy [12]. A framework was developed to detect
the fake and abusive records in the claims received from the customers in healthcare
insurance agencies. In the framework, pairwise comparison using analytical hierar-
chy process method was applied for the calculation of weights of criteria involved.
Expectation maximization was used for clustering similar records. The developed
framework was validated using a real dataset with six different fraudulent
records [13].
Supervised learning algorithms were used to identify the optimal land location
for starting a new retail store. The real dataset collected from New York was used
for studying the prediction accuracies of various machine learning methods [14]. A
non-linear non-parametric forecast model was built using machine learning algo-
rithms such as linear regression to predict the credit risk of the consumers of a com-
mercial bank [15]. A model was developed to classify and predict the claim amount
from the automobile insurance agency.
Machine learning algorithms such as the hierarchical clustering method, heuris-
tic method, and regression algorithm were used to build the model. Clustering algo-
rithms were applied to group the policyholders with similar factors, and the
regression method was used for classification and prediction of claim amount [16].
A machine learning model was developed using an enhanced AdaBoost algorithm
for classification and prediction of sales data in the retail business. The model pro-
cesses the data obtained from daily transactions and utilizes it to predict future sales
for inventory management. Accordingly, proper decisions can be made to improve
the sales of the retail business [17]. A study was conducted by the Korea
Transportation Safety Authority to find the most suitable parameters that affect the
safety measures of the agencies, and based on that, a strategy for promotion can be
suggested. Cluster-based negative binomial regression method was used for the
selection of factors such as socioeconomic factors, demographic factors, conditions
of roadways, driver behavior, and traffic rule violations. Negative binomial regres-
sion models were built to identify the unique parameters for the prediction of deadly
crashes [18].
A machine learning model was built for the prediction of rainfall under the
hydrological research with non-linear pattern of data which is very difficult to pro-
cess. Machine learning algorithms such as recurrent artificial neural network algo-
rithm, support vector machine method, and particle swarm optimization approach
were used to develop the prediction model. Recurrent artificial neural network algo-
rithm was used to predict the rainfall, and support vector machine is used to find the
solution for the non-linear regression and the problems in time-series data. Particle
5 Advanced Machine Learning for Enterprise IoT Modeling 107

Table 5.1 Applications of machine learning algorithms in different domains


S.
no. Purpose Technologies/methods used
1 To assist farmers to take decision on Machine learning algorithms [5–8]
agriculture crop cultivated in their land for
agriculture development
2 Multi-class classification model for Multi-layer perceptron, IoT sensors [11]
agriculture land suitability analysis
3 Prediction algorithm to forecast the chronic A convolutional neural network, latent factor
diseases in patients model [12]
4 A framework for the detection of fake and Analytical hierarchy process [13]
abusive insurance records for the claims
submitted by the customers
5 A model to find the optimal site location Machine learning algorithms [14]
for starting a new retail shop
6 Non-linear non-parameteric forecast model Linear regression method [15]
to predict the credit risk of bank customers
7 A decision model to classify and predict Hierarchical clustering method, heuristic
the insurance claim amount for automobile method, and regression algorithms [16]
insurance agency
8 A model to classify and predict the sales AdaBoost algorithm [17]
data for a retail business
9 A model to find suitable attributes for Cluster-based negative binomial regression
safety measures in transportation agencies method [18]
10 A machine learning model to predict the Recurrent artificial neural network algorithm,
rainfall with non-linear pattern of data support vector machine, and particle swarm
optimization approach [19]

swarm optimization method is used to select the relevant parameters for the devel-
opment of the model [18]. The survey of applications of various machine learning
methods in different problems is summarized in Table 5.1.

5.4 Applications of Machine Learning in E-IoT

From the past few years, the costs of IoT sensors and chips used to store the data
have been diminishing. The data collected by IoT sensors are very huge in the cloud
and nowadays it is very fast also. In order to analyze the data stored in IoT, machine
learning plays a major role. This can be possible by integrating the physical devices,
cloud, and machine learning models which in turn will increase the performance
and efficiency of decisions taken in the business enterprise [20]. Enterprise IoT is
becoming popular due to the technological advancement where the physical sensor
devices are integrated with software modules to take rapid and accurate decisions in
business applications. Enterprise IoT reduces manpower and helps to improve the
overall productivity of businesses. The physical devices are interconnected with the
108 N. Deepa and B. Prabadevi

Fig. 5.6 Applications of machine learning models in E-IoT applications

Internet thus transforming the required data to the cloud for data analysis which
helps to improve the enterprise.The architecture diagram depicting the application
of machine learning models in enterprise IoT is shown in Fig. 5.6.
IoT devices have been recently used in the manufacturing divisions of industries
where the data collected from various sensors are used for predicting some useful
information. A predictive model was developed using autoregressive integrated
moving average method. The data collected from various sensor devices fixed in a
slitting machine were applied to the developed predictive model in order to forecast
the time-series data. Thus machine learning has proved its importance in enterprise
IoT [21]. A machine learning model was built using multistage meta-classifier for
network traffic analysis. This model classifies and identifies the IoT devices which
are connected to a specific network. The model initially distinguishes between the
traffic data generated by the IoT devices and non-IoT devices. Later it classifies and
identifies the correct IoT device from the chosen nine IoT devices. The model has
proved to provide an accuracy of 99.28% [22].
A multi-class model was built using a supervised machine learning algorithm,
namely, random forest, for the identification of fake IoT devices connected to the
organization network. The model extracts the traffic data from the network to iden-
tify the type of IoT device from the given list of devices. In order to train the clas-
sification model, the traffic data collected from 17 IoT devices have been used with
9 class labels. The multi-class model has given 99.4% accuracy in detecting fake
IoT devices [23]. The water crisis is a major problem nowadays for sustainable
agriculture development. There is a need for the effective utilization and
5 Advanced Machine Learning for Enterprise IoT Modeling 109

management of water. A machine learning-based model was developed using mul-


tiple linear regression algorithm for monitoring the water level in the agriculture
field. IoT devices were fixed in the irrigation field which will continuously collect
details about the underground water level and other details such as sunlight, air pres-
sure, rainfall, temperature, etc.; this will help the farmers to get details about the
water scarcity which will hit the land in the future [24].
A model was developed using hidden Markov method to predict the disease in
grapefruits and give an alert to the farmers to apply pesticides through SMS to their
mobile phones. In order to develop the model, various sensors to collect details such
as moisture, humidity, temperature, and leaf wetness were fixed for data collection
[25]. A model was built using a machine learning algorithm to alert about the health
condition of a person. The heartbeat of the person was measured using IoT sensor
and taken as a parameter to predict the stress behavior of the person in this model
[26]. A business intelligent system was proposed to track the emotion and behavior
of the customers using IoT devices which runs on Apache Spark cluster in the retail
industry. An intelligent trolley, namely, EmoMetric, was developed to provide
insights on customer behavior by using the customer’s face. The results were com-
pared with other popular techniques and showed 95% accuracy [27].
A classification model was developed using a supervised neural network algo-
rithm, namely, multi-layer perceptron, to classify the threats in IoT sensor network.
The results of the developed model produced 99% accuracy and proved to give bet-
ter solution in the detection of distributed denial-of-service attacks [28]. A frame-
work was presented for the development of the decision support system to work
inside the ecosystem of IoT. Network communication data related to quality for
electric smart meter is analyzed to make decisions on whether a technician needs to
be sent to the customer address to solve the issue in the electric smart meter. The
framework was implemented using a Bayesian network, and the results were com-
pared with the results obtained from naive Bayes, decision tree, and random forest
algorithms [29].
An artificial intelligence-based bot, namely, SamBot, was developed for a mar-
keting purpose which is to provide interactive answers to customers’ questions in
the corporate website of Samsung IoT. SamBot is integrated with the marketing
domain knowledge such as frequently asked questions, product promotions, etc.,
Sometimes the bot gives random answers if the knowledge is not available. In order
to increase the efficiency of the bot, a supervised machine learning algorithm was
applied to improve the knowledge of the developed SamBot [30]. A big data-based
cloud system architecture was proposed to provide recommendations on a list of
products to the customers. The architecture includes cloud to store the data relevant
to products and purchases ranking. It analyzes the big data from the private and
hybrid cloud for ranking the products to improve the retention in marketing [31].
An integrated system model was developed to monitor the environment and cli-
matic changes. The system integrates cloud platform, Internet of Things, remote
sensing, global positioning system, geographic information system, and artificial
intelligence. Web services, various sensors, and private and public networks have
been used in this model to collect the data from various sources and transport it to
110 N. Deepa and B. Prabadevi

the destination network. IoT-based data collection is done using multiple sensor
devices, satellites, radar, balloons, meteorological instruments, mobile devices,
Bluetooth, ZigBee, Wi-Fi, and RFID. Decision making and planning is scheduled in
the application layer of the developed system architecture. The main functionalities
of this layer are the storage, organization, processing, and distributing of the data
related to climate and environment which are acquired from various sensor devices.
And it also monitors ecological factors, pollution, resource management, and
weather and forecasts disaster. It accomplishes the task of decision making related
to the environment and weather conditions. A case study was conducted by collect-
ing data from China to prove the efficiency of the system developed [32].
Internet of Things-based production system for processing electrical and elec-
tronic equipment waste was developed for a particular type of manufacturing unit
for recycling purpose. In this system, a cloud-based architecture was implemented
to provide services for the production department. The cloud platform is deployed
to deliver computing units, software products, and storage space to the destination
through the network. In the manufacturing division of an industry, it is not possible
to provide resources such as tools, materials, and machines through the network.
Hence Internet of Things is essential to connect and integrate the physical devices
to the cloud platform and named as the Internet of Manufacturing Things (IoMT).
This IoMT is used to exchange the resources and data but in various formats.
Therefore there is a necessity to build an integrated method to assist the industrial
information system among various service shareholders. In the cloud platform,
electrical and electronic equipment waste can be combined with industrial informat-
ics through IoMT [33].
A study is conducted on how to provide service through the development of the
Internet to the manufacturing division of an organization by deploying Internet of
Things platform. The study also analyzes how various benefits of the Internet of
Things can be applied to improve product-service systems. Three case studies have
been taken in which the Internet of Things was successfully implemented in three
manufacturing companies in different sectors of industry such as power distribution
and generation and metal processing units. The empirical results provided an idea to
create different values of providing services to the industrial sectors in the manufac-
turing division [34].
The various enterprise applications of machine learning and the Internet of
Things are summarized in Table 5.2.

5.5 Issues in Enterprise Internet of Things

There are several issues in integrating Internet of Things with enterprise. The major
challenges faced by enterprise IoT are shown in Fig. 5.7 [35].
5 Advanced Machine Learning for Enterprise IoT Modeling 111

Table 5.2 Application of machine learning algorithms on enterprise Internet of Things


S.
no. Purpose Technologies/methods used
1 The predictive model to forecast IoT, autoregressive integrated moving average
time-series data method [22]
2 Machine learning model for network Multistage meta-classifier, IoT [23]
traffic analysis and to distinguish between
IoT and non-IoT devices
3 Multi-class model to identify fake IoT Random forest algorithm, IoT sensors
devices in organization network
4 A decision model to monitor the Multiple linear regression algorithm, IoT
groundwater level in sustainable sensor devices [24]
agriculture development
5 A model to predict the disease in IoT sensor devices, machine learning
grapefruits and recommend suitable algorithm, namely, hidden Markov method
pesticides to the farmers [25]
6 An approach to predict the stress behavior Machine learning algorithms, sensor devices
of a person [26]
7 Business intelligent system to predict the IoT devices, Apache Spark cluster, machine
emotion and behavior of the customers learning methods [27]
8 A classification model to classify the Multi-layer perceptron, IoT sensors [28]
threats in IoT sensor networks in
distributed denial-of-service attacks
9 A decision support system to provide Internet of Things, Bayesian classifier,
technical support for smart electric meter decision tree, random forest [29]
10 Artificial intelligence-based SamBot to The supervised machine learning algorithm,
provide online help for Samsung IoT devices [30]
customers
11 A recommendation model using cloud Cloud platform, IoT, machine learning
system architecture to give product algorithms [31]
suggestions for the customers
12 An integrated model to monitor the IoT, multi-sensor devices, cloud platform,
environment and climate change artificial intelligence, remote sensing, global
positioning system, geographic information
system, web services [32]
13 IoT-based production system to process Cloud platform, IoT, machine learning
electrical and electronic equipment waste algorithms, Internet of Manufacturing Things
in manufacturing division for recycling [33]
purpose
14 A system to provide services to the Internet of Things, cloud platform, machine
manufacturing division of the learning algorithms [34]
organization with the implementation of
the Internet of Things
112 N. Deepa and B. Prabadevi

Fig. 5.7 Major challenges in integrating Internet of Things with enterprise

5.5.1 Data Security Issues

It has been found that most of the Internet attacks are targeted toward IoT devices.
Topmost enterprises also face these security issues. Both users and enterprise have
to face the problem of issues in privacy and security. Hence proper measure needs
to be taken to enhance the privacy and security solutions in enterprise Internet
of things.

5.5.2 Issues in the Network Connection

The major challenge in Internet connectivity is it is not possible to utilize the Internet
facility at the same speed in all the places. It has been revealed that one of the major
challenges in IoT implementation is the Internet connectivity issue. When IoT
devices are fixed in remote locations, it is not possible to transmit the data with high
speed. There may be disruption in Internet connection. Furthermore, based on the
router, the signal quality collected by various sensors which transmit data to the
network differs. The networks should be connected with the latest technologies to
enhance the speed and quality of data transmission in the IoT platform. Moreover,
5 Advanced Machine Learning for Enterprise IoT Modeling 113

the number of IoT devices connected to the network should also be monitored to
obtain good network coverage.

5.5.3 IoT Platform Compatibility with Heterogeneous Network

IoT sensors and network are the major components in implementing IoT. Not every
computer connected to a network is capable of connecting to advanced sensors. And
also not every network has the capability to communicate and share the data effec-
tively. Different sensors have the distinct power-consuming capability and may not
be able to produce the same results. It is mandatory to deploy the latest technologies
to integrate the different sensors to the heterogeneous network which will commu-
nicate data in the same manner.

5.5.4 I ntegration of IoT Devices with Appropriate


IoT Platform

In order to incorporate the IoT devices with enterprises, the IoT physical devices
should be integrated to correct IoT platforms. If it is not properly connected to
appropriate IoT platform, functional abnormalities and inefficiency in the delivery
of solutions to the customers may happen. It has been found that nearly half of the
implementation cost should be spent for the integration of IoT devices with appro-
priate back-end systems and IoT platforms. Only then the enterprises can explore
the data through big data tools and predict the knowledge. Therefore appropriate
IoT devices and platform for integration of application-specific data should be
devised before integration. This will help in avoiding the issues.

5.5.5 Issues in Data Collection

The main purpose of integrating IoT devices with the enterprise is to capture the
data from various physical devices, convert to some standard format, apply machine
learning models, and predict meaningful information for business purposes. If there
are anomalies in the real environment where IoT devices are fixed, say for example
power shutdown or problems due to natural calamities, wrong data may be collected
by the IoT sensor devices.
114 N. Deepa and B. Prabadevi

5.5.6 Issues in Intelligent Predictive Analytics

The major contribution of enterprise Internet of Things is that it transforms the data
to supply knowledge. If there are flaws in the collected data, the prediction capabili-
ties will not be correct. In most of the places, IoT stakeholders find it difficult to
collect correct data, and they confirm that the analytics made by them are not flexi-
ble and strong. Therefore the latest analytic software like machine learning models
can be developed to check the integrity and consistency of the data before intelligent
predictive analytics can be made.

5.5.7 I ntegration of Big Data with IoT for Handling


Unstructured Data

As there is an increase in connected devices, the problems in handling unstructured


data also increase. The other major challenge for the enterprises is to decide which
data is useful because only valuable data can be used for data analytics. The data
collected from various physical devices consists of unstructured data. Unstructured
data cannot be stored in SQL databases. When unstructured data is saved in NoSQL
format, it is a little complex to access the data. Big data frameworks such as
Cassandra and Hadoop reduce the complexity of handling unstructured data. But
the integration of IoT with big data is itself a major challenge.

5.5.8 Lack of Awareness in Users

Many customers use mobile apps, but they are aware of the Internet of Things incor-
porated in it. Even though it is not necessary for the consumers to have technical
awareness about IoT devices, the lack of IoT knowledge may lead to hesitation with
respect to cost and security of such devices. And that may slow down the usage of
such technologies. In a survey conducted by Cisco, it has been found that 53% of
customers do not want their data to be collected from their devices. This shows the
lack of awareness about the IoT platform among the consumers.

5.6 Advanced Machine Learning Techniques in E-IoT

As discussed priorly, machine learning assists in various business activities and


allows the decision makers to make a wise decision in different aspects of business
solutions. Based on the historical data, the machines are trained and then test data is
5 Advanced Machine Learning for Enterprise IoT Modeling 115

given to test the system’s behavior without human intervention. In this section, we
provide a brief overview of advanced machine learning techniques for E-IoT.

5.6.1  pplication of Machine Learning for Demand Forecast


A
in Enterprise Supply Chain

One of the major activities in an enterprise is a business forecast. The supply chain
is the chain that connects the supplier from whom raw material is procured, the
manufacturer who produces the product from raw material procured, the dealer, the
wholesaler, the retailer, and finally the end buyer or the user. To stay connected and
to retain good relationship with various roles in this chain, an enterprise must make
wise decisions in all the activities in this chain. A supplier needs to forecast manu-
facturer demands, and in turn, a manufacturer must forecast various other processes
in predicting the customer. Although customer demand is predictable, the manufac-
turer’s demand will fluctuate randomly. So the biggest challenge is predicting man-
ufacturer’s demand in this collaborative forecasting environment. It has now become
mandatory for an enterprise to realize the integrated information across various
stakeholders in its supply chain. This determines the value of an enterprise. Fig. 5.8

Customer
Customer order

Supply Chain

Replenishment
Supplier
Procurement

Manufacturing Disributor

Manufacturer

Fig. 5.8 A supply chain


116 N. Deepa and B. Prabadevi

Fig. 5.9 Bullwhip effect 60


and distorted
demand signal 50
Orders Placed
40

Quantity
30

20
Actual Sales
10
0
Time

a. BullWhip Effect

b. Distorted Demand Signal

shows various stakeholders in a supply chain. A major challenge in forecasting the


demand of a manufacturer is the demand signal distortion through the supply chain.
This distortion may be due to inaccurate information passed into the supply chain
[36]. The accurate result on forecasting can be attained through supply chain col-
laboration, i.e., by involving all the stakeholders of the supply chain. Since it is not
always possible to involve all the participants every time, distorted demand signal
should be observed in case of full collaboration and partial collaboration. The dis-
torted demand signal and bullwhip effect in manufacturing are shown in Fig. 5.9.
Though various forecasting techniques can be used as mentioned in the prior
section, it is important to analyze these techniques to observe which induces the
bullwhip effect.
The bullwhip effect is a distribution channel phenomenon. Bullwhip effect on the
supply chain will be observed when there are changes in customer demand causing
the companies involved in the supply chain to make more orders until the new
evolved demands of the customer are met [37]. In a multi-level supply chain, if we
move from right to left, that determines the flow of information for customers; on
other side, it is the demand flow as shown in Fig. 5.8 (Fig. 5.9).
R. Carbonneau et al. had carried out an analysis of various advanced ML tech-
niques for supply chain demand forecast and compared them with traditional basic
methods that can be used for forecast [38]. The advanced ML techniques considered
are neural network (NN), feed-forward error backpropagation NN, support vector
machine (SVM), and recurrent neural network (RNN). The basic forecast tech-
niques used are naïve forest, moving average method, average method, trend analy-
sis, and multiple linear regression. They observed that machine learning algorithms
provide better accurate forecasts than traditional methods. They suggest that trend
5 Advanced Machine Learning for Enterprise IoT Modeling 117

analysis and naïve forest for demand signal processing show more errors in fore-
casts. Also Bontempi et al. suggest that the machine learning algorithms like artifi-
cial neural networks (ANNs), decision trees, nearest neighbor, and SVM outperform
the traditional models such as Box-Jenkins and linear regression for time-series
forecasting [39].

5.6.2  pplication of Machine Learning for Enhancing


A
Customer Value in E-IoT

It is a well-known fact that “a product with good features and excellent performance
but with no customer is equivalent to a failed product.” So a successful enterprise is
one with valued customers. Therefore, customer value is more important in any
enterprises. Some of the IoT applications focus on enhancing customer value in an
enterprise. Understanding the mandatory need for customer value in an enterprise is
considered to be a requirement for IoT adoption in an enterprise. Based on research,
three categories of IoT applications were identified to enhance the customer value.
They are monitoring and controlling, information sharing and collaboration, and big
data analytics and business analytics [40]. Monitoring and control is dealt with han-
dling data maintenance. It handles data collection of equipment’s operational condi-
tion and its environment. Decision makers can track and control the performance of
the equipment at any time from any place. I. Lee and K. Lee clarify how these three
categories assist in attaining customer value [41]. Based on how the system is per-
forming and how the customer is reacting to various operations of the pieces of
equipment in the IoT environment, enterprises can decide how they can enhance
their customers’ needs. As IoT is connected with more number of devices, an enor-
mous amount of data need to be collected from various actuators and sensors. These
data will be transmitted to data analytic and business intelligence tools to make
them appear simple for decision makers. These data can be used for forecasting
customer demands in the market and their behavior. The last category of informa-
tion sharing and collaboration helps in creating a better forecast. As discussed
regarding collaboration in the supply chain, here information sharing happens
through people to device, device to device, and people to people. So, collaboration
among all the entities participating in the communication must be considered. When
all the people/entities are connected collaboratively via IoT devices, here communi-
cation will be transparent and tasks will be accomplished on time. Bose and
Mahapatra have done a brief survey on various applications utilizing machine learn-
ing algorithms in business data mining [42]. Ampazis had tried to predict the
demand of customers in the basic and multi-level supply chain using machine learn-
ing algorithms [43]. He used trained ANN and SVM for regression. He found less
uncertainty while using these advanced machine learning algorithms on three differ-
ent huge datasets, namely, Netflix (to predict demand of movie rentals during the
holiday period), Rotten Tomatoes (a movie review aggregator), and Flixster
118 N. Deepa and B. Prabadevi

(movies-­based social network). He has also suggested a computational intelligence


approach for supply chain demand prediction [44].

5.6.3  pplication of Machine Learning for Various Other


A
Activities in E-IoT

There is no enterprise without a customer. Similarly, business partners are also


equally important as customers. So Mori et al. have used machine learning algo-
rithms for predicting best business partners using their enterprise profiles and their
transactional dealings with various enterprises [45]. To predict the customer-­
supplier relationship, they have to build a machine learning-based prediction model.
They suggest that SVM performs well in modeling the customer-supplier relation-
ship. So this approach can be employed in E-IoT architecture to find plausible busi-
ness partners based on their employee count, ranking, and foundation date and to
better classify the customer-supplier relationships. Cumby et al. have proposed a
machine learning-based prediction model for determining customer’s shopping list
from PoS purchase data. They have used decision trees (C4.5) for predicting class
labels, and for learning each class, they have applied perceptron, winnow, and naïve
Bayes [46]. They conclude that their tool can be used to build better customer satis-
faction and can increase the revenue to up to 11%. So this can be employed in E-IoT
for enhancing customer satisfaction.
To retain the customer, better customer relationship management is mandatory in
any enterprise. Buckinx et al. proposed a multi-linear regression model to predict
customer loyalty based on the transactional data available [47]. They have com-
pared their model with other machine learning algorithms like random forest and
relevance determination neural network and obtained better prediction rates from
their models. So this model can be employed for E-IoT for predicting their customer
loyalty.

5.7 Summary

This chapter discusses about enterprise IoT and its uses, challenges and issues,
components, and applications in the real world. Various technologies that can be
used for assisting E-IoT to sort various issues and challenges of E-IoT were pre-
sented. Brief details on machine learning and its importance, functionality, applica-
tions, and various algorithms had been carried out. Specifically how E-IoT is
benefitted with the help of machine learning algorithms was addressed. As a con-
cluding note, machine learning algorithms serve better with best accuracy results
than traditional models existing for prediction of demand in supply chain, customer
value, financial transaction, and business partners.
5 Advanced Machine Learning for Enterprise IoT Modeling 119

References

1. Xia, F., Yang, L. T., Wang, L., & Vinel, A. (2012). Internet of things. International Journal of
Communication Systems, 25(9), 1101.
2. Gharami, S., Prabadevi, B., & Bhimnath, A. (2019). Semantic analysis-internet of things,
study of past, present and future of IoT. Electronic Government, an International Journal,
15(2), 144–165.
3. Haller, S., & Magerkurth, C. (2011). The real-time enterprise: Iot-enabled business processes.
In IETF IAB workshop on interconnecting smart objects with the internet, pp. 1–3.
4. Hanke, J. E., Reitsch, A. G., & Wichern, D. W. (2001). Business forecasting (Vol. 9). Prentice
Hall: Upper Saddle River.
5. Haldorai, A., & Kandaswamy, U. (2019). Cooperative spectrum handovers in cognitive radio
networks. In EAI/Springer innovations in communication and computing (pp. 1–18). Cham:
Springer. https://doi.org/10.1007/978-3-030-15416-5_1.
6. Deepa, N., & Ganesan, K. (2019). Decision-making tool for crop selection for agriculture
development. Neural Computing and Applications, 31(4), 1215–1225.
7. Deepa, N., Ganesan, K., & Sethuramasamyraja, B. (2019). Predictive mathematical model for
solving multi-criteria decision-making problems. Neural Computing and Applications, 31(10),
6733–6746.
8. Haldorai, A., Ramu, A., & Murugan, S. (2019). Social relationship ranking on the smart inter-
net. In Computing and communication systems in urban development (pp. 141–159). Cham:
Springer. https://doi.org/10.1007/978-3-030-26013-2_7.
9. Deepa, N., Ganesan, K., Srinivasan, K., & Chang, C. Y. (2019). Realizing sustainable devel-
opment via modified integrated weighting MCDM model for ranking agrarian dataset.
Sustainability, 11(21), 6060.
10. Deepa, N., Srinivasan, K., Chang, C. Y., & Bashir, A. K. (2019). An efficient ensemble
vtopes multi-criteria decision-making model for sustainable sugarcane farms. Sustainability,
11(16), 4288.
11. Vincent, D. R., Deepa, N., Elavarasan, D., Srinivasan, K., Chauhdary, S. H., & Iwendi,
C. (2019). Sensors driven AI-based agriculture recommendation model for assessing land suit-
ability. Sensors, 19(17), 3667.
12. Chen, M., Hao, Y., Hwang, K., Wang, L., & Wang, L. (2017). Disease prediction by machine
learning over big data from healthcare communities. IEEE Access, 5, 8869–8879.
13. Kose, I., Gokturk, M., & Kilic, K. (2015). An interactive machine-learning-based electronic
fraud and abuse detection system in healthcare insurance. Applied Soft Computing, 36,
283–299.
14. Karamshuk, D., Noulas, A., Scellato, S., Nicosia, V., & Mascolo, C. (2013, August). Geo-­
spotting: Mining online location-based services for optimal retail store placement. In
Proceedings of the 19th ACM SIGKDD international conference on knowledge discovery and
data mining, ACM, pp. 793–801.
15. Khandani, A. E., Kim, A. J., & Lo, A. W. (2010). Consumer credit-risk models via machine-­
learning algorithms. Journal of Banking & Finance, 34(11), 2767–2787.
16. Yeo, A. C., Smith, K. A., Willis, R. J., & Brooks, M. (2001). Clustering technique for risk
classification and prediction of claim costs in the automobile insurance industry. Intelligent
Systems in Accounting, Finance & Management, 10(1), 39–50.
17. Von Kirby, P., Gerardo, B. D., & Medina, R. P. (2017). Implementing enhanced AdaBoost
algorithm for sales classification and prediction. International Journal of Trade, Economics
and Finance, 8(6), 270–273.
18. Jung, S., Qin, X., & Oh, C. (2019). Developing targeted safety strategies based on traffic
safety culture indexes identified in stratified fatality prediction models. KSCE Journal of Civil
Engineering, 2019, 1–8.
19. Hong, W. C. (2008). Rainfall forecasting by technological machine learning models. Applied
Mathematics and Computation, 200(1), 41–57.
120 N. Deepa and B. Prabadevi

20. Patel, P., Kaulgud, V., Chandra, P., & Kumar, A. (2015, December). Building enterprise-­
grade internet of things applications. In 2015 Asia-Pacific Software Engineering Conference
(APSEC), IEEE, pp. 4–5.
21. Kanawaday, A., & Sane, A. (2017, November). Machine learning for predictive maintenance
of industrial machines using iot sensor data. In 2017 8th IEEE International Conference on
Software Engineering and Service Science (ICSESS), IEEE, pp. 87–90.
22. Meidan, Y., Bohadana, M., Shabtai, A., Guarnizo, J. D., Ochoa, M., Tippenhauer, N. O., &
Elovici, Y. (2017, April). ProfilIoT: A machine learning approach for IoT device identification
based on network traffic analysis. In Proceedings of the symposium on applied computing,
ACM, pp. 506–509.
23. Meidan, Y., Bohadana, M., Shabtai, A., Ochoa, M., Tippenhauer, N. O., Guarnizo, J. D., &
Elovici, Y. (2017). Detection of unauthorized iot devices using machine learning techniques.
arXiv preprint arXiv:1709.04647.
24. Ahmed, F. (2017, October). An IoT-big data based machine learning technique for forecasting
water requirement in irrigation field. In International conference on research and practical
issues of enterprise information systems, Springer, pp. 67–77.
25. Patil, S. S., & Thorat, S. A. (2016, August). Early detection of grapes diseases using machine
learning and IoT. In 2016 second international conference on Cognitive Computing and
Information Processing (CCIP), IEEE, pp. 1–5.
26. Pandey, P. S. (2017, July). Machine learning and IoT for prediction and detection of stress. In
2017 17th International Conference on Computational Science and Its Applications (ICCSA),
IEEE, pp. 1–5.
27. Tallapragada, V. S., Rao, N. A., & Kanapala, S. (2017). EMOMETRIC: An IOT integrated big
data analytic system for real time retail customer’s emotion tracking and analysis. International
Journal of Computational Intelligence Research, 13(5), 673–669.
28. Haldorai, A., & Kandaswamy, U. (2018). Cooperative spectrum handovers in cognitive radio
networks. In EAI/Springer innovations in communication and computing (pp. 47–63). Cham:
Springer. https://doi.org/10.1007/978-3-319-91002-4_3.
29. Siryani, J., Tanju, B., & Eveleigh, T. J. (2017). A machine learning decision-support system
improves the internet of things’ smart meter operations. IEEE Internet of Things Journal, 4(4),
1056–1066.
30. Pradana, A. D. I. T. Y. A., Goh, O. S., & Kumar, Y. J. (2018). Intelligent conversational bot for
interactive marketing. Journal of Telecommunication, Electronic and Computer Engineering
(JTEC), 10(1–7), 1–4.
31. Galletta, A., Carnevale, L., Celesti, A., Fazio, M., & Villari, M. (2017). A cloud-based system
for improving retention marketing loyalty programs in industry 4.0: A study on big data stor-
age implications. IEEE Access, 6, 5485–5492.
32. Fang, S., Da Xu, L., Zhu, Y., Ahati, J., Pei, H., Yan, J., & Liu, Z. (2014). An integrated system
for regional environmental monitoring and management based on internet of things. IEEE
Transactions on Industrial Informatics, 10(2), 1596–1605.
33. Wang, X. V., & Wang, L. (2017). A cloud-based production system for information and ser-
vice integration: An internet of things case study on waste electronics. Enterprise Information
Systems, 11(7), 952–968.
34. Rymaszewska, A., Helo, P., & Gunasekaran, A. (2017). IoT powered servitization of man-
ufacturing – An exploratory case study. International Journal of Production Economics,
192, 92–105.
35. https://www.finoit.com/blog/enterprise-challenges-in-iot/
36. Heikkila¨, J. (2002). From supply to demand chain management: Efficiency and customer
satisfaction. Journal of Operations Management, 20(6), 747–767.
37. Dejonckheere, J., Disney, S. M., Lambrecht, M. R., & Towill, D. R. (2003). Measuring and
avoiding the bullwhip effect: A control theoretic approach. European Journal of Operational
Research, 147(3), 567–590.
5 Advanced Machine Learning for Enterprise IoT Modeling 121

38. Carbonneau, R., Laframboise, K., & Vahidov, R. (2008). Application of machine learning
techniques for supply chain demand forecasting. European Journal of Operational Research,
184(3), 1140–1154. https://doi.org/10.1016/j.ejor.2006.12.004.
39. Bontempi, G., Ben, T. S., & Le Borgne, Y. A. (2013). Machine learning strategies for time
series forecasting. In M. A. Aufaure & E. Zimányi (Eds.), Business intelligence. eBISS 2012
(Lecture notes in business information processing) (Vol. 138). Berlin/Heidelberg: Springer.
40. Chui, M., Lo¨ffler, M., & Roberts, R. (2010). The internet of things. McKinsey &
Company. Retrieved from http://www.mckinsey.com/insights/high_tech_telecoms_internet/
the_internet_of_things
41. Lee, I., & Lee, K. (2015). The internet of things (IoT): Applications, investments, and
challenges for enterprises. Business Horizons, 58(4), 431–440. https://doi.org/10.1016/j.
bushor.2015.03.008.
42. Bose, I., & Mahapatra, R. K. (2001). Business data mining – A machine learning perspective.
Information & Management, 39(3), 211–225. https://doi.org/10.1016/s0378-7206(01)00091-x.
43. Ampazis, N. (2015). Forecasting demand in supply chain using machine learning algorithms.
International Journal of Artificial Life Research (IJALR), 5(1), 56–73.
44. Ampazis, N. (2012). A computational intelligence approach to supply chain demand forecast-
ing. In Machine learning: Concepts, methodologies, tools and applications (pp. 1551–1565).
Hershey: IGI Global.
45. Mori, J., Kajikawa, Y., Kashima, H., & Sakata, I. (2012). Machine learning approach for find-
ing business partners and building reciprocal relationships. Expert Systems with Applications,
39(12), 10402–10407. https://doi.org/10.1016/j.eswa.2012.01.202.
46. Cumby, C., Fano, A., Ghani, R., & Krema, M. (2004). Predicting customer shopping
lists from point-of-sale purchase data. In Proceedings of the 2004 ACM SIGKDD inter-
national conference on Knowledge Discovery and Data Mining – KDD’04. https://doi.
org/10.1145/1014052.1014098.
47. Buckinx, W., & Van den Poel, D. (2005). Customer base analysis: partial defection of behav-
iourally loyal clients in a non-contractual FMCG retail setting. European journal of operational
research, 164(1), 252–268.
Chapter 6
Enterprise Architecture for IoT:
Challenges and Business Trends

A. Haldorai , A. Ramu , and M. Suriya

6.1 Introduction

The industrial architecture is categorized as a discipline used for holistically and


proactively initiating industrial responses to disruptive forces. As such, this form of
architecture is vital for analyzing and identifying the relevant execution transition to
obtain the desired outcomes and visions. This paradigm determines the manner in
which enterprises can attain its present and future industrial missions and visions.
Industrial architecture includes the management principles, which are behind the
present discussions concerning the enterprises strategies and the manner it is pre-
sented via the IoTs. The enterprise architecture (EA) was established as an approach
to enhance the specification of a logical blueprint, which explains the operation and
structure of an enterprise. The purpose of the EA is to determine the manner in
which enterprises can effectively attain their present and future missions and visions.
EA aligns technology architecture capabilities with business architecture processes
and services [1]. Historically, this alignment embraced monolithic structures,
including operational systems and data warehouses supported by relational data-
bases and flat files. With today’s Internet of Things (IoT) and all the connected
devices it involves, solution design offers new opportunities, even though designs
can be more challenging. The emerging solution environment has the following
characteristics:

A. Haldorai (*)
Sri Eshwar College of Engineering, Coimbatore, Tamil Nadu, India
A. Ramu
Presidency University, Bangalore, India
M. Suriya
KPR Institute of Engineering and Technology, Coimbatore, Tamil Nadu, India

© Springer Nature Switzerland AG 2020 123


A. Haldorai et al. (eds.), Business Intelligence for Enterprise Internet of Things,
EAI/Springer Innovations in Communication and Computing,
https://doi.org/10.1007/978-3-030-44407-5_6
124 A. Haldorai et al.

• New services and processes for business commerce, lifestyle, social interaction,
culture diversity, and environment management
• A broad range of user platforms such as wearables, smartphones, and a multitude
of appliances
• An expanding use of sensors that replicate human sight, hearing, smell, taste, and
touch and implement real-time concepts.
• Better timing dependency among and between services and processes using large
data volumes and new data analytics to drive the process of decision-making.
This executive update explores the impact of the IoT on traditional business and
technology architectures and the role of EA as an effective methodology for devel-
oping and implementing IoT strategies. We examine business architecture and how
it integrates IoT-driven processes with traditional processes. The IoT is character-
ized by “things” – many of them small in physical size, which can connect to other
things, generating a large network of collaborative IoT and non-IoT things. In this
paradigm, business process management (BPM) extends to accommodate innova-
tive workflows that are more functionally robust with “thing-to-thing” linkages. The
architects behind the EA are the experts who manage every structure to ensure that
IT frameworks are designed based on the present enterprise standards and strate-
gies. The IoTs has proposed a new prototype whereby the global networks of
devices and machines are capable of sharing to establish a digitalized innovation in
industries. As such, the sector of the IoTs has significantly developed to become the
largest sector.
IoTs is large when both viewed as an opportunity for enterprise, including the
kind of data it creates. Approximately 90% of data in the whole world has been
formed over the past 2 years. Presently, there is an output data of about 2.5 quintil-
lion bytes produced daily. Due to the movement of enterprises from the initial phase
of experimentation to complete deployment of the IoTs application, the deluge of
data will progressively be termed as plague to the organizations since they focus on
processing, capturing, and acting on the big data.

6.2 IoT Trends and Challenges

For the purpose of determining the dependability and reliability of the IoTs, a num-
ber of minimal segments of the measures have to be accomplished to attain an
interoperability and integration in the sector of IoTs.
Contract decoupling: The IoTs framework includes the heterogeneous network-
ing devices with the disparate communication paradigms [1]. The integration frame-
work has to be competent to effectively deal with contract decoupling.
Scalability: Provided the developing condition of the IoTs and calculation and
prediction by [2, 3], an effective incorporation system has to be evolvable and scal-
able to efficiently support a lot of things linked to the network.
6 Enterprise Architecture for IoT: Challenges and Business Trends 125

Ease of testing: The integration segment must support the ease of debugging and
testing. Moreover, it must promise the support for debugging failures and defects,
incorporation testing, system testing, component testing, installation testing, com-
patibility testing, non-functional testing, security testing, and performance testing.
Ease of development: The IoTs integration system has to promise a form of ease
of development for Internet developers. The system should potentially eliminate all
the unnecessary complexities to assure an efficient documentation form for both the
developers and the non-developers who have to apply the basic programming skills.
This enables users to easily understand the inner network systems.
Fault tolerance: The IoTs framework should be resilient and dependable. The
smart integration system has to effectively deal with any faults as the IoTs devices
can possibly toggle over the online and offline states.
Lightweight implementation: The incorporation frameworks must be lightweight
overheads in the deployment and development stages. Moreover, it must be easier to
install, activate, uninstall, update, deactivate, and adapt and for versioning [4].
Internal domain operability: The system must be extensible to effectively support
the inner domain communication. For instance, in the intelligent vehicle domain,
the integration system must promise the support of interaction and communication
with the necessary road closures and traffic lights that belong to the intelligent town
domains.

6.3 Enterprise IoT – Overview

The enterprise IoTs is the upcoming sector in technology since it will enhance the
development of the physical things comprising embedded computer devices to
effectively be applied in various business processes. As such, this enables the reduc-
tion of manual work while advancing the general organizational efficacy [5]. With
the application of the various technological advancements that range from the
embedded networking devices with actuators and sensors to the network-based
cloud platforms and communication, the enterprise IoTs applications can poten-
tially automate the various enterprise processes. These procedures depend on the
contextual data that is produced from the programming devices like vehicles, equip-
ment, and machines. The enterprise IoTs is purposed to be the next advancement in
technology, which included the physical things comprising embedded computing
networking devices which are used in organizational processing [6]. This process is
applicable in reducing manual tasks hence enhancing the overall enterprise efficacy.
Based on the combination of a number of technologies that range from the embed-
ded devices with actuators and sensors to the Internet-based communication and the
cloud platforms, the industrial IoTs application can effectively automate the organi-
zational processes, which are dependent on contextual data produced by the pro-
gramming devices like vehicles and machines. Moreover, these enterprise IoTs
applications can transfer control instructions to the devices dependent on the rele-
vant enterprise guidelines. The IoTs is a crucial novel networking concept that
126 A. Haldorai et al.

permits the previously unlinked physical devices to link up with the Internet. In the
future, the IoTs will become the Internet itself due to the possibility of availing a lot
of trials and opportunities to users [7].
The IoTs is projected to develop the efficiency of enterprises [3] through facilita-
tive novel business models to align various physical processes with digitalized
assets on actual-time basis. The big data and cloud data technologies are vital since
they support the IoTs in ensuring smart insights and scalability. The main objective
of the technology is to formulate an insight that allows a faster, appropriate, and
accurate decision-making process in a more client-centric enterprise.
Some of the basic features of every IoTs stack include:
• Loosely coupled: Three critical IoTs stacks have been recognized. However, it is
fundamental that every stack is applied separately from the other. It has to be
possible to apply an IoTs cloud platform from a single distributor with an IoTs
gateway from another supplier and device stacks from a third distributor.
• Modular: Every stack has to permit for the characteristics to be sourced from
various distributors [8].
• Platform independent: Every stack has to have its own entire cloud and hardware
infrastructure. For instance, the device stacks have to be available on a lot of
MCUs and the IoTs cloud platforms, which have to be operated on various
cloud PaaS.
• Dependent on available standards: Information transfer between different stacks
has to be centered on available standards to accomplish interoperability.
• Defined APIs: Every stack has to contain a defined API, which permits an easy
incorporation with the present integration and application of the relevant IoTs
solutions [9].

6.3.1 Enterprise IoT Platform: Key Attributes

To take the pulse of enterprise IoT, you need an IoT platform whose key attributes
align with the requirements of your enterprise solutions. Your enterprise IoT plat-
form should provide you with the following capabilities:
1. Managing a large number of devices –Your enterprise IoT solution requires a
flexible device management system that can easily be on-board or off-board and
manage a large number of devices in bulk. Plus, it should provide a mechanism
to support a variety of contemporary and emerging protocols to deliver the full
control of devices, regardless of their make, model, or manufacturer. Finally,
your IoT platform should scale to meet the pace and magnitude of growth of
connected devices. This kind of platform can store a warehouse of data and man-
age billions of transactions in real time.
2. Multi-tenancy and fine-grained access control – We should emphasize a major
difference between creating several customers, a.k.a. tenants, on the platform
and creating several users with different permissions within a single tenant.
6 Enterprise Architecture for IoT: Challenges and Business Trends 127

­ enants are physically separated data spaces with own users, separate applica-
T
tion management, and no sharing of data by default. Your enterprise solution
needs a natively multi-tenant platform which would allow you to host countless
customers running differently branded digital applications without the risk of
data leakage. Within a single tenant, you also need fine-grained control over your
users and their access rights. For example, not all employees should have access
to every device or its data, nor should they have the ability to change the settings.
That means such devices should have ways to display what data exactly they are
sharing and who they are sharing it with.
3. Integration with enterprise systems – Enterprise IoT platform should provide
you with an easy way to add new integrations to enterprise systems. By integrat-
ing with existing enterprise applications, you can create a master application that
would let you streamline your work and get the most value from your data.
4. Security and privacy at every level of the stack – Security should be embedded at
every level of the platform. Your platform should provide end-to-end data protec-
tion ensuring that personal data is not accessible to unauthorized viewers. Most
importantly, it should support per-device authentication and authorization to
enforce enhanced security. Also, the platform needs to offer a robust and proven
security model that will keep your system secure and data safe. Finally, it should
be ISO 27001 accredited, which demonstrates that you have taken all the neces-
sary steps to implement internal security practices and protect your business.
5. Data analysis – Data analytics should be an integral part of the enterprise con-
nected solution. Why? Through data analysis, raw data is transformed into mean-
ingful information that will help end users to draw key insights to make their
decisions move forward.

6.3.2 Enterprise IoT Architecture

The IoT technology stack is given in Fig. 6.1 which contains three tiers: gateways,
data center or cloud IoT platform, and sensor devices. According to [10], “a typical
IoT solution is characterized by many devices (i.e., things) that may use some form
of gateway to communicate through a network to an enterprise back-end server that
is running an IoT platform that helps integrate the IoT information into the existing
enterprise.”
The devices tier concentrates on data collections through the sensors, which can
be embedded in various forms of devices. These devices include the mobile comput-
ing devices, autonomous appliances, machines, and wearable technologies. These
devices have the potential to capture the details concerning the physical environ-
ments like light, humidity, chemistry, vibration, and pressure. The standard-­centered
wireless and wired network paradigms are applicable in the transfer of telemetry
information ranging from the cloud via the gateways to the networking devices. The
layers of devices form the foundations of the IoTs stacks. The legacy networking
devices, which have been available for many years with the intelligent, modern, and
128 A. Haldorai et al.

Business
Applications

Enterprise Interfaces

M2M / IoT Integrate


Integration Act
Platform Store
Connect & Control
M2M / IoT
Protocol +

Multi-
Collect
Service
Elaborate
Gateway
Communicate
Field Field Interfaces
Data

Fig. 6.1 Enterprise IoT architecture model

interlinked devices, are the vital foundation in IoTs. Every device is fundamental for
obtaining information from the different sensors, which have the potential to track
the crucial parameters. These devices can also be applied on controlling the condi-
tion of equipment, for instance, switching off the faulty machines or rectifying the
RPM of a revolving gear. These device layers have the capacity to assure the final
mile connection to the remote equipment. It therefore presents the present status of
network devices alongside the capability to remotely control these devices.
The gateways, which are also known as the control tiers, are known as the inter-
mediaries that potentially facilitate the transfer of information, offload processing
actions, and drive functions. Due to the fact that some sensors produce a lot of data
points in every second, these gateways assure a place for pre-processing of data
locally before being sent to the relevant cloud tiers. Whenever information has been
aggregated in the gateways, it is summarized and strategically evaluated. Limiting
the amount of information can have a significant implication on the networking
transmission costs, mostly on mobile networks. Moreover, this permits a crucial
organization rule, which is applicable depending on the amount of data streaming
in. The control tier is considered bidirectional. This can possibly provide control
data to the relevant devices to facilitate configuration transitions. During the same
moment, it can possibly respond to information-tier command-and-control requests,
like the security requests used in the process of authentication.
The low-power and legacy devices cannot be used to directly register or com-
municate data to the various IoTs platforms. In this process, the gateways are con-
sidered in the entire networking picture. This is sometimes referred to as the edge
devices, whereby the gateways are used as a proxy to the various networking
6 Enterprise Architecture for IoT: Challenges and Business Trends 129

devices. These gateways are accepted to route the commands that are transferred
from the back end to the relevant networking devices. An edge device or the locally
available gateways present a significant market chance. The networking and hard-
ware vendors are focusing to capture their own share of the market through the
process of augmenting their switches, routers, edge devices, and firewalls meant to
double up the various IoT gateways. The inclusivity of the novel IoT devices, edge
devices, and legacy machines forms a crucial device layer of the IoTs stacks.
The cloud tier, data center, and the IoTs platform are obliged to undertake a
large-scale computation of information to produce the relevant insight, which is
vital in the enterprise. It is therefore relevant in offering a back-end enterprise ana-
lytics to fully execute a complex event processing like the process of analyzing
information that is adaptable in the business world with rules that are inclined to
historical trends. Moreover, this process includes the dissemination of organiza-
tional guidelines downstream. The process requires scalability both horizontally
and vertically to effectively support an increasing number of linked devices. These
are meant to address the various IoTs problems. The vital functions of the IoTs
information center and cloud platform include the messaging routing and connec-
tivity, data storage, event analysis, device management, and application enablement
and integration [11]. The functioning capacity of the industrial IoTs depends on the
software platforms, which manage the status of devices, analyze them, store them,
and present the correct insights to enhance the process of decision-making.
Moreover, it acts as the middleware, which is meant to orchestrate the complete
flow of work. Provided the attributes of the cloud like reliability, scale, and elastic-
ity, it is currently becoming a crucial deployment environment of the IoTs platform.
The device layer denotes the cloud gateways, which are responsible for authoriz-
ing and authenticating devices meant to manage the flow of works. Moreover, it
ensures that secure communication is maintained between the various devices and
the centralized command centers. These gateways are purposeful in dealing with a
lot of protocols and the information formats. The heterogeneous device and locally
available gateways with disparate protocols are used in the process of registering the
cloud gateways. For instance, the locally available gateways and the devices can
possibly communicate with the cloud gateways via the SOAP, REST, AMQP,
XMPP, CoAP, MQTT, and WebSocket. Despite the inbound protocols, the cloud
gateways are significant for assuring a consistent view of devices to the remaining
components. An average enterprise IoTs deployment is responsible for thousands of
devices and sensors that have to be deployed in a lot of sites. Every device requires
registration and maintenance in the centralized repository that serves as the authori-
tative inventory that acts in place of the present condition of deployment. The device
registry represents a centralized inventory where each device is registered into a
system. Every device alongside the metadata is available in the individual registry.
Any component in a platform can query the network device registry to evaluate the
present status of a device alongside its capability.
130 A. Haldorai et al.

6.4 IoT Business Model

Progressively, all forms of businesses are purposed to replace their one-off and dis-
crete sales frameworks with the present subscription-centered framework, which
links up products and services within a framework for a long-term relation. These
subscription models are effective and relevant to both the customers and businesses.
The business is designated to more recurrent and consistent streams of revenues.
Clients are no longer subjected to massive capital investments but they are being
offered opportunities of scaling their capacities meant to access superior services
and support. In the current years, advance cloud computing, technologies, and con-
nectivity have made the process possible since the advent of the “as-a-service”
enterprise model. This process began with the software-as-a-service (SaaS) before
becoming adjusted software. The IoTs has assured a powerful novel enterprise
framework, which is considered as a shot in the arm. Moreover, the low-cost, inter-
linked IoTs sensors and the monitored devices, including the sold products, enhance
the kind of services produced in an enterprise. The actual-time and historical infor-
mation generated by the IoTs devices permits users to exercise preventive mainte-
nance. For instance, the automated alerting and the predictive maintenance have
become possible in the modern age. These developments are useful in enhancing the
technological investment connected to the IoTs [12].

6.4.1 Types of IoT Architecture Models

Device-to-device: “IoT devices within the same network that generally connect
using wireless PAN protocols, such as Bluetooth and ZigBee, are device-to-device
architectures,” the report illustrates.
Device-to-cloud: In such architectures, “IoT devices connect directly to the
cloud, typically using a long range communications network, such as cellular. For
example, IoT-enabled vehicle monitoring devices (such as those provided by car
insurance companies to drivers) collect data on the vehicle, such as distances and
speeds driven, and acceleration and braking rates. These data are then transmitted to
the cloud, analyzed in the cloud, and used by insurance companies to create tailored
insurance rates based on the driving data.”
Device-to-gateway: “Device-to-gateway architectures transfer information from
sensors to the cloud via a gateway device. The gateway collects the data and then
communicates the data to the cloud through additional network connectivity, such
as Wi-Fi or cellular connection.”
Cloud-to-cloud: “Cloud-to-cloud architecture, also known as back-end data
sharing, enables third parties to access uploaded data from IoT devices. For exam-
ple, smart buildings receiving data from smart thermostats and smart light bulbs can
send the data to a cloud via Wi-Fi. The collected data are then aggregated in cloud
1, which may be owned by the building as the conduit, a user can set a smart
6 Enterprise Architecture for IoT: Challenges and Business Trends 131

Fig. 6.2 Types of IoT architecture models

thermostat to activate when the user’s car approaches their house (the conditional
trigger in this example being the location sensor within the car).”
IOT-enabled enterprise architecture given in the figure is a blueprint for the
deployment of IT resources that support business services and processes. It is a
high-level view of six fundamental domains: service, process, information, applica-
tion, data, and infrastructure.
EA aligns new technology capabilities with innovative business architecture pro-
cesses and services. In Fig. 6.2, new architecture elements associated with the IoT
are accented in red. In the business architecture with IoT, commodity services with
standard processes and operational data are augmented with custom services deliv-
ered through configurable processes. These processes provide on-demand informa-
tion tailored to user preferences or customized processes delivering unique services
associated with multiple, diverse connected devices. Due to larger consumer mar-
kets, more configurable processes will be required to meet broader customer prefer-
ences [13].
In the emerging EA [14] in which the IoT is a force, the architecture is more
diverse and complicated. New data platforms, including Hadoop, NoSQL, and
cloud data, will be used to house huge volumes of data. The IoT will enable people
in their daily lives to access custom services implemented using configurable
­processes. For example, a smart refrigerator user can set food expiration policies
and rely upon automated leading signals before actual expiration dates are realized.
132 A. Haldorai et al.

A configurable process minimizes food loss and reduces consumer grocery cost. In
another example, a food manufacturer can signal consumers if a product has been
found to pose public health risks (e.g., the possible presence of E. coli) to minimize
consumer exposure to this risk. With real-time design, a food producer might search
purchased smart refrigerator data, identify the location of tainted food, and send an
urgent message to affected consumers, which can be broadcast among a cluster of
product users, increasing the speed and reach of the message within the targeted
customer sector. These innovative workflows are enabled through smart refrigerator
integration with applications and data located across selected infrastructure
platforms.

6.5 Enterprise IoT Monetization

One of the most vital objectives of the IoTs is to transform the mindset of doing
things in an enterprise. This mindset is relevant in the process of determining if
users can use a product-based monetizing approach or a service-based approach.
CSPs, ISVs, OEMs, and other relevant stakeholders are invited to contribute their
intellectual properties in advancing a significant interlinked ecosystem. This pro-
cess necessitates monetization of a framework meant to permit all the contributors
to effectively leverage the novel IoTs organizational models to assure the agility to
deploy novel application to accomplish quicker ROI. The application enablement
platforms such as SensorLogic are vital for bridging this necessity with a competent
device pre-built and on-boarding to the IoTs services.

6.5.1 Enable Flexible Monetization Models

The IoT is developing a novel enterprise framework, for instance, the PaaS (product-­
as-­a-service), whereby the OEM provides the device, but in the instance of charging
the client upfront, the OEM permits them to pay via flexible frameworks (monthly,
pay-per-use or metered, etc.). The Sentinel Software Monetization resolutions per-
mit the CSP and OEM to effectively apply the flexible monetization frameworks via
tested technologies.

6.5.2 License and Entitlement Management

The objective of software is gradually developing and will continue progressing


since there are a lot of interlinked applications which have emerged in the recent
years. ISVs have, for a long time, handled various problems to enhance the moneti-
zation of their smart properties, i.e., software. This involves challenges such as
6 Enterprise Architecture for IoT: Challenges and Business Trends 133

reverse engineering and piracy. Whereas the software industries have increased, the
incumbent and emerging device producers are still new to their pending concerns.
To effectively monetize the intellectual properties, there should be effective enforce-
ment of the licensing policies for all the devices, which must also be broken down
to the individual features in the devices. The Sentinel Embedded and Cloud Service
Monetization resolutions permit the OEM and cloud service vendors to effectively
enforce the licensing frameworks.

6.5.3 Software Upgrades

Due to the rapid development of the IoTs services, the OEM will necessitate a
robust infrastructure for the purpose of remote upgrading of features and software
on the devices. However, these upgrades necessitate advanced levels of reliability
and security to assure that the devices in the fields cannot be tampered with making
them safe. The Sentinel Embedded resolutions assure a system for secure down-
loading of the software upgrades, which make sure that the features can be deployed
for the purpose of generating new streams of revenues from the interlinked devices.

6.5.4 Device Management

As a result of proliferation of different endpoints, the difficulty of different device


protocols has gradually enhanced. Ranging from the standard-based to the proprie-
tary implementation, the device integration into the IoTs environment is considered
as a time-consuming process. To assist devices to communicate with the business or
cloud resources, the SensorLogic platform is relevant as an orchestrator. Based on
the application of the device translators, the device on-boarding is considered easier,
which reduces the set-up process of the devices in the network.

6.5.5 IoT Application Development

The SensorLogic platform promises a system that can be used for the rapid develop-
ment of novel applications for IoTs use cases. This system involves (but is not lim-
ited to) a wide range of web services for the purpose of administration, authentication,
alarm notification, and location identification. Moreover, it assures a pre-integration
of a third-party service such as connectivity and geo-fencing resolutions. The pre-­
developed building blocks are used by IoTs developers to enhance the formulation
of new use cases, in addition to the present blocks. Developer kits, such as the
Cinterion Concept Board, and application platforms, such as Java, are relevant for
prototyping and designing. For the client, this represents a quicker timeframe of
deployment and ROI for the purpose of IoTs application.
134 A. Haldorai et al.

6.6 Case Study

6.6.1 Waste Management

M1 Limited (M1), one of Singapore’s leading full-service communications provid-


ers, is working with OTTO Waste Systems Singapore Pte. Ltd. (OTTO), to provide
a litter bin management system to the National Environment Agency (NEA). The
new system is designed to enable the NEA to leverage technology to better manage
the deployment of litter bins, as well as to optimize cleaning resources. The NEA
has been exploring how data can be used to enhance the effectiveness and efficiency
of public cleaning. The new system utilizes IoT sensors fitted within litter bins to
track how full they are, so that the cleaning crew can be notified when they need to
empty these litter bins. The NEA can also monitor the usage of litter bins to gauge
if there are adequate bins in a particular area to serve the public. OTTO aims to
deploy up to 500 of these smart bins during the first quarter of 2019. M1 says the
reliable and secure city-wide coverage provided by its NB-IoT, together with its
support for industry standards, makes the technology well suited for large-scale
smart city applications, such as the proposed litter bin management system. Smart
city solutions can also benefit from NB-IoT’s power efficiency, which makes it via-
ble to use batteries in connected devices, thus reducing infrastructure and mainte-
nance costs. M1 developed the litter bin monitoring solution together with OTTO,
who supplies the litter bin receptacles, and SmartCity, who provides the centralized
management system. “The collaboration with M1, using their NB-IoT network for
smart waste management, allows our customers to enjoy easy access to useful real-­
time data for smarter planning and resourcing on waste management nationwide,”
says Christopher Lopez, Managing Director of OTTO Waste Systems. “We also see
the potential of such implementations to help consumers have a greener environ-
ment to live in.” “Extensive research and development were carried out to produce
the hardware and the methodology of installation to maximise the accuracy of the
measurement in waste level and pollution in the environment,” adds P. Renganathan,
Director of SmartCity. “Through the strategic cooperation with M1, we will help
companies to achieve greater cost savings and reach higher productivity.”

6.6.2  mart Cold Chain – Tracking Temperature


S
During Transit

In Thailand, mobile operator AIS has developed a mobile IoT-based solution for
monitoring the temperature of perishable goods during transportation as depicted in
Fig. 6.3. Fresh food, frozen food, medicine, and some other goods need to be kept
at a constant temperature during distribution to ensure that they don’t decay and that
they reach end customers in a pristine condition. Connected “cold chain” solutions
can be used to monitor the temperature of a cold storage container during transit and
maintain the quality of goods, reducing the number of claims from customers that
goods have decayed or been damaged during transportation.
6 Enterprise Architecture for IoT: Challenges and Business Trends 135

Fig. 6.3 Smart cold chain model

These solutions can be configured to send a notification to the supplier if the


temperature rises beyond a specific threshold. To meet the demand for a low-cost
solution that can be installed easily without impacting logistics companies’ existing
systems, AIS is using NB-IoT to connect on-board thermometers to its IoT plat-
form, which can be used to record, analyze, and display the resulting temperature
data. AIS says the compact battery-powered thermometer is cost-effective enough
to be deployed at scale, while its small size and independent power supply mean it
can be quickly and easily installed or moved to another location. The device can
measure the temperature between −50 and 20 degree Celsius. It can be configured
to transfer temperature data to AIS IoT platform every x interval such as every
3 min and alert when the temperature changes by more than x degree Celsius such
as one degree Celsius. AIS says it is also using NB-IoT to monitor the electrical
supply of the cooling system, allowing it to ensure there is sufficient power to cool
the goods being transported. If the power supply is not working properly, the system
is designed to relay the relevant data to the AIS IoT platform, thereby allowing the
logistics company to proactively resolve the issues before any serious damage is
caused. “By installing the temperature-measuring devices in cold chain logistic sys-
tems, the quality of perishable goods can be assured – Mobile IoT connectivity can
be used to notify the operating parties when issues arise and take necessary actions
to prevent any damage to the goods,” explains Asnee Wipatawate, Head of Enterprise
Product and International Service of AIS. “The quality of IoT solutions becomes
critical to mitigate this problem and therefore yield competitive advantages.”

6.6.3 Smart Security

The remote access management and monitoring of high-value, distributed infra-


structure assets is a historically difficult, labor-intensive, disconnected, and non-­
scaling burden. Singtel’s vision of a frictionless, smart, highly scalable perimeter
access control solution begins today with the planned commercial launch of
igloohome’s connected digital lock system as shown in Fig. 6.4. Singtel and
136 A. Haldorai et al.

Fig. 6.4 Smart


security lock

igloohome (a Singtel-funded start-up) are excited to announce the upcoming com-


mercial launch of their connected perimeter access solution, based on igloohome’s
connected lock technology.
The resolutions assure an actual-time management, scalable remote control, and
monitoring of the distributed infrastructure perimeter accessibility. igloohome
works with leading property developers throughout Asia, including but not limited
to Sansiri (Thailand), Capitaland (Vietnam), and Mitsubishi (Japan). Matthew Ng,
VP of Product of igloohome says: “We have adopted carrier-grade IoT network
technologies like LTE-M and NB-IoT as they are increasingly prevalent among
global operator IoT solutions deployments. In Singapore, we rely on Singtel’s
NB-IoT cellular network because of its wide coverage and high availability. These
high-quality public IoT networks give us faster time-to-market, and obviate the
need to deploy our own private network, or implement discrete connectivity hubs/
gateways. LTE-M and NB-IoT are very power efficient, making our battery-­powered
smart locks more appealing to end users.” igloohome is a worldwide partner of
Airbnb, Booking.com, and Agoda, works with over 50 distributors, and ships to
more than 90 countries. A global operator-deployed standard like NB-IoT is an
essential element in support of igloohome’s global business.
Rahul Mehta, IoT Product Lead of igloohome comments: “We have extensively
tested our NB-IoT locks across many countries in Asia, and they have performed
well among all deployment scenarios – from deep indoor, to outdoor, and even
remote locations. We are excited to meet the global market demand for our con-
nected smart locks solution, a task that’s simplified by a global IoT network stan-
dard like NB-IoT.” igloohome first proved its technology in the vacation rental
space, partnering with Airbnb to simplify host-controlled guest access without the
need for a physical key exchange. The solution further enhanced host peace of mind
with on-demand, detailed visibility of guest-specific room, site, or location access.
igloohome then broadened its offering to address the needs of different categories of
home and property owners, addressing the operational and security limitations of
physical keys and enabling use cases like time-sensitive, remote-monitored, and
controlled access for delivery and trade services and access expiration for former
tenants.
6 Enterprise Architecture for IoT: Challenges and Business Trends 137

6.7 Summary

Enterprise architecture is an important methodology for improving and cultivating


the business value of IT. The emerging capabilities of the IoT are valuable process-
ing of data and analytical contributions to new enterprise services and processes.
Consequently, the business value of the IoT can be realized and validated when an
EA framework is applied to IoT strategy and projects. When an EA framework for
IoT is constructed, it is not isolated from non-IoT components. Architectures should
integrate non-IoT and IoT into a unified blueprint. This framework is composed of
reference models, standards, and principles that encompass the expanded function-
ality of new non-IoT and IoT products and services. Today, new data technologies
and flexible platform configurations can be implemented to manage this data. From
a market perspective, user-enabled products are a major growth sector targeted by
many companies. Popular user-enabled products, including smartphones and wear-
ables, are in demand in broad consumer markets. User-enabled products are increas-
ingly powered by the IoT.

References

1. Shroff, G. (n.d.). Enterprise architecture: Role and evolution. Enterprise Cloud Computing,
39–48. https://doi.org/10.1017/cbo9780511778476.006.
2. Kale, V. (2019). Internet of things (IoT) computing. In Digital transformation of enterprise
architecture (pp. 413–435). Milton: CRC Press. https://doi.org/10.1201/9781351029148-22.
3. Tambo, T. (2017). Enterprise architecture beyond the enterprise – Extended enterprise archi-
tecture revisited. In Proceedings of the 19th international conference on enterprise informa-
tion systems. https://doi.org/10.5220/0006277103810390.
4. Anandakumar, H., & Umamaheswari, K. (2017, March). Supervised machine learning tech-
niques in cognitive radio networks during cooperative spectrum handovers. Cluster Computing,
20(2), 1505–1515.
5. Anandakumar, H., & Umamaheswari, K. (2018, October). A bio-inspired swarm intelligence
technique for social aware cognitive radio handovers. Computers & Electrical Engineering,
71, 925–937. https://doi.org/10.1016/j.compeleceng.2017.09.016.
6. Arulmurugan, R., & Anandakumar, H. (2018). Early detection of lung cancer using wavelet
feature descriptor and feed forward back propagation neural networks classifier. In Lecture
notes in computational vision and biomechanics (pp. 103–110). Cham: Springer. https://doi.
org/10.1007/978-3-319-71767-8_9.
7. Haldorai, A. R., & Murugan, S. Social aware cognitive radio networks. In Social net-
work analytics for contemporary business organizations (pp. 188–202). https://doi.
org/10.4018/978-1-5225-5097-6.ch010.
8. Arulmurugan, R., & Anandakumar, H. (2018). Region-based seed point cell segmentation and
detection for biomedical image analysis. International Journal of Biomedical Engineering and
Technology, 27(4), 273.
9. Suganya, M.,& Anandakumar, H.(2013,December). Handover based spectrum allocation in
cognitive radio networks. 2013 International Conference on Green Computing, Communication
and Conservation of Energy (ICGCE). https://doi.org/10.1109/icgce.2013.6823431, https://
doi.org/10.4018/978-1-5225-5246-8.ch012.
138 A. Haldorai et al.

10. Implementation: Developing Enterprise Architecture. (n.d.). From enterprise architecture to IT


governance, 167–194. https://doi.org/10.1007/978-3-8348-9011-5_7.
11. Rana, M. M., & Bo, R. (2019). IoT-based cyber-physical communication architecture:
Challenges and research directions. IET Cyber-Physical Systems: Theory & Applications.
https://doi.org/10.1049/iet-cps.2019.0028.
12. Al-Qaseemi, S. A., Almulhim, H. A., Almulhim, M. F., & Chaudhry, S. R. (2016). IoT archi-
tecture challenges and issues: Lack of standardization. 2016 Future Technologies Conference
(FTC). https://doi.org/10.1109/ftc.2016.7821686.
13. Planning: Creating Enterprise Architecture. (n.d.). From enterprise architecture to IT gover-
nance, 153–166. https://doi.org/10.1007/978-3-8348-9011-5_6.
14. Armour, F., Kaisler, S., & Bitner, J. (2007). Enterprise architecture: Challenges and implemen-
tations. 2007 40th annual Hawaii International Conference on System Sciences (HICSS’07).
https://doi.org/10.1109/hicss.2007.211.
Chapter 7
Semi-Supervised Machine Learning
Algorithm for Predicting Diabetes Using
Big Data Analytics

Senthilkumar Subramaniyan, R. Regan, Thiyagarajan Perumal,


and K. Venkatachalam

7.1 Introduction

This chapter introduces the concept of Big Data for predicting complications of
diabetes mellitus. The discussion includes descriptive and predictive investigations,
knowledge discovery and information disclosure, cloud processing, Hadoop
MapReduce, and machine learning strategies.

7.1.1 Machine Learning and Knowledge Discovery

1. Big Data
Big Data uses an assortment of components: social systems administration,
mobile computing, data analytics, and the cloud, which are collectively known as
SMAC. Modern research areas require huge amounts of information that are diffi-
cult to store and process quickly in a reasonable amount of time using customary
approaches. Thus, such information requires an innovative approach, such as Big
Data [1]: “Big data alludes to the instruments, procedures, and strategies enabling
an association to make, control, and oversee gigantic informational indexes and
storage facilities” [2].

S. Subramaniyan · T. Perumal
Dept of CSE, University College of Engineering, Pattukkottai, Rajamadam, India
R. Regan
Computer Science and Engineering, University College of Engineering, Villupuram, India
K. Venkatachalam (*)
School of Computer Science and Engineering, IT Bhopal University, Bhopal, India

© Springer Nature Switzerland AG 2020 139


A. Haldorai et al. (eds.), Business Intelligence for Enterprise Internet of Things,
EAI/Springer Innovations in Communication and Computing,
https://doi.org/10.1007/978-3-030-44407-5_7
140 S. Subramaniyan et al.

2. Big Data in Healthcare


The rapid growth of information in healthcare applications is related to the digi-
tization of patient information, installed frameworks in social insurance, and the
need for accessible and efficient mobile systems. Breaking down and deciphering
the enormous and complex datasets in healthcare is a challenge for specialists and
is known as a Big Data issue.
Big Data management is a significant job in healthcare. To a great extent, Big
Data analytics enables patients Well-Being of mind, framework and legitimate man-
agement. Big Data methodology considers issues related to the 4 Vs: volume, veloc-
ity, variety, and veracity.
3. Diabetes Mellitus
Diabetes mellitus (DM) is a metabolic issue with abnormal insulin regulation.
Insulin insufficiency results in higher blood glucose levels and impaired absorption
of sugar, fat, and proteins.
DM is a common endocrine issue, affecting approximately 200 million individu-
als worldwide. The incidence of DM is expected to increase drastically in the com-
ing years. DM can be generally classified as one of two types: type 1 diabetes (T1D)
and type 2 diabetes (T2D), based on etiology. T2D is the most common type, chiefly
characterized by insulin resistance. Fundamental causes of T2D include lifestyle,
physical activity, diet, and genetics. T1D is believed to result from the autoimmuno-
logical devastation of Langerhans islets facilitating pancreatic-β cells. T1D accounts
for 10% of diabetes cases around the world. Other types of diabetes can be charac-
terized by their insulin emission profiles and include gestational diabetes, endocri-
nopathies, maturity onset diabetes of the young (MODY), neonatal diabetes, and
mitochondrial diabetes. The main complications of DM include polyurea, polydip-
sia, and critical weight loss. Diagnosis relies upon blood glucose levels (fasting
plasma glucose = 7.0 mmol/L).

7.2 Related Work

DM has been unequivocally associated with a few complications, mostly because of


the constant hyperglycemia. DM has a broad scope of different pathophysiologies.
Many well-known complications have been identified, including vascular issues,
diabetic nephropathy, retinopathy, neuropathy, diabetic coma, and cardiovascular
disease. Because of the great morbidity and mortality associated with DM, preven-
tion and management have attracted considerable attention. Insulin is a fundamental
treatment for T1D, but is only given in specific instances of T2D, such as when
hyperglycemia cannot be controlled through diet, weight loss, exercise, and oral
medication. The targets of medication aim to
(a) Spare a patient’s life and mitigate complications.
(b) Prevent complications and thus extend life span.
7 Semi-Supervised Machine Learning Algorithm for Predicting Diabetes Using Big… 141

The most widely recognized anti-diabetic medications include sulfonylurea,


metformin, alpha-glucosidase inhibitors, peptide analogs, and non-sulfonylurea
secretagogues. Most of these medications have various side effects. Furthermore,
insulin treatment is associated with weight gain and hypoglycemic episodes. Hence,
antidiabetic medications and treatments are of extraordinary importance and simul-
taneously a clincial challenge.
1. Big Data Analytics
An immense amount of information is created each day by specialists, treatment
plans, medical reports, imaging, and body sensor devices, such as IoT gadgets,
which result in Big Data. Creating Big Data is certainly not a significant challenge
for healthcare experts, but obtaining helpful information from Big Data is.
Investigations are required to contribute Big Data to medical services. Accessible
medical information in a Big Data investigation enable health care providers to
provide efficient results, from which patients can be prescribed the best medications
or treatments. Currently, Cloud-based Big Data analytic environment is needed to
analyze semi structured, structured and unstructured Big Data from healthcare sector.
2. Descriptive/Predictive Analytics
Insightful investigations are fundamental to predict disease in a patient based on
the patient’s health parameters. Using machine learning, such as support vector
machine (SVM) calculations with data frameworks such as Hadoop, it is conceiv-
able to execute a huge volume of data sets effectively and quickly to predict the
likelihood of diabetes based on a patient’s lifestyle. Along these lines, a Hadoop
distribution in a conveyed condition is the most recent innovation.
3. Hadoop MapReduce
A lot of processing capacity is needed to separate helpful information in Big
Data. Distribution and/or circulation enable us to take quick calculations by utiliz-
ing the numerous CPUs or CPU centers to run a few calculations simultaneously.
Although parallel and disseminated registering are not identical, they fill a similar
principal need of separating huge problems into smaller ones that can be analyzed
simultaneously. In distribution or parallelization, parameters such as execution
speed, memory, and simultaneousness are significant elements. The memory limit
of a few machines can be used for distribution and/or parallel or circulated handling;
however, this can be overcome by using a single device to process the enormous
data collections.
4. Machine Learning Techniques
In computer science, artificial intelligence (AI), sometimes called machine intel-
ligence, is intelligence demonstrated by machines, in contrast to the natural intelli-
gence displayed by humans and animals. Machine learning technique takes over the
activity involving automated learning from data set. It initially learns the knowledge
and applies this knowledge to distribute for predictions. A wide area of artificial
intelligence enabled devices are implemented through information.
142 S. Subramaniyan et al.

Selection Preprocessing Transformation Data Mining Interpretation /


Evaluation

Selected Preprocessed Data Transformed


Data Data Subset Data Patterns Knowledge

Fig. 7.1 The general flow of a KDD process

The general meaning of machine learning was denoted by Mitchel:


A computer program is said to learn from experience Ex with respect to some class of tasks
Ta and performance measure Pe, if its performance at tasks in Ta, as measured by Pe,
improves with experience Ex [3].

Knowledge discovery in databases (KDD) is an area associating theories, tools, and


techniques to create data and mine useful patterns or knowledge. The many steps of
KDD (gathering, feature selection, transformation, knowledge discovery, pattern
evaluation and creation) are shown in Fig. 7.1. A full overview of KDD was pro-
vided by Fayyad et al., who defined it as follows:
KDD is the nontrivial process identifying valid, novel, potentially useful, and ultimately
understandable patterns in data.

Types of Machine Learning Tasks


Machine learning [4] tasks can be categorized into three main types, as shown in
Fig. 7.2:
1. Supervised learning – The system can learn the data by labeling 30% as train-
ing data,
2. Unsupervised learning – The system can adapt by unlabeling the data
3. Reinforcement learning – The system can communicate with static/dynamic
environments.

1. Supervised Learning
In supervised learning, a framework can “learn” things inductively with a capac-
ity known as target work, which is represented as an outflow of an output model
depicting the data and information. A target work can be utilized to estimate an
input variable, also known as a subordinate instance/variable or a yield instance/
variable, depicted from autonomous factors, information factors/attributes, or char-
acteristics. The arrangement for conceivable information estimations of the
7 Semi-Supervised Machine Learning Algorithm for Predicting Diabetes Using Big… 143

Fig. 7.2 Machine learning tasks

capacity, such as its area, are known as occurrences. Each case can be portrayed
with many of the qualities (traits/characteristics). The subset of everything being
equal; in which, it provides each and every variable plays an important role. The
implementation of data was denoted as a models. Therefore, to derive an objective
capacity, learning framework, or preparation subset, theory/elective capacities and
known theories, signified by a hypothesis, must be considered. Supervised learning
consists of two types of learning methods:
• Grouping
• Relapse
Arrangement techniques attempt to anticipate particular methods, such as blood
collections, whereas relapse techniques anticipate numerical qualities. The most
commonly used supervised systems are decision trees (DT), rule learning, and
instance-based learning (IBL), such as k-nearest neighbors (k-NN), genetic algo-
rithms (GA), artificial neural networks (ANN), and support vector machines (SVM).
2. Unsupervised Learning
In unsupervised learning, the machine is used to provide the hidden variable and
structure of the input data/associations among the different variables. Therefore, the
training data consists of variables without the target labels.

7.3 Association Rule Learning

Association rule mining denotes an interrelationship between data items that is used
in many techniques. It is receiving more attention in research on databases [5, 6].
This approach was proposed in the 1990s by Rakesh Agrawal in a market bin
144 S. Subramaniyan et al.

examination. In this shopping/marketing basket model, association rules were rep-


resented as the structure of {A1, …, An} → B, which implies that when you find all
of A1, …, An in a truck, then it is conceivable to also find B. The Apriori algorithm
was proposed in 1994 by Rakesh Agrawal.
Association rule mining is mainly used for market bin investigations, including
bioinformatics. Another application in science and bioinformatics incorporates an
organic succession investigation, the examination of high-quality information, and
others. A careful survey of finding regular examples and association rules originates
from organic information, calculations and unsupervised applications.

7.3.1 Clustering

Clusters are educational examples that occur through grouping, such as the partition
of an entire dataset into groups of information [7]. The goal is to gather similar
examples for comparison under a particular circumstance.
3. Reinforcement Learning
Reinforcement learning is a type of machine learning used to group methods
where the framework covers to learn and act through direct contact with nature for
enhancing the general ides for execution. Reinforcement learning is predominantly
applicable to self-ruling frameworks, because of the freedom to connect its condi-
tions [8].

7.4 Diabetic Mellitus and Machine Learning

The next section describes the significant processes for DM [9].

7.4.1 Biomarker Identification and Prediction of DM

An enormous number of components are regarded as significant in DM. Obesity is


considered to be significant hazard impact, particularly in T2D, providing a very
strong connection among the types of DM. The diagnosis of DM can be aided by a
few tests. In the early stages, the findings in both T1D and T2D can affect the
following:
(a) Slowing progression of the disease.
(b) Directed choice of medication.
(c) Preventing future complications.
(d) Identifying related problems.
7 Semi-Supervised Machine Learning Algorithm for Predicting Diabetes Using Big… 145

Biomarkers (BM) (e.g., organic atoms) are quantifiable markers of a specific


condition associated with healthy and unhealthy states. Normally, BMs are described
as follows.
(a) Found in bodily fluids (blood, saliva, or urine).
(b) Experienced data and information was decided based on the autonomous etio-
pathogenic robotic pathway.
(c) Used to screen for disease status and response to medical treatment.
Biomarkers will be the immediate conclusion of the use of infection itself with
different inconveniences. Current advances, including metabolomics, proteomics,
and heredity, will improve the use of BMs. In cases of DM, BMs can indicate hyper-
glycemia and other relevant issues in diabetes.
The following sections discuss diagnostic and predictive markers, as well as dis-
ease prediction.

7.4.2 Diagnostic and Predictive Markers

The primary class manages biomarker disclosure, which is a task for the most part
that performed through component choice methods. Following an element determi-
nation step, an order calculation is used to evaluate the expected precision of the
chosen characteristics [10–16].
Primarily, setup techniques are utilized in biomarker assessment. A medical
dataset containing 803 prediabetic females with 55 characteristics was used for a
few basic element determination calculations to predict DcMs. It presumed that the
better presentation that has been accomplished with wrapper strategies. In addition,
among the channel strategies utilized, balanced vulnerability showed the best out-
comes. It consolidated electromagnetic (E-M) calculations of the closest classifier
and an inverse sign test (IST). As a totally unique methodology to manage charac-
teristics in a medical dataset for diabetes, cluster-based (progressive grouping)
includes an extraction system using data on disease symptoms. Their approach
delivered groups to be used as characteristics for disease severity and patient read-
mission hazard forecasts.

7.5 Prediction of DM

The second class manages disease expectation and analysis. Various calculations
and methodologies have been applied, such as conventional AI calculations, outfit
learning methods, and association rule methods, to accomplish the order exactness.
In addition to the previously mentioned technique, Calisir and Dogantekin imple-
mented LDA–MWSVM, a framework used to diagnosis diabetes. This framework
146 S. Subramaniyan et al.

includes feature extraction techniques, such as Linear Discriminant Analysis (LDA),


followed by a Morlet Wavelet Support Vector Machine (MWSVM) classifier.
High-dimensional datasets may be used. For example, one dataset consisted of
4.1 million people and 42,000 factors from regulatory cases, pharmacy records,
social insurance use, and laboratory results for the years from 2005 to 2009. This
was used to assemble prescient models (in light of calculated relapses) for various
onsets of T2D. Various learning calculations appear to be viable methods for
improving arrangement exactness. The methodologies have additionally been used
in DM prediction.

7.5.1 Diabetic Complications

Complications from elevated glucose levels are an extraordinary field. The negative
effects of hyperglycemia can be classified as follows:
(a) Macrovascular complications, such as coronary supply route infection, fringe
blood vessel malady, and stroke.
(b) Microvascular complications, including diabetic neuropathy, nephropathy, and
retinopathy.
Hyperglycemia is considered to be a fundamental contributor to morbidity and
mortality in T1D and T2D. Many studies have demonstrated an association between
glycemia and diabetic microvascular complications in both T1D and T2D. Diabetic
complications can likewise be grouped by their severity and onset.
Additionally, both insulin resistance and hyperglycemia are involved in the
pathogenesis of diabetic dyslipidemia. This is significant because DM affects indi-
viduals with well-controlled blood glucose levels. Many of these instances have
been considered in AI and information mining applications [17–24].
Huang et al. [19] used a decision tree-based expectation device that joined both
genetic and clinical characteristics to identify diabetic nephropathy in individuals
with T2D. Leung et al. [20] examined a few AI strategies that incorporated halfway
least squares relapse, characterization and relapse trees, the C5.0 Decision Tree,
random forest, innocent Bayes, neural systems, and bolster vector machines. The
dataset was composed of both genetic (single nucleotide polymorphisms) and clini-
cal information. Age, time of diagnosis, systolic blood pressure, and hereditary
polymorphisms of uteroglobin and lipid digestion emerged as the most useful
indicators.
DuBrava et al. [21] used random forest to select specific characteristics for pre-
dicting diabetic peripheral neuropathy. In order of importance, the characteristics
selected were the Charlson Comorbidity Index score (100%), age (37.11%), num-
ber of pre-record methodology and administrations (29.701%), number of pre-file
outpatient treatments (24.223%), number of pre-list outpatient visits (18.302%),
number of pre-file research center visits (16.912%), number of pre-list outpatient
office visits (12.12%), number of inpatient treatments (5.913%), and number of
7 Semi-Supervised Machine Learning Algorithm for Predicting Diabetes Using Big… 147

torment related medicine remedies (4.404%). The general precision of the model
arrived at 89.01%. The Diabetes Preprocessing Research Activity (DiScRi) data-
base was used to predict cardiovascular autonomic neuropathy (CAN).
Staniery et al., [22] used choice trees and ideal choice way discoverer (ODPF) to
locate the ideal succession of tests to diagnose CAN. Abawajy et al. used relapse
and meta-relapse, in combination with the Ewing equation, to distinguish the
classesofn CAN, while conquering the issue of missing information. Cardiac abnor-
malities are common diabetic complications. They are considered to be a critical
connection between diabetes, coronary disease, and stroke.
Hypoglycemia (low glucose levels) mainly occurs from anti-diabetic medica-
tions and has an extraordinary effect on individuals with DM. AI strategies such as
random forest, bolster vector machines (SVM), k-closest neighbor, and innocent
Bayes were used by Wang et al., [24] to predict hypoglycemia in individuals with
T2D. Bolster vector relapse was used by Georga et al. for a similar investigation.
Furthermore, effectively distributed calculations were used by Jensen in a simi-
lar study.
Diabetic retinopathy (DR) is an eye disorder that occurs in individuals with T1D
or T2D. The longer a patient has diabetes, the higher is the risk of acquiring this
particular pathophysiological condition. D_R is a rule shows early warning sign &
it is portrayed with significant diabetic confusion. DR has two primary stages: non-
proliferative (NPDR) and proliferative (PDR) Information mining and AI have been
used to predict DR, principally by imaging. A comprehensive review of computa-
tional strategies for DR was published in 2013. A two-stage technique, diabetic
fundus image recuperation (DFIR), has been used to predict DR. The first stage
includes an analysis of advanced retinal fundus images. The subsequent stage uses
a help vector machine for the forecast. This resulted in an approach to determine a
patient’s need for referral based on evidence of DR-related injuries on retinal
imaging.

7.5.2 Classification

For semi-supervised learning strategies, the objective is to create a model that


results in one of a variety of potential classes using named and unlabeled datasets.
The easiest class of semi-supervised learning methods trains one model using one
learning calculation with many characteristics. For example, self-preparing first
trains a model on the marked models. The model is then applied to all unlabeled
information, whereby the models are positioned by the confidence in their predic-
tions. The most certain predictions are then included in the named models. This
procedure repeats until all unlabeled models are marked.
Another class prepares different classifiers by inspecting the preparation infor-
mation several times and preparing a model for each example. For example, tri-­
preparing first trains three models on the marked models using bagging for the outfit
learning calculation. At that point, each model is refreshed iteratively, whereby the
148 S. Subramaniyan et al.

other two models make predictions on the unlabeled models. Only models with
similar forecasts are used to with the first named guides to re-train the model. The
cycle stops when no model changes. Finally, the unlabeled models are marked uti-
lizing lion’s share casting a ballot in which at any rate two models must concur with
one another.

7.6 Summary

This chapter focused on predictive analytics with machine learning to analyze Big
Data in order to predict future complications in diabetic patients. A dataset of
4.101 million people and 42,000 factors from case management, pharmacy records,
medical records, and laboratory results between 2005 and 2009 was used to create
prescient models (to calculate relapse) for various onsets of T2D. The most certain
predictions were then included in the marked models. This procedure repeated until
all unlabeled models were named. The following class prepared different classifiers
by inspecting the preparation information several times and preparing a model for
each example. At that point, each model was refreshed iteratively, whereby the other
two models made predictions on the unlabeled models; only the models with similar
forecasts were used with the first named guides to re-train the model. The cycle
stopped when no model changes. Finally, the unlabeled models were named utiliz-
ing dominant part casting a ballot in which at any rate two models must concur with
one another.

References

1. American Diabetes Association. (2009). Diagnosis and classification of diabetes mellitus.


Diabetes Care, 32(Suppl. 1), S62–S67.
2. Bagherzadeh, F., Khiabani, A., Ramezankhani, F., Azizi, F., Hadaegh, E. W., & Steyerberg,
D. (2016, March). A tutorial on variable selection for medical prediction models: Feature
selection methods in data mining could improve the results. Journal of Clinical Epidemiology,
71, 76–85. https://doi.org/10.1016/j.jclinepi.2015.10.002.
3. Agrawal, R., & Srikant, R. (1994). Fast algorithms for mining association rules in large
databases. In Proceedings of the 20th international conference on very large databases
(pp. 478–499).
4. Cai, L., Wu, L., Li, D., Zhou, K., & Zou, F. (2015). Type 2 diabetes biomarkers of human
gut microbiota selected via iterative sure independent screening method. PLoS One, 10(10),
e0140827. https://doi.org/10.1371/journal.pone.0140827.
5. Haldorai, A. R., & Murugan, S. Social aware cognitive radio networks. In Social net-
work analytics for contemporary business organizations (pp. 188–202). https://doi.
org/10.4018/978-1-5225-5097-6.ch010
6. Fayyad, U., Piatetsky-Shapiro, G., & Smyth, P. (1996). From data mining to knowledge dis-
covery in databases. AI Magazine, 17, 37–54.
7. Abawajy, J., Kelarev, A., Chowdhury, M., Stranieri, A., & Jelinek, H. F. (2013). Predicting car-
diac autonomic neuropathy category for diabetic data with missing values. Computers in Biology
and Medicine, 43(10), 1328–1333. https://doi.org/10.1016/j.compbiomed.2013.07.002.
7 Semi-Supervised Machine Learning Algorithm for Predicting Diabetes Using Big… 149

8. Agrawal, R., Imielinski, T., & Swami, A. (1993). Mining association rules between sets of
items in large databases. In Proceedings of the ACM SIGMOD conference on management of
data (pp. 207–216).
9. Georga, E. I., Protopappas, V. C., Polyzos, D., & Fotiadis, D. I. (2015). Evaluation of short-­
term predictors of glucose concentration in type 1 diabetes combining feature ranking with
regression models. Medical & Biological Engineering & Computing, 53(12), 1305–1318.
https://doi.org/10.1007/s11517–015–1263-1.
10. Anandakumar, H., & Umamaheswari, K. (2017, September). An efficient optimized handover
in cognitive radio networks using cooperative spectrum sensing. Intelligent Automation & Soft
Computing, 1–8. https://doi.org/10.1080/10798587.2017.1364931.
11. Huang, G.-M., Huang, T.-Y., & Lee, J. (2015). An interpretable rule-based diagnostic clas-
sification of diabetic nephropathy among type 2 diabetes patients. BMC Bioinformatics,
16(S-1), S5.
12. Jelinek, H. F., Stranieri, A., Yatsko, A., & Venkatraman, S. (2016). Data analytics identify gly-
cated haemoglobin co-markers for type 2 diabetes mellitus diagnosis. Computers in Biology
and Medicine, 75, 90–97. https://doi.org/10.1016/j.compbiomed.2016.05.005.
13. Lagani, V., Chiarugi, F., Thomson, S., Fursse, J., Lakasing, E., Jones, R. W., et al. (2015,
May–June). Development and validation of risk assessment models for diabetes-related com-
plications based on the DCCT/EDIC data. Journal of Diabetes and its Complications, 29(4),
479–487. https://doi.org/10.1016/j.jdiacomp.2015.03.001.
14. Lagani, V., Chiarugi, F., Manousos, D., Verma, V., Fursse, J., Marias, K., et al. (2015, July).
Realization of a service for the long-term risk assessment of diabetes-related complica-
tions. Journal of Diabetes and its Complications, 29(5), 691–698. https://doi.org/10.1016/j.
jdiacomp.2015.03.011.
15. Lee, B. J., & Kim, J. Y. (2016). Identification of type 2 diabetes risk factors using phenotypes
consisting of anthropometry and triglycerides based on machine learning. IEEE Journal of
Biomedical and Health Informatics, 20(1), 39–46. https://doi.org/10.1109/JBHI.2015.2396520.
16. Leung, R. K., Wang, Y., Ma, R. C., Luk, A. O., Lam, V., Ng, M., et al. (2013). Using a multi-­
staged strategy based on machine learning and mathematical modeling to predict genotype–
phenotype risk patterns in diabetic kidney disease: A prospective case–control cohort analysis.
BMC Nephrology, 14, 162. https://doi.org/10.1186/1471-2369-14-162.
17. Anandakumar, H., & Umamaheswari, K. (2017, March). Supervised machine learning tech-
niques in cognitive radio networks during cooperative spectrum handovers. Cluster Computing,
20(2), 1505–1515.
18. Marling, C. R., Struble, N. W., Bunescu, R. C., Shubrook, J. H., & Schwartz, F. L. (2013). A
consensus perceived glycemic variability metric. Journal of Diabetes Science and Technology,
7(4), 871–879.
19. Marx, V. (2013, June 13). Biology: The big challenges of big data. Nature, 498(7453),
255–260. https://doi.org/10.1038/498255a.
20. Mitchell, T. (1997). Machine learning (p. 2). Singapore: McGraw Hill. ISBN:0-07-042807-7.
21. Haldorai, A. R., & Murugan, S. Social aware cognitive radio networks. In Social net-
work analytics for contemporary business organizations (pp. 188–202). https://doi.
org/10.4018/978-1-5225-5097-6.ch010.
22. Sacchi, L., Dagliati, A., Segagni, D., Leporati, P., Chiovato, L., & Bellazzi, R. (2015). Improving
risk-stratification of diabetes complications using temporal data mining. In Conference pro-
ceedings – IEEE engineering in medicine and biology society (pp. 2131–2134). https://doi.
org/10.1109/EMBC.2015.7318810.
23. Stranieri, A., Abawajy, J., Kelarev, A., Huda, S., Chowdhury, M., & Jelinek, H. F. (2013). An
approach for Ewing test selection to support the clinical assessment of cardiac autonomic
neuropathy. Artificial Intelligence in Medicine, 58(3), 185–193. https://doi.org/10.1016/j.
artmed.2013.04.007.
24. Wang, K. J., Adrian, A. M., Chen, K. H., & Wang, K. M. (2015). An improved electromagnetism-­
like mechanism algorithm and its application to the prediction of diabetes mellitus. Journal of
Biomedical Informatics, 54, 220–229. https://doi.org/10.1016/j.jbi.2015.02.001.
Chapter 8
On-the-Go Network Establishment of IoT
Devices to Meet the Need of Processing Big
Data Using Machine Learning Algorithms

S. Sountharrajan, E. Suganya, M. Karthiga, S. S. Nandhini,


B. Vishnupriya, and B. Sathiskumar

8.1 Introduction

With the advancement of technologies, a lot of devices are coupled together, gener-
ating data, and that data is very much bigger and complex. This huge volume of data
is termed as “big data.” Conventional method of data processing cannot manage
such huge voluminous data. A lot of business-related queries are answered using
this massive collection of data sets. So, this big data has to be processed in some
manner to solve our business problems. Recent developments in open-source plat-
form, namely, Spark and Hadoop, paved a way to handle and store this big data in a
cheaper and more efficient way. Machines apart from humans are solely responsible
in the generation of this massive data which is increasing day by day. Also with the
evolution of Internet of Things (IoT) and machine learning, the generation of mas-
sive data has skyrocketed. With the growth of big data increasing day by day, the
usefulness of this data is just underpinning. Cloud computing with its scalability
and elasticity expands the usefulness of big data and provides a path for the devel-
opers to build ad hoc clusters for small data subsets and to test it.

S. Sountharrajan (*)
VIT Bhopal University, Bhopal, India
e-mail: s.sountharrajans@vitbhopal.ac.in
E. Suganya
Anna University, Chennai, India
M. Karthiga · S. S. Nandhini · B. Vishnupriya
Bannari Amman Institute of Technology, Sathyamangalam, India
e-mail: karthigam@bitsathy.ac.in; nandhiniss@bitsathy.ac.in; vishnupriya@bitsathy.ac.in
B. Sathiskumar
VIT University, Chennai, India
e-mail: sathiskumar.b@vit.ac.in

© Springer Nature Switzerland AG 2020 151


A. Haldorai et al. (eds.), Business Intelligence for Enterprise Internet of Things,
EAI/Springer Innovations in Communication and Computing,
https://doi.org/10.1007/978-3-030-44407-5_8
152 S. Sountharrajan et al.

8.2 Business Use Cases of Big Data

The massive big data helps in addressing a wide range of business actions from
customer feedback to good analytics. Some of the use cases are listed below.

8.2.1 New Product Establishment

Top companies use big data to understand the needs of their customers. By analyz-
ing the past and latest services, key attributes are classified and predictive models
are created by understanding the relationship between these previous and current
attributes which in turn builds new products/services. Any plan to produce and
launch new products is made after specific analytics of the data from focused
groups, media, markets, and stores.

8.2.2 Predictions About Mechanical Failures

By analyzing structured and unstructured data hidden in organizations such as


equipment manufactured year, model, entries of the log, errors, and engine and
room temperature, mechanical failure of the machines can be predicted earlier
before a problem breaks out. This prediction helps the company to reduce the
amount spent for maintenance and to uplift their income.

8.2.3 Experience of the Customers

The experience of the customers can be known well than before with the availability
of big data. With the help of social media, log files, web surfing, and other data, big
data improves the customer experience knowledge and the delivery value. It helps
in handling the issues positively and thereby diminishes the customer’s stress.

8.2.4 Security Breaches

Security has become a big landscape nowadays. Complaints and security issues are
growing constantly. Data patterns related to fraudulence are easily identified from
big data and the valuable information is alienated to induce fast reporting.
8 On-the-Go Network Establishment of IoT Devices to Meet the Need of Processing… 153

8.3 Handing Huge Data

Each day 2.5 billion gigabytes of data are generated with an increase in complex
patterns to interpret. To process all those, we need an intelligent system. For exam-
ple, data from social media and sensory information collected by satellite in space
need to be recorded for analysis. The following primitive systems can be utilized to
break the enormous amount of data, which takes more time for human analysis [15].
Intelligent learning – is a journey, which indicates maturity level and capability of
automation of the business process to improve efficiency and reduce operational
cost. The system can do, think, learn, and adapt which in turn lead to artificial
intelligence.
Artificial intelligence – is where a machine performs the task that requires human
judgment, learning, or problem-solving skills using techniques like NLP, speech
recognition, machine learning, and many more.
Machine learning – is a technique to implement artificial intelligence, which uses
algorithms to understand patterns in a huge amount of data.
Deep learning – is a technique of learning complicated patterns in a huge amount of
data using different types of neural networking concepts.
Cognitive learning – is a combination of all these techniques to automate a task that
involves unstructured content decision-making or any complex challenge.
Among these buzzing techniques, the widely used techniques are machine learn-
ing and cognitive computing or learning.

8.4 Cognitive Learning (CL)

Efficient steering through a flood of unstructured information requires a new era of


computing called cognitive computing. Before understanding cognitive learning,
one should first know the meaning of cognition, which means context and reasoning
form the basis of the cognition system, which mimics humans’ reason and the pro-
cess through which they analyze information and draw an inference. Cognitive
learning plays a vital role in fetching data from the rational business aspects and
implements the solution in an efficient way [1]. CL is used in the field of analyzing
social media data and is the perfect assistant for treating a serious medical condition
as it speeds up interaction and decision-making accordingly. CL can merge data
from various sources of information upon balancing context and conflicting evi-
dence to suggest the best possible answers. To process the above-said data, cogni-
tive learning includes some learning technologies that use data mining, pattern
recognition, and NLP to mimic the way the human brain works, and these were used
to achieve the learning systems. To enact the above-said technologies, cognitive
learning uses its adjacent computing technique called machine learning.
154 S. Sountharrajan et al.

Machine learning algorithms are used to perform typical tasks and solve differ-
ent types of problems, which require a vast amount of structured and unstructured
data [2]. A refined way of identifying patterns and processing data leads the cogni-
tive learning system to move to the next level of anticipating new problems and
modeling possible solutions. To attain all these capabilities, it must have the follow-
ing five key attributes/features:

8.5 Machine Learning

Machine learning is a data-driven approach with a class of self-learning algorithms.


In traditional programming, the data is given as an input and the output is predicted.
In the machine learning approach, the data is provided and output is predicted by the
machines; in turn, machines learn from the data, find the hidden insights, and create
a model. Examples of machine learning in the industry are found in Google Maps
and Facebook – face recognition, virtual personal assistant, speech recognition,
natural language processing, and so on. The objectives of machine learning include
greater accuracy, greater coverage of problems, greater economy in obtaining solu-
tions, and greater simplicity of knowledge representation. Machine learning, like
human learning, has two aspects: a behavioral component and knowledge acquisi-
tion component [5].
The following are the features of the machine learning technique:
1. It adjusts the program action making use of the data to detect patterns in a data
set.
2. It provides an efficient way of implementing complex programs.
3. It uses iterative algorithms to explicitly find the hidden perspectives.
4. It automates analytical model building.
A typical machine learning life cycle has six steps:
1. Collecting data – relevant data is collected from various sources.
2. Data wrangling – it is the process of cleaning and converting data into a specific
format.
3. Analyzing data – it is the process of selecting and filtering the required data
using machine learning algorithms.
4. Train and test algorithm – use the appropriate algorithm.
5. Prediction – a new set of input is given during testing and the model will classify
the input according to the training data set.
8 On-the-Go Network Establishment of IoT Devices to Meet the Need of Processing… 155

8.6 Cognitive Computing and Machine Learning

Cognitive computing and machine learning can be used to support decision-making,


deliver highly relevant information, and optimize the available attention to avoid
missing key developments. Cognitive machine learning can perform operations
analogous to learning and decision-making in humans [6]. Intelligent personal
assistants can recognize voice commands and queries, respond with information, or
take desired actions quickly, efficiently, and effectively.
Implementing cognitive computing can help achieve these desirable results:
• It helps people make better decisions, take action more quickly, and achieve
more successful outcomes.
• It delivers relevant information and advice at the time of need.
• It reduces information overload and optimizes people’s available attention span.
• It allows people to act more efficiently and effectively.
• It reduces errors, minimizes loss and damage, and improves health and safety.
Cognitive computing tools such as IBM Watson, artificial intelligence tools such
as expert systems, and intelligent personal assistant tools such as Amazon Echo,
Apple Siri, Google Assistant, and Microsoft Cortana can be used to extend the abil-
ity of humans to understand, decide, act, learn, and avoid problems. Using these
approaches, one can enhance the capabilities of humans by augmenting their pow-
ers of observation, analysis, decision-making, processing, and responding to other
people and to routine or challenging situations. Cognitive learning principles are
ways to apply sentient thought to learning activities and in turn control learning
behavior and motivate toward more profitable results.
The following are the three basic cognitive principles of learning strategies [3]:
1. Effective learning – rather than responding to stimuli, its focus is on what you
know.
2. Emphasize structure – connect new information and sort the same.
3. Effective learning strategy – validate the information and act accordingly.

8.7 Cognitive-Machine Learning Algorithms

Machine learning deals with artificial intelligence algorithms to enable the com-
puter to think and learn as efficiently as humans. The main goal of machine learning
algorithm over cognitive computing is to design the process that helps in recogniz-
ing patterns and compose decisions based on input data. Machine learning is applied
for training cognitive networks.
One of the best machine learning techniques is supervised learning which leads
cognitive learning to work efficiently [4]. The feature of machine learning algorithm
design depends on the appropriate selection of one or more basic principles, which
are selected based on its learning tasks. The combination of these basic principles is
156 S. Sountharrajan et al.

suitable for learning tasks if only a short time is available for the solution and if an
optimal solution is not necessary, which should be solved by the algorithm. Ordered
version space search with the aid of score function and reduction of the number of
concept versions solves a convoluted learning task with a large number of training
examples that include the following:

8.7.1 Particle Swarm Optimization (PSO)

In 1995, James Kennedy and Russell Eberhart first described a nondeterministic


optimization technique named particle swarm optimization (PSO), which is a
swarm-intelligence-based approximate technique. In the area of cognitive comput-
ing, optimization techniques can be used to find the parameters/objectives that pro-
vide the maximum (or minimum) value of a target function for classification
algorithms such as artificial neural networks and support vector machines. These
classification algorithms often require the user to define certain coefficients, which
are to be found by trial and error or exhaustive search [9].
The following are the perceptions of particle swarm optimization:
• The PSO algorithm retains multiple hidden solutions at one time.
• Objective functions determine the fitness value at each iteration.
• At the search space, each solution is represented by a particle in the fitness.
• The particles “fly” or “swarm” through the search space and determine the maxi-
mum value returned by the objective function.

8.7.2 Bayesian Cognitive Learning Method

1. Bayes theorem is a linear classification to analyze feature probabilities and


assumes feature independence. It learns and predicts very fast and it does not
require lots of storage. The prevailing applications of Bayes algorithms are real-­
time prediction, multiclass prediction, text classification, recommendation sys-
tem, cognitive systems, and image processing [7]. Cognitive machine learning is
sympathetic in the best hypothesis h from some space H, given observed training
data. Bayesian learning is relevant for two reasons:
• Explicit manipulation of probabilities
• Perspective for understanding learning methodology
2. The following is a feature of Bayesian learning methods:
• Each observed training sample that could incrementally decrease or increase
the estimated probability is correct.
8 On-the-Go Network Establishment of IoT Devices to Meet the Need of Processing… 157

3. Final probability of a hypothesis is determined from the prior knowledge from


the observed data.
• Weighted probabilities with combined predictions of multiple hypotheses
were used to classify a new instance.
4. Bayes theorem and cognitive learning can be used for designing straightforward
learning algorithms, which is achieved through a brute-force MAP learning
algorithm. A Bayesian algorithm is a way to distinguish the behavior of learning
algorithms. The advantage of using the Bayesian approach is that it uses inde-
pendence assumption, which might not hold true in certain databases that could
contain many interactions between the predictor (independent) variables. It has
been applied to a variety of data mining analyses in a large number of domains.

8.7.3 Hill Climbing-GA (HCA-GA)

5. Hill climbing is a heuristic approach used for numerical optimization problems


in the field of cognitive learning. The features of hill climbing include (1) vari-
ants of test algorithms as it takes the feedback from the test procedures. It is an
iterative calculation that begins with an arbitrary answer for an issue, and after
those attempts, it improves the answer by making changes. The cognitive engine
has three major functions: optimization, decision-making, and learning.
6. HCA is good for finding a local minimum but not for a global solution [6]. Hill
climbing is coupled with the genetic algorithm to find a global solution in the
cognitive system. GA is a computer science search technique used to discover
approximate alternatives for issues of optimization and search.
7. In particular, it falls within the category of local search methods and is thus usu-
ally an unfinished search. HCA-GA is a method for hybrid optimization. In each
genetic iterative, after the genetic operation, we get an ideal person from the
entire population. Then the hill climbing algorithm optimizes the optimal indi-
vidual again.

8.7.4 Associative Memory Algorithm

8. Associative memory is a framework of content-addressable memory that maps


particular representations of inputs to particular depictions of output. Associative
memories store content in such a manner that information can be recovered later
by adding a tiny part of the material to the memory instead of giving the memory
an address. Associative memories are used in database engine algorithms, anom-
aly detection systems, compression algorithms, cognitive learning, and face rec-
ognition systems as construction blocks. Hopfield’s neural network is a classic
illustration of associative memory [8].
158 S. Sountharrajan et al.

It is a system that “associates” two patterns (X, Y) such that when one is encoun-
tered, the other can be recalled. Typically, XÎ {−1, +1} m, Y Î {−1, +1} n and m and
n are the length of vectors X and Y, respectively. The vector components can be
thought of as pixels when the two patterns are considered as bitmap images. There
are two associative memory classes: auto-associative and hetero-associative. An
auto-associative memory is used to collect a pattern earlier stored that is most simi-
lar to the present pattern, i.e., X = Y. On the other hand, in a hetero-associative
memory, the retrieved pattern is, in general, different from the input pattern not only
in content but possibly also different in type and format, i.e., X 1Y.

8.8 Big Data: A Challenge

Though big data offers plenty of use cases, there are lots of challenges in utilizing
it. First and foremost the size of the data is big. In spite of the advancement in new
technologies in storing and managing the data, the size of the data is increasing
rapidly day by day. Yet organizations are trying to determine the best solution to
store and retrieve their data in an efficient manner. Challenges not only occur in data
storage but also in using the valuable information the data hold according to the
needs. Data cleaning and data organization in a meaningful manner with respect to
the needs of the clients are still a challenging task. Nearly 60–70% of the time is
spent on the data preparation task itself. In spite of the challenges in storing and
retrieving big data, the technologies used in handling the big data are varying rap-
idly. Apache Hadoop technology is utilized a few years back and then Spark tech-
nology is introduced. Presently both Apache Hadoop and Spark are used together.

8.9 Complexity of Big Data in Internet of Things (IoT)

One of the huge revolutions brought in the technological world is the Internet of
Things (IoT) [10–12]. IoT helps to know about the things happening around us
using sensors and to infer useful information from the sensors to utilize our world
in a smart manner. Thus the life of the people has improved a lot in a smarter way
using these real-world applications. Smart city is one of the use cases of IoT where
a lot of applications operate together to make the city smart [13, 14, 16, 17]. The
biggest challenge in IoT is its huge collection of data. Data collected from the appli-
cations cannot be utilized as it has to be processed to obtain useful information.
Three approaches are utilized in data processing of the big data collected from
IoT applications:
• Cloud computing
• Edge processing
• Local computing
8 On-the-Go Network Establishment of IoT Devices to Meet the Need of Processing… 159

8.9.1 Local Computing

In this approach processing of data will be done in the place where data is com-
posed. No data will be communicated to the servers located in remote places. Useful
information if necessary to take advance decisions will only be communicated to
remote places. This in turn enhances the whole efficiency and performance and also
diminishes the added burden of transmission. The technology of local processing is
used in smart sensors. Smart sensors are sensors with inbuilt computation and com-
munication paradigm. A smart sensor collects data and then processes the collected
data to make smart decisions and also store the data for future reference. It also
provides two-way communication. It has become nowadays an integral tool in intel-
ligent systems and an important perspective in advance IoT applications. One of the
best examples is “smart wearables”. These wearables obtain information from the
environment, process and produce the user’s required information, and also com-
municate the relevant information to the external platforms if needed.
The new implementation of intelligent systems in industries using IoT brought
revolutionary changes thereby leading to the next phase of industrial revolution
through “industrial IoT (IIoT)” [19]. Network virtualization plays a noteworthy role
in providing flexible and manageable network connections [20]. Thus the intricacy
of the infrastructure is reduced because the resources of the network are utilized as
logical services rather than physical services. This facilitates the execution of set-
ting up smart methodologies to calculate the network usability and data flows from
an IoT application. But this resource utilization has to be properly carried out to
enhance the effectiveness and efficiency. This is a challenging task to the network
operators since active monitoring will produce an added overhead in the network.
Yet, a promising methodology has to be implemented in an intelligent manner based
on monitoring data partially and not as a whole.

8.9.2 Edge Processing

It is one of the emerging paradigms nowadays. It involves deploying storage and


processing capability at the border or at the edge of the network. It is created
between the data sensor area and the data cloud centers. Computation capability,
storage ability, and resources for network are deployed. This capability of gaining
computation capability near the data centers helps in attaining low latency, more
bandwidth, and less jitter to services [21]. Some of the approaches that are imple-
mented on account of edge processing are as follows:
1. Fog computing: It provides storage and computation capability by using fog
nodes which consist of devices like routers, switches, and gateways of the net-
work. These fog nodes with devices are considered as virtual nodes thereby con-
tributing to network virtualization facility. This capability leads to the larger
usage of fog processing in mobiles as well as in IoT devices.
160 S. Sountharrajan et al.

2. Mobile edge computing (MEC): It involves deploying cloud computation in the


base stations [22].
3. Cloudlets: It involves cloud computation facility without the facility of wide area
networks (WANs). Local data acquisition network area contains the servers
which are known as cloudlets.

8.9.3 Cloud Computing

It provides flexibility in yielding computation resources on pay basis. Currently, it


is the most widely available disruptive technology. The data centers are deployed in
large numbers and they are used by the users virtually with greater efficiency. This
helps the usage of IoT use cases to operate in various environments in a lively man-
ner without changing the infrastructure [40]. This facility increases the usage of IoT
as a service [23]. It leads to efficient computation by integrating various IoT devices
and other embedded systems thereby outsourcing advanced services and use cases
based on collected data. The big data that is accumulated in the cloud is analyzed
and useful information is gathered and outsourced to the stakeholders using data
mining techniques. Complexity and constraint arising during operation are resolved
and more favorable dynamic solutions are obtained.

8.9.4 Context-Aware Computing

Context-aware computing systems have a variety of functions like gathering context


information from sensor devices and application interface and storing in a reposi-
tory. Then context filtering of raw data is done to obtain useful information and
model it into a useful context. This context information is utilized by the system to
react and make appropriate decisions to the users via the interface. A recent advance-
ment in context-aware computing is web-service-based context-aware systems [18].
A wide variety of middleware are also available in providing context-aware compu-
tations. Context-responsive applications that are aware about the different situations
are designed as objects in RCSM [24]. The modeled objects gather data from differ-
ent situations and accumulate it into object containers where analysis in respect to
the situations is made. Context-aware applications are created using the infrastruc-
ture and programming framework Java Context Awareness Framework (JCAF)
[25]. The issues related to big data analytics are also addressed by JCAF. User-­
preferred decision supporting system with a programming framework is provided
by the middleware PACE [26]. Network-operated intelligent bots are represented in
CAMPUS infrastructure for supporting context reasoning related to users and envi-
ronment [27]. CoWSAMI [28] supports context-aware computing in pervasive
environments. Context awareness related to disaster management is supported by
web-service-related middleware ESCAPE [29]. It has both front end and back end.
8 On-the-Go Network Establishment of IoT Devices to Meet the Need of Processing… 161

Front-end components support context data sensing and data sharing for contextual
information processing through web services. Back-end components support stor-
age and processing of data from different front ends. Another web-service-related
intensive working environment for context sensing and reasoning is provided by
InContext [29]. Context-aware computing systems not only analyze and respond
according to the behavior of the environment but also perk up the response of the
software. It is a dynamic infrastructure with self-adaptable and configuring ability.

8.10 Issues in IoT Things and Big Data

Things in Internet of Things communicate with themselves and with surroundings


for data aggregation and to respond unaccompanied to the real world without any
human intrusion when taking important business decisions and social processes. It
also creates services for interactions with the smart devices connected over the
Internet and also changes their states when security breaches are accounted in main-
taining the privacy of the users. The sensor devices in IoT operate anywhere and
anytime. They gather different kinds of data for which it is meant for, like collecting
humidity data for monitoring temperature, gathering noise level in case of environ-
mental monitoring, obtaining biomedical data for patient monitoring system, etc.
These data are big data and there are a lot of research issues related to the data col-
lected. Sensor data are used by context-aware use cases to acclimatize to the changes
and respond accordingly. Sensor data not only are complex, highly dynamic, and
inaccurate and alter according to the time, but also these data have to be integrated
in a repository and the information has to be analyzed specific to the domain.
To determine the complex relations between situations and data from sensor,
appropriate machine learning techniques are employed. But the performance is
highly dependable upon the complexity of the sensor data. Hidden Markov model
(HMM) is utilized in most of the context-aware use cases where a Markov chain
method (series of events) is used to model the system [30]. The system contains lot
of finite hidden states and responses generated from these states. SVM (support vec-
tor machine) is a machine learning technique utilized for linear and non-linear data
classification [31]. The protective nature of SVM against over-fitting helps in man-
aging large feature spaces irrespective of the number of features.
With the wide knowledge of the environment and the needed context and how it
is to be used, developers decide which features from data are needed to support their
applications. But moving to the actual implementation from a design is a big task.
Help from different services is needed by the developers to achieve this task.
Webinos EU [32] is a funded project which provides a platform for a wide variety
of wireless applications. Multi-platform, application-specific systems related to
web technology are provided by Webinos for connected things. To do so, Webinos
at any time makes use of the knowledge about the recent state of the IoT device and
users to provide decisions based on the context. Thus Webinos is a good example in
providing cross-platform environment for third-party context-aware services.
162 S. Sountharrajan et al.

8.11 Opportunistic Data Distribution Maintenance for IoT

Access of data content from many devices has widely increased nowadays. The
most widely accessed data by most of the people via mobile devices are short text
messages, voice messages, and contacts. People access their SMS, mails, social
network accounts, calendar data, photos, videos, music, games, and apps from any-
where and anytime using Wi-Fi. These accesses via wireless media produce an ava-
lanche of data which has to be managed and filtered using big data platforms. In
opportunistic networks, mobile devices are used for communication to achieve
larger throughputs during failure of direct communication. These opportunistic net-
works are used for short-range communication when users are nearer to each other.
The network routes are dynamically created. If a node wishes to transfer data, it first
stores it and then passes the data across the network till it reaches the destination
node that is in closer proximity to the sender node. From this node the data has to
be relayed and this is one of the biggest challenges in opportunistic networks. A lot
of solutions have been proposed in determining the suitable relay node such as
choosing the highest centrality node [33], predicting the nodes that follow the same
interaction pattern everyday [34], and deciding upon contact and inter-contact time
of the node.
Data distribution is an imperative research area in opportunistic networks. In
opportunistic networks, the topology is unstable due to its dynamic nature. New
techniques based on design models in distributing data are proposed by different
authors where data is distributed from the sources to the really interested receivers
because of the unawareness of the sources and receivers previously. A lot of research
fields have emerged based on delay-tolerant networks (DTNs) and opportunistic
networks (ONs) for IoT use cases. These networks do not have a permanent infra-
structure as like traditional networks whose nodes are assumed as permanent and
topology is known beforehand. In traditional network, if a node likes to transmit a
message, it embeds the message into the network path, and through the path, the
message will reach its destination. In DTNs and ONs, the technique of message
transfer is different. No two nodes are neighbors at a different time period. Nodes
that are in close proximity can come in contact while there is data exchange. The
buffer space is not available for the nodes in DTNs and ONs as they are mobile
devices, so all the data cannot be exchanged at a time. It is relayed on store-carry-­
forward paradigm. These DTNs’ and ONs’ techniques are used mostly in IoT sce-
narios where the communication via traditional network approach is difficult
and costly.
Nowadays online mobile advertisers generate revenues by targeting a users’ wish
and preference. For revealing appropriate ads based on users’ preferences, precious
information about the users is collected by these advertisers. The security and pri-
vacy of the users’ data is not guaranteed. MobiAd is a solution provided for ensur-
ing users’ privacy by revealing the users’ preferences to the advertisers yet preserving
the users’ privacy [35, 36]. An example of a context-aware mobile application solu-
tion is CAPIM (context-aware platform using integrated mobile services) [37].
8 On-the-Go Network Establishment of IoT Devices to Meet the Need of Processing… 163

It is a next-generation application integrated with a lot of services to gather con-


text data. It uses the sensing ability of modern smartphones and other external sen-
sors while dynamically loading the needed smart services. For smart city application,
DTNs and ONs are the best methods. In smart city application, the sensors are
meant for monitoring and integrating the significant infrastructure of the cities, for
optimizing the usage of resources, and for planning future maintenance activities.
The changes in the environment are guessed and reacted upon automatically by the
sensors. The cumulative intelligence of cities is employed through smart city use
case which connects the physical, economic, social, and business activities of the
cities. The communication by smart city sensors is done through ONs and DTNs.
The data collected from all these sensors are gathered and processed efficiently
through big data platforms.

8.12  ata Distribution Techniques in Opportunistic


D
Networks

Data distribution in an opportunistic network is carried out through four popular


techniques. First, socio-aware overlay algorithm develops an overlay that consists
of nodes with high centrality values and high visibility [38]. Before starting the
communication, the technique assumes that an infrastructure exists and forms an
overlay with visible nodes from each community sector. The nodes are interacted by
starting gossiping after community detection. The gossiping distribution begins by
sending the message to random node groups. The node centrality measurement unit
is used to choose a hub in a network by the socio-aware overlay algorithm. Bluetooth
and Wi-Fi devices are utilized for discovering the nodes. Subscribe approach or
unsubscribe approach is followed by the socio-aware method. During communica-
tion, broker node information along with the centrality list and timestamp informa-
tion is passed to the nodes for subscriptions or unsubscriptions. During change of a
broker node due to loss of closeness, a new subscription list is exchanged. Updates
about the new broker nodes are transferred to all the brokers. Community brokers
help in propagating the information about subscriptions during gossiping. When a
broker receives a publication, it forwards it to all the other brokers, and then all the
brokers update their own publication list. When there are lots of members in its
community, the information is flooded to all the members through proper channels.
The community detection methodology is provided by community-related algo-
rithms such as socio-aware algorithm. Depending upon the knowledge of standpoint
of each and every node in a network, the socio-aware algorithm classifies the nodes
and builds a community structure. In this community structure, the first level of
nodes belongs to the same community, and these nodes have a large number of con-
tacts and a stable standpoint duration. The next level of nodes is familiar stranger
nodes; these also have a larger number of contacts, but the standpoint duration of
these nodes is short. The third level is completely stranger nodes which have a
164 S. Sountharrajan et al.

smaller number of contacts with shorter contact duration. Finally, the nodes with
smaller contacts but with higher standpoint duration are friend nodes. Community
detection of nodes is done in a decentralized manner due to the unreliable and
unfixed structure of the opportunistic networks. Each node should determine its
own local community. Two algorithms are there for community detection in distrib-
uted environments, namely, simple algorithm and k-clique algorithm. To identify
the local community, the nodes interact with the nearby devices and run a detection
algorithm. This methodology is used in the data exchange level between the nodes.
Each node maintains the content detail about the nearby confront nodes, the dura-
tion of their contacts, and the identified local community. In a wireless environment,
infrastructure is not possible, so a wireless podcasting method is used to facilitate
the content distribution among the podcasting devices in that wireless range [39].
In resource-constrained networks, the content is made available to all the users in
contact without high usage of resources by ContentPlace method [41]. This method
utilizes the knowledge of relationships among the users to decide upon the area to
position the user data. The method of design is based on two preferences: one is
grouping the users according to their interests and the other one is based on their
social relationships.

8.13 Research Path in Big Data Platforms for IoT

The actual research path leads to the evolution of Internet of Things which provides
a huge amount of data by connecting all things in the environment thereby raising
the demand requirements in a complex way. The increasing growth of data volume
day by day requires a highly scalable platform for processing the data, for managing
the traffic in the network, and for efficiently storing the data. For maintaining the
connection between the things during poor connection of Wi-Fi/wireless links, good
communication protocols are desired to enable the connection and to maintain the
network traffic. For storing, processing, and manipulating the data in a proficient
manner, new algorithms are needed. IoT helps in developing a lot of applications
that connect the people and the things around the environment. They help the urban
population in gathering information at their fingertips and also in mounting their
activities to improve the competitiveness.
In addition to this, IoT also enables outdoor-computing facility and user purpose
approaches to boost the lifestyle of the urban population like smart transportation,
social services, anytime healthcare applications, agency administrations, and smart
education. IoT is not a single technique; it is a combination of multiple technologies
which tend to modify the society in the forthcoming years. The advancement of IoT
depends on the growth of a huge number of research projects and journal publica-
tions. As estimated, the growth of IoT will accompany more than 50,000 billion
things of different types. To accompany the growth of IoT, new protocols for com-
munication and standard methodologies should be invented for communication
8 On-the-Go Network Establishment of IoT Devices to Meet the Need of Processing… 165

between device and people. Web services can be used to embed the smart IoT
devices and their applications. By increasing the consumption of resource time,
response time, throughput time, and reliability, the quality of providing services
could be enhanced. The availability of a lot of services and instant publishing/sub-
scribing and notifying capability pave a path for easy management of the complex
architecture of IoT.
The information obtained from IoT devices should be well processed so that the
quality of the information is improved. The number of IoT devices is increasing, so
proper management is required to provide a highly robust, autonomous, and intel-
ligent operation. Solutions for self-organization, self-healing, self-protection, and
self-optimization of IoT devices are needed. For distributed information storage and
sharing, new computational services are available. For providing security and pri-
vacy to all IoT devices, new protocol mechanisms are essential. For context-aware
IoT models, stronger security measures are essential. For saving the energy and for
developing self-sustainable models, a new methodology is required.
A lot of research projects are going on to make the smart IoT objects gain their
required energy from the environment itself, thereby facilitating smart energy man-
agement. In addition to this, new platforms and advanced techniques are essential
for real-time storage and processing of big data to accomplish guaranteed data
availability and provisioning in real time. One available technique for big data is
cloud computing which simplifies the construction of big data infrastructure. It also
facilitates massive big data storage and processing thereby satisfying the customer
demands. New cloud computing capabilities are required for global collection of
different kinds of big data and for distributed processing and secure communication
among different platforms throughout the globe. When interoperability of IoT
devices is considered, a huge variety of technologies and design procedures are
involved. One approach is using standardized protocols for enabling inter-­
communication, and in addition to this, the self-configuring and self-managing
abilities of smart devices are essential for inter-operability and inter-communication
within the smart devices and with the surroundings. This methodology is highly
preferable than centralized management of complex IoT infrastructure. The autono-
mous nature of IoT devices is preferred in their operational level. These automatic
respondents of smart devices enable the building of complex infrastructure accord-
ing to the changes in the environment. Special methodologies are needed for
enabling low-power-consumption devices. Power management should be done
starting at the operational level of devices to network routing level. This would
enable us to develop complex expandable infrastructure at a low cost rate.
Due to the distributed nature of the surroundings, different issues may arise in
the operations and decision-making of the collaborated devices. The major issue
lies in device convergence and facilitating a quality solution. The other issue lies in
aiming to provide a correct solution thereby avoiding duplication and malpractice.
Altogether, things should be capable of preserving confidentiality, integrity, and
data availability.
166 S. Sountharrajan et al.

8.14 Summary

The evolution of Internet of Things leads to the growth of highly complex environ-
ments thereby raising demand requirements toward current and future research. The
tremendous growth of data from IoT things requires highly reliable and scalable
environments for supporting network traffic and storage and processing of data as
needed by the consumers. For continuous connectivity of devices in wired and wire-
less links, standard communication protocols are essential to manage the high traf-
fic in the network as well. In addition to this, new solutions are required for efficient
storage, fetching, and searching of data in these complex environments. Such a
solution would include forming dynamic networks of IoT devices based on the
needs to process big data using any of the listed machine learning algorithms to
provide useful information to the enterprise, which in turn helps in making the right
decisions.

References

1. Ahmed, E., & Rehmani, M. H. (2017). Mobile edge computing: Opportunities, solutions, and
challenges. Future Generation Computer Systems, 70, 59–63.
2. Anandakumar, H., & Umamaheswari, K. (2017, March). Supervised machine learning tech-
niques in cognitive radio networks during cooperative spectrum handovers. Cluster Computing,
20(2), 1505–1515.
3. Anandakumar, H., & Umamaheswari, K. (2017, September). An efficient optimized handover
in cognitive radio networks using cooperative spectrum sensing. Intelligent Automation & Soft
Computing, 1–8. https://doi.org/10.1080/10798587.2017.1364931.
4. Arulmurugan, R., & Anandakumar, H. (2018). Early detection of lung cancer using wave-
let feature descriptor and feed forward back propagation neural networks classifier. In
Lecture notes in computational vision and biomechanics (pp. 103–110). https://doi.
org/10.1007/978-3-319-71767-8_9.
5. Athanasopoulos, D., et al. (2008). CoWSAMI: Interface-aware context gathering in ambient
intelligence environments. Pervasive and Mobile Computing, 4(3), 360–389.
6. Boldrini, C., Marco, C., & Andrea, P. (2008). ContentPlace: Social-aware data dissemination
in opportunistic networks. In Proceedings of the 11th international symposium on Modeling,
analysis and simulation of wireless and mobile systems. ACM, Vancouver, British Columbia,
Canada.
7. Catherine, F., & Tamilarasi, J. S. (2018). Artificial intelligence – Deep learning-based cognitive
ontology model. International Journal of Innovative Technology and Exploring Engineering
(ITITEE), 8(2S2).
8. Celesti, A., et al. (2016). Exploring container virtualization in IoT clouds. In 2016 IEEE inter-
national conference on Smart Computing (SMARTCOMP). IEEE.
9. Ciobanu, R. I., & Ciprian, D. (2012). Predicting encounters in opportunistic networks. In
Proceedings of the 1st ACM workshop on high performance mobile opportunistic systems.
ACM.
10. Dobre, C., Flavius, M., & Valentin, C. (2011). CAPIM: A context-aware platform using inte-
grated mobile services. In 2011 IEEE 7th international conference on intelligent computer
communication and processing. IEEE, Cluj-Napoca, Romania.
8 On-the-Go Network Establishment of IoT Devices to Meet the Need of Processing… 167

11. Doukas, C., et al. (2007). Patient fall detection using support vector machines. In IFIP inter-
national conference on artificial intelligence applications and innovations. Boston: Springer.
12. Giacobbe, M., et al. (2018). Evaluating information quality in delivering IoT-as-a-service. In
2018 IEEE international conference on Smart Computing (SMARTCOMP). IEEE, Taormina,
Sicily, Italy.
13. Gil, D., et al. (2016). Internet of things: A review of surveys based on context aware intelligent
services. Sensors, 16(7), 1069.
14. Haddadi, H., et al. (2011). Targeted advertising on the handset: Privacy and security chal-
lenges. In Pervasive advertising (pp. 119–137). London: Springer.
15. Haldorai, A., Ramu, A., & Murugan, S. Social aware cognitive radio networks. In Social
network analytics for contemporary business organizations (pp. 188–202). https://doi.
org/10.4018/978-1-5225-5097-6.ch010.
16. Henricksen, K., & Ricky, R. (2006). A survey of middleware for sensor networks: State-of-­
the-art and future directions. In Proceedings of the international workshop on Middleware for
sensor networks. ACM, Melbourne, Australia.
17. Hui, P., Jon, C., & Eiko, Y. (2010). Bubble rap: Social-based forwarding in delay-tolerant
networks. IEEE Transactions on Mobile Computing, 10(11), 1576–1589.
18. Kalyan, A., Srividya, G., & Sridhar, V. (2005). Hybrid context model based on multilevel
situation theory and ontology for contact centers. In Third IEEE international conference on
pervasive computing and communications workshops. IEEE, Kauai Island, HI, USA.
19. Kanda, T., et al. (2008). Who will be the customer?: A social robot that anticipates people’s
behavior from their trajectories. In Proceedings of the 10th international conference on
Ubiquitous computing. ACM, Seoul, Korea.
20. Kim, H., Young-Jo, C., & Sang-Rok, O. (2005). CAMUS: A middleware supporting context-­
aware services for network-based robots. In IEEE workshop on advanced robotics and its
social impacts, 2005. IEEE, Nagoya, Japan.
21. Kristina, M., & Jan, P. (2003). Basic principles of cognitive algorithms design. In Proc. of the
IEEE International Conference Computational Cybernetics, Siófok, Hungary (pp. 245–247).
22. Lenders, V., Gunnar, K., & Martin, M. (2007). Wireless ad hoc podcasting. In 20074th annual
IEEE communications society conference on sensor, mesh and Ad Hoc communications and
networks. IEEE.
23. Macke, J., et al. (2018). Smart city and quality of life: Citizens’ perception in a Brazilian case
study. Journal of Cleaner Production, 182, 717–726.
24. March, H. (2018). The Smart City and other ICT-led techno-imaginaries: Any room for dia-
logue with Degrowth? Journal of Cleaner Production, 197, 1694–1703.
25. Mayank, A., Joshua, C. P., & Thomas, L. G. (2019). Using Machine Learning to guide cogni-
tive modeling: A case study in moral reasoning. arXiv preprint arXiv:1902.06744.
26. Mohammadi, M. (2018). Enabling cognitive smart cities using big data and machine learning:
Approaches and challenges. IEEE.
27. Ng, I. C. L., & Susan, Y. L. (2017). The Internet-of-Things: Review and research directions.
International Journal of Research in Marketing, 34(1), 3–21.
28. Saarikko, T., Ulrika, H. W., & Tomas, B. (2017). The Internet of Things: Are you ready for
what’s coming? Business Horizons, 60(5), 667–676.
29. Samir, M. (2017). Cognitive computing architectures for machine (deep) learning at scale.
MDPI, Presented at the IS4SI 2017 Summit DIGITALISATION FOR A SUSTAINABLE
SOCIETY, Gothenburg, Sweden.
30. Satyanarayanan, M. (2017). The emergence of edge computing. Computer, 50(1), 30–39.
31. Su, Y., et al. (2018). Dynamic virtual network reconfiguration method for hybrid multiple
failures based on weighted relative entropy. Entropy, 20(9), 711.
32. Tomáš, K., Štěpán, B., & Johannes, F. (2018, April). A review of possible effects of cognitive
biases on the interpretation of rule-based machine learning models. arXiv:1804.02969v1 [stat.
ML].
168 S. Sountharrajan et al.

33. Truong, H., & Schahram, D. (2009). A survey on context-aware web service systems.
International Journal of Web Information Systems, 5(1), 5–31.
34. Van Kasteren, T., et al. (2008). Accurate activity recognition in a home setting. In Proceedings
of the 10th international conference on Ubiquitous computing. ACM.
35. Vergori, P., et al. (2012). The webinos architecture: A developer’s point of view. In International
conference on mobile computing, applications, and services. Berlin/Heidelberg: Springer.
36. Williams, J. A., Weakley, A., & Cook, D. J. (2013). Machine learning techniques for diagnos-
tic differentiation of mild cognitive impairment and dementia. AAAI Workshop.
37. Yau, S. S., & Fariaz, K. (2004). A context-sensitive middleware for dynamic integration of
mobile devices with network infrastructures. Journal of Parallel and Distributed Computing,
64(2), 301–317.
38. Yingxu,W. (2017).Novel machine learning algorithm for cognitive concept elicitation by cog-
nitive robots. IJCINI.
39. Yoneki, E., et al. (2007). A socio-aware overlay for publish/subscribe communication in delay
tolerant networks. In Proceedings of the 10th ACM symposium on modeling, analysis, and
simulation of wireless and mobile systems. ACM, Chania, Crete Island, Greece.
40. Zhihan, L., et al. (2018). Government affairs service platform for smart city. Future Generation
Computer Systems, 81, 443–451.
41. Zhong, R. Y., et al. (2017). Intelligent manufacturing in the context of industry 4.0: A review.
Engineering, 3(5), 616–630.
Chapter 9
Analysis of Virtual Machine Placement
and Optimization Using Swarm
Intelligence Algorithms

R. B. Madhumala and Harshvardhan Tiwari

9.1 Introduction

Virtualization is one of the powerful techniques that introduced a wide variety of


applications to the real world. Cloud computing is considered as a computer para-
digm for computing and delivering services over the Internet with many exciting
features and provides three major service frameworks such as the software as a
service (SaaS), the infrastructure as a service (IaaS), and the platform as a service
(PaaS). Consumers are always concerned about the performance of the application
that they are using and are very much interested in providing efficient resources.
The predominant issues in effective cloud service provisioning are optimal resource
utilization and energy conservation over the network. Due to their emerging growth,
cloud providers also provide infrastructure as a service (IaaS) to their main opera-
tions. In cloud service models, the IaaS model provides virtualized hardware
resources where a single physical server is partitioned into multiple logical servers;
in turn, each logical server acts as a single physical server for problem execution.
The total energy used and the cost utilization in cloud services mainly depend on
virtual machine scheduling. Efficient resource management through virtual machine
placement is a great concern in data centers. Allocating resources for cloud data
centers to formulate a virtual machine from a physical machine makes it a big prob-
lem. While allocating, there is a need to consider the basic parameters to optimize
resource allocation. We are mainly concentrating on the energy consumption as well
as the cost minimization of cloud data centers.

R. B. Madhumala (*)
Department of Computer Science Engineering, Jain University, Bangalore, India
H. Tiwari
CIIRC, Jyothy Institute of Technology, Bangalore, India

© Springer Nature Switzerland AG 2020 169


A. Haldorai et al. (eds.), Business Intelligence for Enterprise Internet of Things,
EAI/Springer Innovations in Communication and Computing,
https://doi.org/10.1007/978-3-030-44407-5_9
170 R. B. Madhumala and H. Tiwari

The cloud service providers provide a group of computational memory and stor-
age units in the form of virtual machines. Each physical bare-metal unit is divided
into a number of VMs. The main concern of the cloud provider is to identify a
proper physical machine to deploy the VMs meant at reducing the physical
machine’s running while maintaining service quality. This process is termed as vir-
tual machine placement optimization.
The researchers continuously carried out the effective optimization of VMs. Few
kinds of research used heuristic methods while others used metaheuristic methods
with a flavor of artificial intelligence.
Optimization helps to completely utilize all the resources of a currently running
physical machine (Fig. 9.1). Optimization helps in reducing power consumption
thereby improving the green computing. In real-life situations, there will be many
constraints for optimizing the cloud center. Maximizing the efficiency of the cloud
center and minimizing the undesired factors is a challenging task to solve.
Optimization is a trial and error method. Many new algorithms will be proposed
and their results are tested against the desired metrics and the successful ones are
continuously modified to get better results. Broadly we can classify the optimiza-
tion algorithms as conventional and non-conventional algorithms. Much of the work
is already done in the conventional algorithms, but very few are proposed in non-­
conventional algorithms. The non-conventional algorithms are primarily the bio- or
nature-inspired algorithms. These nature-inspired algorithms need intelligence and
the AI tools are used for inducing the required intelligence into the algorithms. Few

Fig. 9.1 Virtual machine placement in cloud computing


9 Analysis of Virtual Machine Placement and Optimization Using Swarm Intelligence… 171

optimization algorithms are listed under heuristic methods as well as metaheuristic


methods. In virtual machine selection, the overloaded host is detected and VMs are
further selected to offload the host mainly to address the performance issue [1].
After selecting VMs, the host once again is checked if it still remains overloaded. In
the next step, it is applicable to select another VM, which is purposed to migrate
from its host [2]. This process is thus repeated unless the host is considered as
unoverlooked. The aspect of virtualization is a vital concept of cloud computing
which determines the manner in which a suitable host for the virtual machine is
analyzed. As indicated above, the inspiration behind the placement of VM is
inscribed in the traffic-aware, application-aware, topology-aware, and energy-aware
prospects in Fig. 9.2.
Energy management—minimizing the cost at the level of hardware and server
consolidation: green computing. Resource management—resources should be
available based on the need and cannot be allocated statistically based on the peak
workload elasticity of the cloud traffic engineering: to maintain the data center
application efficiency and also for accurate planning of the network architecture.
Resources are limited to cloud data centers simply because cloud providers try to
minimize over-utilized resources. This helps them to have a minimum number of
servers as well as reduce the cost for both server hardware and software. Resources
like CPU, memory, and hard disk need to be kept at minimum to avoid unnecessary
costs. So some improvements are needed to reduce energy cost. This research has
analyzed the recent researches concerning the provision of resources in the process
of virtualizing computing ecological segment. It reduces response time and maxi-
mizes resource utilization but QoS factors are not considered [3]. It achieves sched-
ule with a lower makespan, but its time complexity is more. It minimizes the total
cost and high risk in resource management. Existing methods cannot operate on the
issues of non-coordinate framework like resolution to power field.

Optimization
algorithms

Approximate
Exact methods
methods

Iterative Enumerative Ad-hoc


Metaheuristics
methods methods heuristics

Constructive Population
Local search Trajectory
heuristics based

Fig. 9.2 Optimization algorithm classification


172 R. B. Madhumala and H. Tiwari

9.2 Literature Survey

This section provides insight into existing algorithms developed which can be used
for optimal virtual selection of machines and the inclusivity of cloud computing,
which is placed in the group of issues known as the NP-hard issues because of large
resolution spaces. Cloud computing includes tasks for mapping on an unlimited
computing resource, nature-inspired algorithm in ant colony optimization (ACO),
particle swarm optimization (PSO), and gravitational search algorithm (GSA),
which are vital for dealing with the NP-hard issues mentioned in this chapter. The
condition-motivated algorithm in the area of privacy is meant to safeguard the data
connected with cloud computing and task computing.
• Kumar et al. [4] presents a critical survey of nature clever algorithms that are
used in artificial intelligence and automation in real-life domains. Nature-­
inspired algorithms are an emerging area of research on an algorithm based on
physics and biology. This paper explains the idea of duplication in artificial
frameworks. The condition-inspired computing and computation intelligence
will assure maximum resolution to issues on a new venue of development and
research.
• Nature with its massively adverse, robust, dynamic, and complex idea is a great
source of information for solving problems in computer science. Biology-­
inspired computing has a wide field for research; in particular, there are great
opportunities in exploring a new approach. Nature-inspired algorithm is a vital
algorithm which increases the field of future-generation computing. Binitha et al.
[5] presents a broad view of biology-inspired optimization algorithms which in
turn increases the areas where these algorithms can be successfully applied.
• Cloud computing needs optimal resource utilization of cloud resources which in
turn needs a certain novel scheme with enhanced dynamic resource allocation,
collective control, resource management, and maximum distribution in network-
ing and computing resources. The present trend is using virtualization in resource
mobilization for data centers through machine migration techniques. Theja and
Babu [6] discusses proposed approaches for the optimization of resources for
cloud infrastructure. Virtual machines in association with physical machines can
be an effective solution for resource optimization by considering certain increased
predictive schemes, load balancing, and mapping which could be an increasing
factor in virtualization for better performance.
• The VM placement in the cloud is enhanced based on objects such as VM time
allotment, power consumption, SLA violation utility of resources, etc. [7]. This
includes proposed algorithms such as the multiplied hybrid ACO-PSO algo-
rithms that minimize the resource wastage and consumption of power to assure a
more balanced server. It helps in the reduction of server costs.
• Usmani and Singh [8] deals with details regarding VM placement algorithm aim-
ing at maximum utilization to reach optimal solution for minimization of power
consumption. These algorithms aim at studying the workload variability and
transforming the application demands, including the minimization of trade-­off
9 Analysis of Virtual Machine Placement and Optimization Using Swarm Intelligence… 173

between power consumption and better performance by using a hybrid technique


for server energy efficiency. This is designed to be a two-staged procedure that is
comprised of green computing and overload avoidance.
• VM migration is a source-intensive process to address VMs’ progressive demand
which is effective for the CPU cycles, communication bandwidth, and cache
memory capability. The continuous movement becomes essential in managing
the efficiency of data centers and in smoothing the application service. Choudhary
et al. [9] deal with the problem faced in VM migration by having them shifted
while there is a progressive operation making it possible for the VMs to be
changed with zero downtimes. This shows the various forms of content, which
include migration of the CPU states, storage contents, and memory contents. This
analyzes the post-copy, pre-copy, and hybrid method of VM migration. The VM
migration methods are sub-divided into two categories: models and frameworks.
• Cloud computing has attained remarkable growth in every field; provisioning,
scaling, and maintenance of applications are achieved and it serves in a breeze.
Rani and Bhardwaj [10] focuses on task scheduling using ant colony optimiza-
tion genetic algorithm, PSO, and GSA. This survey deals with task scheduling in
cloud computing based on current information and sources to build a good map-
ping relationship between tasks and resources. It compares the ant colony opti-
mization with other techniques to prove the former is better in comparison.
• The main goal of [11] is to provide an understanding of the present algorithms
and approaches that ensure an effective VM placement in the contest of cloud
identification and computing that will be applied in future systems. The sophis-
ticated VM placement optimization aims to reduce work, power, and cost and
prevent congestion of data flow. The migration of VM requires a secure connec-
tion between the source and target servers. This aims to make a way for further
work to address this problem for establishing and managing better
communication.
• Cloud computing which is an important development in sharing and pooling of
resources over Internet services is still in its infancy to achieve improvement;
much research is required in various directions: one is proceduring the objective
of scheduling to trace enough resources. This is placed in the segment of prob-
lems known as the NP-hard problems. There is no algorithm that provides opti-
mized remedies within the polynomial timeframe to resolve this issue.
Metaheuristic-based methods provide some remedies within the speculated
timeframe. The metaheuristic techniques like ACO, GA, and PSO and two novel
approaches like the League Championship Algorithm (LCA) and bat algorithm
(getting inspiration from the echolocation behavior of bats, Yang introduced the
bat algorithm) form these techniques. The comparative framework of these algo-
rithms is centered on metaheuristic methods utilized for optimizing methods,
nature of obligations, and ecology where these algorithms are applied. Son and
Buyya [12] proposes priority-aware VM allocation (PAVA) which uses network
topology information to allocate VM on the host which is nearest to the requester
of the resource. The priority of the task is also considered as a parameter.
174 R. B. Madhumala and H. Tiwari

9.3 Optimization Techniques

9.3.1 Ant Colony Optimization Algorithm (ACO)

The ant colony optimization (ACO) algorithm is considered as one of the most
recent algorithms that is competitive to other forms of algorithms [13]. The ACO
algorithm is developed based on its combination with the ABC algorithm. The
hybrid algorithm is formulated to link up the advantages of the global research abil-
ity of the ABC algorithm and the localized search ability of the ACO algorithm
alongside the merits provided by the maximum and minimum algorithms. The algo-
rithm initially utilizes the advanced max and min algorithm meant to schedule
requests, whereas balancing is performed by ACO that is enhanced by the ABC
algorithm. The vital aim of using multiple-objective ant colony system algorithm
for the virtual machine placement problem is to obtain a collection of non-­dominated
resolution, the Pareto set, which minimizes the complete resource wastage and the
consumption of energy [14]. The projected algorithm is analyzed with the instances
recognized in literature. The remedies are performed and compared to the present
multi-purpose genetic algorithm and the two-objective algorithms referred to as the
bin-packing algorithm and the maximum and minimum ant system algorithm in
Fig. 9.3.
Whenever the ant has to locate the shortened path between the colonies to the
feeding sources, the moving ant produces some pheromone on the ground, which is
responsible for forming paths by a trail of substances. Whereas the isolated ant
migrates at a fundamentally randomized way, the ant that encounters the laid trail
may potentially determine and decide based on a high probability to embrace it. As
such, this reinforces a trail with the pheromone. This inclusive behavior that pops up
has created an autocatalytic condition whereby more ants following the trail are
attracted thus becoming a followed path. The procedure is grouped based on a posi-

Fig. 9.3 Ant colony


behavior in the search
space
9 Analysis of Virtual Machine Placement and Optimization Using Swarm Intelligence… 175

tive feedback loop, whereas the probability of the ant to select the path is developed
through the number of ants which have chosen a single path.

9.3.2 Variants of Ant Colony Optimization (ACO)

Ant system: The initial ACO algorithm promised was the ant system (AS). This seg-
ment was applied to a number of small cases of travelling salesman issues in
about 75 urban environments. On November 15, 1991, Dorigo submitted a man-
uscript for the ACO which was again revised on September 3, 1993, July 2, 1994,
and December 28, 1994, and it was finally published on IEEE in 1996. It was an
attempt by Dorigo to extend the travelling salesman (TSP) problem to the asym-
metric travelling salesman problem (ATSP) [15].
Dorigo proposed the following algorithm:
1. Initialize the time counter, cycle counter, and trail intensity to zero.
2. Place m ants in n nodes.
3. Make the ants move to different towns with a probability function.
4. Each time a path is retraced, update the trail counter and prepare a tabu list with
the values.
5. Steps 2 to 4 are repeated till the tabu list is full (all the towns are covered for each
ant).
6. Steps 2 to 5 are repeated until the cycle counter reaches a preset value.
7. From the table, find the path with the highest trail intensity, and this is considered
as the shortest path.
The complexity of the algorithms is determined by:
NC = number of cycles
n = number of towns
m = number of ants
Dorigo has even proposed two other extended versions of the algorithm, namely,
ant-density and ant-quantity algorithms, but there was not enough research done as
they didn’t become more popular like the original ant system. The main problem of
the ant colony algorithm is stagnation. When many ants travel in the problem space,
a lot of pheromones will be put on that path, and since the path with a lot of phero-
mones is considered as the best path, there is a chance that the other available paths
get ignored and this creates the stagnation in Fig. 9.4.
One issue which pops up based on the algorithms is stagnation. Whenever many
ants travel in this issue space, they are placed down on pheromones. However, as
many of them gravitate over the best resolution to locate it, it possibly creates star-
vation to the remaining datasets. Though few ants may take other paths, their phero-
mones will get evaporated eventually making that path as an unreliable path. This
means that only the path where many ants travel will have pheromone replenish-
176 R. B. Madhumala and H. Tiwari

Fig. 9.4 Basic flow chart of ant colony optimization algorithm (ACO)
9 Analysis of Virtual Machine Placement and Optimization Using Swarm Intelligence… 177

ment and all other paths will be ignored. When this happens, there is no possibility
of further optimization, and to combat this, many modified and advanced versions
of the ant system are engineered.
Elitist ant system: In the elitist AS, they have formulated the specialist ants along-
side the normal ant system. In the elite ant system, the bonus pheromones are
provided to the most relevant resolutions by the basis of multiplication of some
pheromones provided by the specialist ants. At each moment, the ultimate path
found is based on massive deposition of the pheromones that is initiated for a
relevantly long duration of time. While the usual paths degrade as time goes, the
elitist paths are rendered as a huge concentration of pheromones that stay and are
a validated option for a long duration of time. As such, one elitist path is suffi-
cient enough to diverge from the present issue space therefore preventing any
possibility of stagnation.
Rank-based ant system: Ranked AS utilizes a form of a significant ant as seen in the
case of the elitist ant system. But in this case, they spread the pheromones on
various promising paths instead of providing one best path. Each path is ranked
according to its length, whereby the best-ranked one gets more pheromones and
the one that has been ranked the least obtains a few pheromones by the specialist
ants. In the ranked AS, in case there are less specialists than the usual ants, the
worst-ranked path shall not be assigned any pheromones.
Min-max ant system: In the minimum and maximum ant system, there exists no
specialist ant. It is utilizing the normal ants. In this framework, there is a cap
reflecting the max and min values of the pheromone, which can be placed on a
certain path. Since the algorithm places the cap on a minimum pheromone
amount, the path pheromone cannot be dropped as low to make the path so abso-
lute. In the same case, the maximum amount of pheromone on paths is fixed as
paths cannot get saturated in the process of overshadowing all the other paths.
Whenever one or more paths are closer to the min or max levels, the pheromone
levels are considered smooth. This therefore promotes the paths with low levels
of pheromone while maintaining the standings over the paths. In this methodol-
ogy, the pheromones are applicable on the most effective path while ignoring the
others. All these additions make the basic ant system a powerful and competitive
algorithm.

9.4 Particle Swarm Optimization Algorithm (PSO)

9.4.1 Introduction

In 1995, J. Kennedy and R. Eberhart based on their earlier investigations suggested


that the group members can benefit not only from their unidirectional memories but
also from the collective memory with a multidimensional group. Particle swarm
optimization (PSO) method is based on bird flocking. It is a metaheuristic optimiza-
178 R. B. Madhumala and H. Tiwari

tion algorithm. The major PSO algorithm operates based on a population or swarm
of candidate resolutions known as particles. All these particles revolve around the
issue search space as per the simple formula [16].
This is the concept behind the particle swarm optimization algorithm. We start
with randomly grouped particles on the points of the forms K = <k1, k2,…,kn > and
each is characterized with a randomized selected speed V = <v1, v2,…,vn>. Every
particle is composed of three elements such as inertia, memory, and group memo-
ries in Fig. 9.5.
The movements of the particles are guided by:
1. Their individually known positions in the searching spaces (pbest)
2. The complete swarm’s best-known position (gbest)
3. The velocity V which is generally capped to some maximum value Vmax to avoid
searching out of the search space
Basic pseudo-code for PSO algorithm is provided below:
Step 1: Initialization
For each particle i
Initialize the particle’s position with a uniform distribution and represent the
lower and upper bounds of the search space.
(a) Initialize pbest to its initial positions: pbest.
(b) Initialize gbest to the minimal value of the swarm: gbest.
(c) Initialize velocity: V.
Step 2: Repeat until a termination criterion is met.

Best Swarm Member


er-j
emb
of M
Inert

city

ive
ia

at
Velo

r
plo
Ex
New

Exploitive
r
ecto Best Location in the History
tyV
rsi of Member-j
ve
Di Swarm Member-j

Fig. 9.5 Particles move around the problem search


9 Analysis of Virtual Machine Placement and Optimization Using Swarm Intelligence… 179

For each particle i


Pick random numbers: r1, r2.
Update the particle’s velocity. See formula (2).
(a) Update the particle’s position. See formula (3).
(b) Update the best-known position of particle i: pbest.
Step 3: Output gbest(t) holds the best position found in a given problem search
space.
The initial segment is termed as inertia that signifies the previous speed of par-
ticles. The following part is referred to as the cognitive element that is determined
by every particle that signifies the other, including the third parameter that signifies
the collaborative influence of the particles to locate the novel resolutions [17].

9.4.2 Basic Variants of PSO

Jau et al. [18] projected a modified QPSO that is applicable to the high-breakdown
regression estimators and the minimally trimmed square methodology. Jamalipour
et al. recommended the QPSO based on various mutation operators used to optimize
the WWER-1000 core fuel processing. Tang et al. [19] projected a memetic algo-
rithm and the memory technique. The memetic algorithm gives some basic experi-
ence to the particles in their local space, and after gaining experience in the local
space, they are released to search the global space. Davoodi et al. [20] proposed a
new approach, by combining improved QPSO and simplex algorithms. Li and Xiao
[21] recommended the encoding technique that is centered on qubits described on
the Bloch sphere. Yumin and Li [22] combined the artificial fish swarm to the QPSO
and exploited adaptive constraints meant to be overlooked. Jia et al. [23] projected
a developed QPSO centered on GA to evaluate the asynchronous optimization of
the various sensor arrays and classifiers. Gholizadeh and Moghadas [24] focused
and developed QPSO metaheuristic algorithm purposed to initiate a performance-
based optimized designing process. The two number samples have been initiated to
represent the most effective methodology. The two numerical examples have been
included to represent the effectiveness of the proposed methodology. J. Kennedy
proposed bare bones particle swarm optimization (BBPSO) in the year 2003.
Modified the BBPSO algorithm by utilizing the ­crossover and mutation operator of
the DE algorithms. In the year 2015 projected a binary BBPSO by formulating a
reinforced memory technique to update all the local leader particles targeting to
evade degradation of pending genes in particles to solve optimal feature subset and
classification problems. Chaotic PSO (CPSO) was implemented by integrating the
chaotic theory with the PSO. Many combinations were done with the CPSO algo-
rithm like combining it with the K2 algorithm as done by Zhang et al. Pluhacek
et al. [25] used various chaotic frameworks as a pseudorandom figure generator for
the calculation of PSO algorithm. Juang et al. [26] projected the adaptive FPSO
180 R. B. Madhumala and H. Tiwari

(AFPSO) algorithm that uses fuzzy logic to enhance the model. PSO with time-
varying acceleration coefficients (TVAC) was anticipated to enhance the perfor-
mance of standard PSO by Cai et al.

9.5  omparison of the Particle Swarm Optimization


C
Algorithms

Author and
year Parameter used Limitations Advantages Conclusion
Singh and Execution time It does not consider QoS-aware
Chana [26] multiple levels of resource
QoS requirements scheduling in
cloud environment
Dashti and Makespan It is more efficient The modified PSO
Rahmani for scheduling algorithm
[27]
Lalwani CPU-based and Comparison Study of variations
et al. [28] GPU-based between different of PSO algorithm
strategies PSO approaches
Xu and Yu Contraction-­ It does not fully Markov properties The convergence
[29] expansion address the issue of SPSO are of SPSO is studied
coefficient, the related to analyzed using Martingale
wave function convergence rate and theory
running time
Selvaraj Turnaround time, It is efficient when VM selection
et al. [30] waiting time, and compared to using swarm
CPU utilization SPSO, GA, and intelligence
DPACO models approach

9.6 Applications of PSO

Particle swarm is more than just a collection of particles wherein these particles
have no power to solve any type of problem and progress occurs only when they
interact with each other. The diagram represents the different applications of PSO in
various domains as shown in Fig. 9.6.
PSO does not utilize the gradient of the issue but is centered on optimizing the
necessary optimization issues that are defined by the classical optimization method-
ologies. PSO can be utilized in the process of optimizing the irregular issues. The
US military is investigating swarm intelligence techniques for controlling unmanned
vehicles. NASA is also investigating the use of PSO technology for planetary map-
9 Analysis of Virtual Machine Placement and Optimization Using Swarm Intelligence… 181

Load Balancing
Task scheduling
Workflow Scheduling
Cost Optimization
Others

Fig. 9.6 Applications of PSO in cloud computing

ping. PSO is also applied in computer graphics and visualization. It is also used in
designing and optimizing engines as well as electrical motors.

9.7 Summary

This chapter surveys different nature-inspired algorithms and their applications in


different domains. We have also surveyed different algorithms used in optimal
placement and selection of virtual machines in the cloud-centered ecological sys-
tems. All the algorithms discussed have considered various parameters like memory
optimization, bandwidth optimization, computation time and cost, etc. As noted
each algorithm has its limitations, and we need to develop a hybrid solution for
optimizing the placement of virtual machines in the cloud.

References

1. Abdelsamea, A., et al.(2017, November). Virtual machine consolidation enhancement using


hybrid regression algorithms. Egyptian Informatics Journal, ,18(3), 161–170.
2. Guo, Z., et al. (2017). Heuristic algorithms for energy and performance dynamic optimiza-
tion in cloud computing. Computing and Informatics, 36, 1335–1360. https://doi.org/10.4149/
cai201761335.
3. Dhanda, M., & Sangwan, O. P (2019). Quality of Service in Dynamic Resource Provisioning:
Literature Review. In: Minz S., Karmakar S., Kharb L. (eds) Information, Communication
and Computing Technology. ICICCT 2018. Communications in Computer and Information
Science, vol 835. Springer, Singapore.
4. Kumar, M. S., Sindhuja, P., & Ramamoorthy, P. (2018). A brief survey on nature inspired
algorithms: Clever algorithms for optimization. Asian Journal of Computer Science and
Technology, 7, 27–32.
5. Binitha, S., & Sathya, S. S. (2012). A survey of bio inspired optimization algorithms.
International Journal of Soft Computing and Engineering, 2(2), 137–151.
6. Theja, P. R., & Babu, S. K. (2014). Resource optimization for dynamic cloud comput-
ing environment: A survey. International Journal of Applied Engineering Research, 9(24),
26029–26042.
182 R. B. Madhumala and H. Tiwari

7. Suseela, B. B. J., & Jeyakrishnan, V. (2014). A multi-objective hybrid ACO-PSO optimiza-


tion algorithm for virtual machine placement in cloud computing. International Journal of
Research in Engineering and Technology, 3(4), 474–476.
8. Usmani, Z. & S. Singh (2016). “A survey of virtual machine placement techniques in a cloud
data center.” Procedia Computer Science 78: 491–498.
9. Choudhary, A., Govil, M. C., Singh, G., Awasthi, L. K., Pilli, E. S., & Kapil, D. (2017). A criti-
cal survey of live virtual machine migration techniques. Journal of Cloud Computing, 6(1), 23.
10. Rani, P. and A. K. Bhardwaj (2017). “A Review: Metaheuristic Technique in Cloud Computing.”
International Research Journal of Engineering and Technology (IRJET) Volume 4.
11. St́ephanie, C., Paraiso, F., Merle, P. (2017). A Study of Virtual Machine Placement
Optimization in Data Centers. 7th International Conference on Cloud Computing and Services
Science.
12. Son, J., & Buyya, R. (2019). Priority-aware VM allocation and network bandwidth provi-
sioning in Software-Defined Networking (SDN)-enabled clouds. IEEE Transactions on
Sustainable Computing, 4(1), 17–28.
13. Blum, C. (2005). Ant colony optimization: Introduction and recent trends. Physics of Life
Reviews, 2(4), 353–373.
14. Gao, Y., Guan, H., Qi, Z., Hou, Y., &Liu, L. (2013, December). A multi-objective ant colony
system algorithm for virtual machine placement in cloud computing. Journal of Computer and
System Sciences,79(8), 1230–1242.
15. Dorigo, M., & Gambardella, L. M. (1997). Ant colony system: A cooperative learning approach
to the traveling salesman problem. IEEE Transactions on Evolutionary Computation, 1(1),
53–66.
16. Bin, W., Qinke, P., Jing, Z., & Xiao, C. (2012, June 1). A binary particle swarm optimiza-
tion algorithm inspired by multi-level organizational learning behavior. European Journal of
Operational Research, 219(2), 224–233.
17. Zhang, S., et al. (2017). Numerical study on attenuation of bubble pulse through tuning the air-­
gun array with the particle swarm optimization method. Applied Ocean Research, 66, 13–22.
18. Jau, Y.-M., Su, K.-L., Wu, C.-J., & Jeng, J.-T. (2013). Modified quantum-behaved particle
swarm optimization for parameters estimation of generalized nonlinear multi-regressions
model based on Choquet integral with outliers. Applied Mathematics and Computation, 221,
282–295. https://doi.org/10.1016/j.amc.2013.06.050.
19. Tang, D., Cai, Y., Zhao, J., & Xue, Y. (2014). A quantum-behaved particle swarm optimiza-
tion with memetic algorithm and memory for continuous non-linear large scale problems.
Information Sciences, 289, 162–189. https://doi.org/10.1016/j.ins.2014.08.030.
20. Davoodi, E., TarafdarHagh, M., & GhassemZadeh, S. (2014). A hybrid Improved Quantum-­
behaved Particle Swarm Optimization–Simplex method (IQPSOS) to solve power sys-
tem load flow problems. Applied Soft Computing, 21, 171–179. https://doi.org/10.1016/j.
asoc.2014.03.004.
21. Li, P. (2008). Quantum genetic algorithm based on Bloch coordinates of qubits and its applica-
tion. Control Theory & Applications, 25(6), 985–989.
22. Yumin, D., & Li, Z. (2014). Quantum behaved particle swarm optimization algorithm based
on artificial fish swarm. Mathematical Problems in Engineering, 2014, 1–10. https://doi.
org/10.1155/2014/592682.
23. Jia, P., Tian, F., Fan, S., He, Q., Feng, J., & Yang, S. (2014). A novel sensor array and classi-
fier optimization method of the electronic nose based on enhanced quantum-behaved particle
swarm optimization. Sensor Review, 34(3), 304–311. https://doi.org/10.1108/SR-02-2013-630.
24. Gholizadeh, S., & Moghadas, R. (2014). Performance-based optimum design of steel frames
by an improved quantum particle swarm optimization. Advances in Structural Engineering,
17, 143–156. https://doi.org/10.1260/1369-4332.17.2.143.
25. Pluhacek, M., Senkerik, R., & Zelinka, I. (2014). Particle swarm optimization algorithm
is driven by multichaotic number generator. Soft Computing, 18, 631–639. https://doi.
org/10.1007/s00500-014-1222-z.
9 Analysis of Virtual Machine Placement and Optimization Using Swarm Intelligence… 183

26. Singh, S. and I. Chana (2015). “QRSF: QoS-aware resource scheduling framework in cloud
computing.” The Journal of Supercomputing 71(1), 241–292.
27. Dashti, S. E., & Rahmani, A. M. (2016). Dynamic VMs placement for energy efficiency by
PSO in cloud computing. Journal of Experimental & Theoretical Artificial Intelligence, 28(1–
2), 97–112.
28. Lalwani, S., H. Sharma, S. C. Satapathy, K. Deep and J. C. Bansal (2019). “A Survey on Parallel
Particle Swarm Optimization Algorithms.” Arabian Journal for Science and Engineering
44(4), 2899–2923.
29. Xu, G. and G. Yu (2018). “Reprint of: On convergence analysis of particle swarm optimization
algorithm.” Journal of Computational and Applied Mathematics 340, 709–717.
30. Selvaraj, A., R. Patan, A. H. Gandomi, G. G. Deverajan and M. Pushparaj (2019). “Optimal
virtual machine selection for anomaly detection using a swarm intelligence approach.” Applied
Soft Computing 84, 105686.
Chapter 10
Performance Evaluation of Different
Neural Network Classifiers for Sanskrit
Character Recognition

R. Dinesh Kumar, C. Sridhathan, and M. Senthil Kumar

10.1 Introduction

Classifying the input characters according to the predefined character data set is a
challenging task. Many researchers using the image processing techniques have
contributed their work on English character recognition. There is an increasing
interest for computer applications which ease the human jobs and solve complex
problems. Optical character recognition is used for classifying the alphanumeric
characters which are in the digital image format. Classical methods for handwritten
character recognition have drawbacks. First, same characters will differ in style,
shape, and size from person to person. Second, the characters will vary from time to
time for the same person. Third, visual characters are affected with noise near the
edges while reading through OCR techniques. Optical character recognition system
has a variety of commercial and practical applications such as in banks, postal ser-
vice, license plate recognition system, smart card processing system, and so
on [1–3].
The third most commonly utilized language after English and Chinese is Hindi.
There are around 500 billion people all over the world who write and speak in
Hindi. Hindi is the fundamental script of many Indian languages that originated
from Sanskrit. Sanskrit being an ancient language is no longer spoken. It is a very
expressive language, which has been enriched and influenced by Farsi, Turkish,
Dravidian, Portuguese, English, and Arabic.
Printed Sanskrit characters are effortlessly identified by computer machines.
But, handwritten Sanskrit characters are not being identified accurately and

R. D. Kumar (*)
Siddhartha Institute of Technology & Science, Hyderabad, Telangana, India
C. Sridhathan · M. S. Kumar
Nalla Malla Reddy Engineering College, Hyderabad, Telangana, India

© Springer Nature Switzerland AG 2020 185


A. Haldorai et al. (eds.), Business Intelligence for Enterprise Internet of Things,
EAI/Springer Innovations in Communication and Computing,
https://doi.org/10.1007/978-3-030-44407-5_10
186 R. D. Kumar et al.

efficiently by the computer machine. Numerous researchers have proposed different


kinds of methods and algorithms for recognizing the Sanskrit characters. Several
software are used for optical Sanskrit character recognition. However, for recogniz-
ing the handwritten Sanskrit characters, several procedures are to be adopted; no
single machine or single process can perform Sanskrit character recognition.
Artificial neural network (ANN) can be used for character recognition because of its
versatility and simplicity of working design.

10.2 Literature Review

The functioning of the human brain has inspired many scientists to come up with
computing systems termed as neural networks. Some researchers applied optical
character recognition using feed-forward neural networks, backpropagation algo-
rithm model, and other techniques for pattern recognition [1, 3, 4].
Recognition of handwritten characters is the most mesmerizing research area in
the field of image processing and pattern recognition. In recent years, character
recognition has been a popular research area for many researchers. This is because
of its potential application in several fields like number plate recognition, sorting
letter based on postal codes, etc. The first research on Devanagari characters was
published in 1977 [5]. At present researchers are working on handwritten character
recognition of many languages such as English, Chinese, Hindi, Tamil, Bangla, and
Telugu. Few researchers are focusing on neural network and artificial intelligence
technique for reduced processing time with higher recognition accuracy [1–3, 6–9].
Character recognition is mainly classified into online and offline which are further
classified into clustering template matching, etc. as shown in Fig. 10.1 [8]. Existing

Character
Recognition

Online Off line

ANN KNN

Clustering

Template Matching Direction Based

Feature Extraction

Fig. 10.1 Classification of character recognition system


10 Performance Evaluation of Different Neural Network Classifiers for Sanskrit… 187

methods for handwriting identifications are template matching, statistical and struc-
tural techniques, fuzzy logic, support vector machines, and neural networks (NNs).
Recognition of handwritten character involves (1) image acquisition, (2) image
pre-processing, (3) image segmentation, (4) feature extraction, and (5) classifica-
tion. Classification stage is the decision-making part of image processing systems.
Feature extraction and classification play a significant role in handwritten character
recognition [1, 2, 6].

10.3 Character Recognition Process

Character recognition process for offline handwritten patterns consists of the fol-
lowing stages as shown in Fig. 10.2.

10.3.1 Image Acquisition

The handwritten Sanskrit characters are scanned as an input as shown in Fig. 10.3.

10.3.2 Pre-processing

Scanned handwritten Sanskrit characters are transferred as an image. Pre-


processing involves binarization, denoising, thinning, skew detection, and correc-
tion. In binarization, color images are converted into the gray-scale image with the
help of the balanced histogram thresholding technique. Based on the histogram

Input Document Pre-processing Segmentation Feature Extraction

Training Data Image Matrix Mapping

Recognised Document Training Network Classification

Fig. 10.2 Stages of Sanskrit handwritten character recognition system

Fig. 10.3 Scanned Sanskrit handwritten character


188 R. D. Kumar et al.

Fig. 10.4 Pre-processed Sanskrit handwritten character

Fig. 10.5 Character segmentation

weights, this method divides the images into two: foreground and background.
Denoising is performed using non-local median filters for eliminating and sup-
pressing the noise level. Zhang-Suen thinning algorithm has been utilized to pro-
duce another binary image with a thickness of one pixel for better edge detection.
This minimizes the noises without losing the quality of the image. The final step in
pre-processing is skew detection where the orientation of the character is adjusted
with respect to the true horizontal axis. The angle of orientation of the character is
rotated such that it is ±15 degrees. Pre-processed Sanskrit handwritten character is
shown in Fig. 10.4.

10.3.3 Segmentation

In segmentation, the required region of the image is separated. Here, improved


multi-scale segmentation approach is used to divide the characters into the
­horizontal and vertical projections with the help of the fragmentation process.
After the fragmentation process, the texts are split into the paragraphs and then
lines, words, and characters with the help of the histogram approach. Line, word,
and character segmentation is carried out and segmented character images are
shown in Fig. 10.5.
10 Performance Evaluation of Different Neural Network Classifiers for Sanskrit… 189

10.3.4 Feature Extraction

The statistical and structural procedure can be used for extracting features such as
height, width, horizontal lines, vertical lines, circles, slope lines, and arcs. Here the
height and width of the characters are estimated with the glyph and related bound-
ary values. These detected line values are decomposed by applying the wavelet
function described in Eq. 10.1.
−∞
F ( a,b ) = ∫ f ( x )ψ ∗ ( a,b )( x ) dx (10.1)

where ∗ is the complex conjugated and ψ is the Daubechies wavelet function.


In addition mean, variance, standard deviation, entropy, and energy are also cal-
culated and their corresponding formulae are given in Eqs. 10.2, 10.3, and 10.4.
i =0
Mean = ∑ i. px + y ( i ) (10.2)
2 ( n −1)

i =0 j =0
Variance = ∑ ∑ ( i − µ ) . p ( i,j )
2
(10.3)
n −1 n −1

i, j =0
Entropy = ∑ − ln ( pij ) pij (10.4)
n −1

10.4 Classification

Feature extracted characters are used in the classification stage for decision-making.
The classification methods can be statistical methods, artificial neural networks
(ANNs), kernel methods, and multiple classifier combination. In this work RCS
with BPNN, BPNN with RBF, and multilayer perceptron model are used for
classification.

10.4.1 RCS with BPNN

Random candidate selection (RCS) with backpropagation neural network (BPNN)


is one of the efficient supervised learning models which maps each input to the
particular output. The main goal of the RCS with BPNN is to train the network in
190 R. D. Kumar et al.

an efficient manner that can learn the appropriate internal representations, which
allows the arbitrary mapping of the input to the output with minimum error rate.
Steps involved in the proposed algorithm are as follows:
Step 1: Initialize all weights.
Step 2: Input vector consists of Fourier descriptors and border transition values
which are used as classifiers.
Step 3: By using non-linear sigmoid function, the weights are adjusted to obtain the
outputs yj.
Step 4: The weights are adjusted such that all the training inputs are used for weight
stabilization.
The structure of the network and the training algorithm used in RCS improve the
recognition of Sanskrit character and reduce the error rate based on the learning rate.

10.4.2 BPNN with RBF Network

The backpropagation neural network with radial basis function is one of the better
classification algorithms with training process to get the desired output. RBFN con-
sists of input layer, hidden layer, and output layer. Each layer produces the linear
output even for non-linear inputs. In radial basis function neural network, extracted
features like character height, weight, direction, mean, standard deviation, and vari-
ance are fed as inputs and transferred to the hidden layer that uses the radial basis
(radbas) as the activation function. K-mean clustering algorithm is used to obtain
the center value as given in Eq. 10.5.

i =1 2
argmin = ∑ ∑ X − µi (10.5)
k X∈Si

In the recognition phase, the characters are identified based on the shape and the
distance between the centers. Finally, the features of the handwritten character are
stored as a template in the database and compared with the new character as given
in Eq. 10.6.

j =0
( ) ( )
yk x p == ∑ Wkjϕ j x p
M
(10.6)

10.4.3 Multilayer Perceptron Neural Network

Multilayer perceptron neural network’s working principle is relatively based on the


human brain. Normally a human brain stores the information as a pattern and gains
the knowledge to solve the complex problems by experience. The multilayer
10 Performance Evaluation of Different Neural Network Classifiers for Sanskrit… 191

perceptron neural network recognizes the patterns by supervised training algorithm


with feed-forward from input to output layers. Activation function is computed as
given in Eq. 10.7.

 j =1…N 
m

xi m +1 ( n ) = f  ∑ Wij m x j ( n )  (10.7)
 M

Based on the error rate, the weightage factor is modified as given in Eq. 10.8.
t =1
new.w ji m = wij m −1 + γ ∑ δ i m ( n ) x j m −1 ( n ) (10.8)
T

10.5 Performance Analysis

The performance of RCS with BPNN, BPNN with RBF, and multilayer perceptron
neural network models is evaluated based on the following metrics:

10.5.1 Accuracy

Accuracy is a statistical measure which is used to analyze how well the classifier
recognizes the Sanskrit character with optimized way in Eq. 10.9.

No of True Positive + No of True Negative)


Accuracy = (10.9)
Total No of Samples
Accuracy can also be expressed in terms of sensitivity and specificity as given in
Eq. 10.10.

Accuracy = ( sensitivity )( prevalence ) + ( specificity ) (1 − prevalence ) (10.10)

10.5.2 Sensitivity

Sensitivity is a measure of how the proposed system correctly classifies the hand-
written characters with efficient manner. The sensitivity is measured using Eq. 10.11.

True Positive
Sensitivity = (10.11)
True Positive + False Negative
192 R. D. Kumar et al.

10.5.3 Specificity

Specificity measures how the proposed system correctly identifies the negative clas-
sifiers during the character recognition process. It is expressed as shown in Eq. 10.12.

True Negative
Specificity = (10.12)
True Negative + False Positive
The minimal error rate and precision recognition are used so that the efficiency
of the Sanskrit character recognition system is increased.

10.6 Results

The experimental results of the random candidate selection with backpropagation


neural networks, backpropagation neural networks with radial basis function net-
works, and multilayer perceptron neural network models are evaluated in terms of
error rate, sensitivity, specificity, and accuracy measures. The results of the pro-
posed classifiers are compared with the traditional classifiers such as SVM, SOM,
etc., and the comparison results are shown below in Tables 10.1, 10.2, and 10.3.

Table 10.1 Performance evaluation of proposed RCS with BPNN and existing methods
Different classifiers Sensitivity Specificity Accuracy
SVM 85 87 93
SOM 86.5 87.43 94.23
RBFN 88.23 91.23 94.34
Fuzzy NN 89.46 94.23 95.35
Proposed RCS with BPNN 92.13 95.63 96.45

Table 10.2 Performance evaluation of BPNN with RBF and existing methods
Different classifiers Sensitivity Specificity Accuracy
SVM 85 86 92
SOM 83.5 86.43 93.23
RBFN 87.23 90.23 94.34
Fuzzy NN 88.46 93.23 95.35
BPNN with RBF 93.13 96.63 97.45

Table 10.4 Efficiency comparison for proposed character recognition methods


Metrics RCS with BPNN BPNN with RBF MLP networks
Sensitivity 92.13 93.15 94.13
Specificity 95.63 96.63 96.73
Accuracy 96.45 97.45 98.46
10 Performance Evaluation of Different Neural Network Classifiers for Sanskrit… 193

Table 10.3 Performance evaluation of proposed MLP and existing methods


Different classifiers Sensitivity Specificity Accuracy
SVM 86 87 97
SOM 87.5 86.42 96.89
RBFN 89.23 95.23 97.54
Fuzzy NN 91.46 91.43 97.77
Proposed MLP 94.13 96.73 98.46

The experimental metrics results of the MLP network models were compared
with other models such as RCS with BPNN and BPNN with RBF; it had higher
overall efficiency. The comparison results are shown in Table 10.4 below. It can be
seen from Table 10.4 that the sensitivity, specificity, and accuracy levels of MLP
level are higher. The methods used for recognizing the handwritten Sanskrit charac-
ters produce minimum error rate and high accuracy when compared to other exist-
ing methods.

10.7 Summary

Handwritten Sanskrit character recognition was done using the following methods:
RCS with BPNN, BPNN with RBF, and MLP network. The MLP network gave an
effective and efficient result in both feature extraction and recognition. The accu-
racy level of MLP level was 98% and execution time required was very less. This
work can be extended in face recognition also.

References

1. Abbas, M. (2016). Offline handwriting recognition using neural networks. Thesis.


2. Aicha, E., Mohamed, K. K., & Hacene, B. (2015). Ontologies and bigram-based approach for
isolated non-word errors correction in OCR system. International Journal of Electrical and
Computer Engineering (IJECE), 5(6), 1458–1467.
3. Chirag, I. P., Ripal, P., & Palak, P. (2011). Handwritten character recognition using neural
network. International Journal of Scientific & Engineering Research, 2(5), 1–6.
4. Khushbu, S. M. (2013). Image pre-processing on character recognition using neural networks.
International Journal of Computer Applications, 82(13), 11–15.
5. Dimple, B., Gulshan, G., & Maitreyee, D. (2014). Design of an effective preprocessing
approach for offline handwritten images. International Journal of Computer Applications,
98(1), 17–23.
6. Shoba, R., Sanjay, K. V., & Anitta, J. (2016). A zone based approach for classification and
recognition of telugu handwritten characters. International Journal of Electrical and Computer
Engineering, 6(4), 1647–1653.
194 R. D. Kumar et al.

7. Shobha, R., Neethu, O. P., & Nila, P. (2015). Automatic vehicle tracking system based on fixed
thresholding and histogram based edge processing. International Journal of Electrical and
Computer Engineering, 5(4), 869–878.
8. Shruti, S. K., Hoshank, J. M., Sakshi, S. G., Sarang, S. S., & Chitre, D. K. (2017). Handwriting
recognition using neural network. International Journal of Engineering Development and
Research, 5(4), 1179–1181.
9. Sudarshan, S., & Seema, B. (2016). Handwritten character and word recognition using
their geometrical features through neural network. International Journal of Application or
Innovation in Engineering & Management, 5(7), 77–85.
Chapter 11
GA with Repeated Crossover
for Rectifying Optimization Problems

Mayank Jha and Sunita Singhal

11.1 Introduction

Optimization is at the core of many processes that solve problems in the real world.
Nonetheless, identifying the most optimal resolution for these issues is normally
tiresome, mostly in the availability of non-linear, high-dimensionality, and multiple
modalities. The evolving algorithm has indicated a more considerable success
recorded over the past few years to deal with the complex issues of optimization.
Whereas the EA family includes a number of various algorithms, the most promi-
nent and commonly used in practice is the genetic algorithm (GA). There has been
an increase in studies relating to the actual-parameter genetic algorithm (GA) in the
past decades irrespective of the presence of one similar actual-parameter evolution-
ary algorithm, such as evolutionary technique and the differential evolution. In con-
trast, in theory, the actual-parameter GA studies indicated the same theoretical
behavior in fitness landscapes with appropriate parameter tuning, as shown in the
previous study.
Crossover and mutation include the vital operation utilized in genetic algorithm.
The genetic algorithm (GA) is significantly utilized in dealing with the practical
optimization issues. GA has the capability of handling both the discrete and con-
tinuous variables, which is suited for the purpose of handling the parallel computing
and fitness environments that are extremely complex. Some researchers have desig-
nated that the genetic algorithms have a better adaptability to effectively deal with
the various models instead of ancient mathematics programming techniques. In this

M. Jha (*)
Citicorp Services India Limited, Pune, India
S. Singhal (*)
School of Computing and Information Technology, Manipal University Jaipur, Rajasthan,
India
e-mail: sunita.singhal@jaipur.manipal.edu

© Springer Nature Switzerland AG 2020 195


A. Haldorai et al. (eds.), Business Intelligence for Enterprise Internet of Things,
EAI/Springer Innovations in Communication and Computing,
https://doi.org/10.1007/978-3-030-44407-5_11
196 M. Jha and S. Singhal

chapter, the main aim is to enhance the status of task scheduling based on the
­comparison of the proposed genetic algorithm with the present techniques to obtain
the accuracy, fitness value, and runtime and know the turnaround time of the task
allocated by various task schedulers like FCFS, RR, etc.

11.2 Proposed System

In the proposed system, we have purposed a genetic algorithm to calculate and mea-
sure the fitness value from the assigned tasks in the scheduler. The tasks schedulers
could be in the form of FCFS, RR, etc. By using GA we will come to know the mini-
mum turnaround time or execution time required for processing the tasks as shown
in Fig. 11.1.

11.2.1 General Arrangement of a Genetic Algorithm

Generally, GA includes five elements:

Population P(t) Offspring C(t)


start chromosome chromosome

encoding 1100101010 crossover


Initial 1100101010
1011101110
solutions t¬0 1011101110 CC (t)
▪ 1100101110

t¬t+1 ▪ 1011101010

0011011001 mutation 0011011001


1100110001
0011001001
decoding
N evaluation CM (t)
terminating Solution candidates decoding
Y condition? fitness computation

selection P(t)+C(t)

stop New Population


best solution
roulette wheel

Fig. 11.1 The general structure of a genetic algorithm


11 GA with Repeated Crossover for Rectifying Optimization Problems 197

1. A genetic account of probable resolutions to the issue


2. A technique of formulating a population (a preliminary assortment of possible
solutions)
3. Testing functions in terms of fitness ranking solutions
4. Genetic operators altering offspring’s genetic composition (crossover, mutation,
selection, etc.)
5. The constraint values utilized by the genetic algorithm (population size, genetic
operator probabilities, etc.)

11.2.2 Algorithm Techniques

The proposed algorithm technique for task scheduling is genetic algorithm as


given below:
Algorithm: Genetic Algorithm

1 START
2 Generate the initial population
3 Compute fitness
4 REPEAT
5 Selection
6 Crossover
7 Mutation
8 Compute fitness
9 UNTIL secluding has converged
10 STOP

The given algorithm shows the execution process of the tasks allocated to the
CPU. The GA includes primarily the initialization, discovery, crossover, and muta-
tion phases. Initially all the tasks are given for processing. A set of parameters
(variables) known as genes describe a person. The fitness function determines how
competitive a person is (an individual’s ability to compete with other people). The
probability that an individual will be selected for processing is based on its fitness
score. The whole process is repeated until the fit individuals are not obtained. Next
step is for fitness score. The fitness element refers to the fitness of someone’s state
(an individual’s ability to compete with other folks). It gives a fitness score to each
individual. Next step is selection. Fitness-related selection, also known as selec-
tion of roulette wheels, is a genetic operator used in genetic algorithms to select
potentially useful recombination solutions. In fitness-related selection, as in all
selection strategies, the fitness function assigns fitness to possible solutions. A
crossover is a simple operation wherein a pair of strings randomly swaps their
substrings with each other. This operator’s role in GA is more critical. It imple-
ments the principle of evolution. Next step is mutation. An operation of mutation
198 M. Jha and S. Singhal

is formulated to diminish the processors’ idle time that is spent waiting for the
other processors’ information. Again the fitness of the process is computed for the
selection procedure until the scheduling process is converged.

11.3 Literature Review

Shigeyoshi Tsutsui et al. [1] projected simplex crossover (SPX), a new multi-par-
ent recombination operator for real-coded GAs. The SPX is considered as a simpli-
fied crossover operator, which utilizes the search space property of the simplex.
The SPX is characterized with a balance between the development and discovery
and is distinct in the production of offspring from coordinate systems. The results
of the experiment showed high performance with multimodality and/or epistasis
on test functions.
Isao Ono et al. [2] proposed a novel actual-coded genetic algorithm that utilizes
the unimodal normal distribution crossover (UNDX) improved by the uniform
crossover (UX). The UNDX is characterized by an advantage that is not found in
other crossover operators. As such, this is effective for the process of optimizing the
functions in operators with firm epistasis. UNDX is assured with the most crossover
operators; hence, it is vital for optimizing various functions in the parameters with
firm epistasis.
Zhu Can et al. [3] applied the genetic algorithms with actual codes to resolve
issues meant to find the novel individual promised by the crossover operators of the
seeds in various localized minimal points disrupting schemata of parents. This was
responsible in acting as mutation operators. With reference to this, the idea is sepa-
rated from the optimal sub-population and proposed by the population division cen-
tered on the Euclidean distances between the present optimal individuals. Kalyanmoy
Deb et al. [4] recommended the standardized parent-centered recombination opera-
tors and the steady-state, scalable, elite-conserving, and rapid population shift mod-
els. The performance of the G3 system, including the PCX operator, is evaluated on
three extensively utilized test issues as contrasted with various classical and evolu-
tionary optimization algorithms, which include other actual-parameter GAs com-
posed of uniform modal normal distribution crossover (UNDX) and simplex
crossover (SPX) operators and the related self-adaptive evolutionary algorithm. The
technique of covariance matrix adaptation evolution (CMA-ES), the methodology
of differential evolution, and the quasi-Newton method are some of the initiatives.
It was noticed in the analysis that the recommended algorithm performed an effec-
tive, reliable, and consistent approach. Saber M. Elsayed et al. [5] presented an
objective meant to enhance the evaluation of the genetic algorithm through the pro-
cess of initiating novel crossover with random operator targeted at replacing muta-
tion. The projected crossover uses three parents vital for generating three new
11 GA with Repeated Crossover for Rectifying Optimization Problems 199

offspring relevant for exploitation, whereas a third offspring is purposed to enhance


exploration. The randomized operator is helpful to facilitate an escape and prema-
ture convergence.
Can et al. [6] analyzed the unfair use of fair selection laws to overcome
genetic algorithm restricted optimization. A form of population framework with
the ideology of Pareto dominance, including the sequence of user factory
designs. As a result, this balances the feasible limits and regions along an opti-
mal search path of the resolution that makes the genetic algorithm of constrained
optimization on all sides of the region feasible to consider the optimal resolu-
tion. There is a relation between the feasible part in demes and the revolutionary
generation. It considers both the algorithmic search quality and efficiency opti-
mization. The example of computational research and engineering has shown
that the improved genetic algorithm of constrained optimization includes a sim-
plified algorithm structure with a significant quality of resolutions. Tetsuyuki
Takahama et al. [7] recommended a constrained DE with the archive and gradi-
ent-based mutation (εDEag) to enhance the stability, usability, and efficiency of
the εDEg.
K. Sunitha et al. [8] designed an algorithm to schedule the DAG tasks on hetero-
geneous processors in such a way that minimizes the total completion time (makes-
pan). Makespan is a measure of the throughput of the heterogeneous computing
system (execution time + waiting time or idle time). Comparing makespan for dif-
ferent number of processors, number of tasks, population size, and number of gen-
erations to be done. Yan Kang et al. [9] have shown the use of GAs in the process of
dealing with task scheduling issues through various means. The two vital techniques
appear to be the methodologies that apply the GA in the linking of other list sched-
uling methods and techniques that apply the GA to produce the real assignment of
transforming tasks to processors.

11.3.1 Expected Results

This section contrasts simulation results of our proposed algorithm with the results
of several other algorithms. Table 11.1 lists the parameters of our proposed algo-
rithm and the other algorithms used in the performance analysis. Table 11.2 indi-
cates the detailed specifications.
By studying various techniques in literature survey, we will improve the methods
and techniques in Figs. 11.2, 11.3, and 11.4. We also applied our algorithm to a task
scheduling problem [10]. Task scheduling’s primary goal is to schedule tasks on
processors and reduce the make-up of the program, i.e., the accomplishment period
of the last task compared to the start time of the first task. The output of the problem
is the assignment of tasks to processors.
200 M. Jha and S. Singhal

Table 11.1 Comparison of all algorithms along with their parameters and values
Algorithm Parameter Value
GGA P-CROSSOVER 0.8
P-MUTATION 0.8
NUMBER OF GENERATION VARIABLE
SCGA P-CROSSOVER 0.8
P-MUTATION 0.8
NUMBER OF GENERATION VARIABLE
SSGA P-CROSSOVER 0.8
P-MUTATION 0.8
NUMBER OF GENERATION VARIABLE
GA-RA P-CROSSOVER 0.8
P-MUTATION 0.8
NUMBER OF GENERATION VARIABLE
CLOUDSIM MIPS(VM) 20(incremented by 10)
CLOUDLET LENGTH 4000(incremented by 50)
SCHEDULING POLICY VARIED

Table 11.2 Algorithm’s runtime and objective value for OneMax function
OneMax
gGA GA-RA
Iterations Runtime Value Runtime Value
5000 158 315 332 512
10000 372 332 550 512
15000 563 350 808 512

3.85
3.8
3.75
3.7
SpeedUP

3.65
3.6
3.55
3.5 SpeedUP
3.45
3.4
3.35
0 10 20 30 40 50 60
Number of Tasks

Fig. 11.2 Speedup vs. number of tasks


11 GA with Repeated Crossover for Rectifying Optimization Problems 201

1.2

0.8
Speedup
0.6 RR

0.4 GA

0.2 TS-GA

0
25 50 75 100
number of Task

Fig. 11.3 The comparison speedup of three algorithms RR, GA, and TS-GA

1.2
1
0.8
Efficiency

RR
0.6
0.4 GA

0.2 TS-GA

0
25 50 75 100
number of Task

Fig. 11.4 The comparison efficiency of three algorithms RR, GA, and TS-GA

11.4 Summary

GA is a heuristic remedy search method motivated by the physical evolution. It


belongs to the class of flexible and robust approaches. GA can be used in a wide
range of optimization and learning issues. This is one of the major reasons for its
enormous popularity. It is certainly suited to issues whereby traditional optimiza-
tion systems break down due to irregular structure of search spaces or due to
searches becoming computationally intractable. In this research, we have proposed
a genetic algorithm for solving optimization problems. We also tested the algorithm
on task scheduling problem and found it efficient in reducing the makespan time.
The experimental examination presented that the algorithm congregates quicker
than its counterparts to the optimal solution. Also, the results produced were effec-
tive based on the projected value, thus exhibiting a superior performance.
202 M. Jha and S. Singhal

References

1. Shigeyoshi Tsutsui, Masayuki Yamamura, & Higuchi, T.. (1999). Multi-parent recombination
with simplex crossover in real coded genetic algorithms. In IEEE (pp 657–664).
2. Ono, I., Kita, H., & Kobayashi, S. (1999). A Robust real-coded genetic algorithm using
unimodal normal distribution crossover augmented by uniform crossover: Effects of self-­
adaptation of crossover probabilities. IEEE, 1, 496–503.
3. Zhu Can, & Liang Xi-Ming. (2009). Improved genetic algorithms to solving constrained opti-
mization problems, international conference on computational intelligence and natural com-
puting. In IEEE (pp 486–489).
4. Deb, K., Anand, A., & Joshi, D. (2002, December). A computationally efficient evolutionary
algorithm for real-parameter optimization, evolutionary computation. IEEE, 10(4), 371–395.
5. Elsayed, S. M., Sarker, R. A., & Essam, D. L. (2011) GA with a new multi-parent crossover
for constrained optimization. In 2011 IEEE Congress of Evolutionary Computation (CEC)
(pp. 857–864). New Orleans.
6. Can, Z., Xi-Ming, L., & Shu-renhu, Z. (2009). Improved genetic algorithms to solving con-
strained optimization problems. In 2009 International conference on computational intelli-
gence and natural computing (pp. 486–489). IEEE.
7. Takahama, T., & Sakai, S. (2010). Constrained optimization by the ε constrained differen-
tial evolution with an archive and gradient-based mutation. In Congress on Evolutionary
Computation (pp. 1–9). Barcelona: IEEE.
8. Sunita, K., & Sudha, P. V. (2013). An efficient task scheduling in distributed computing
systems by improved genetic algorithm. International Journal of Communication Network
Security, IEEE, 2, 24–30.
9. Kang, Y., & Zhang, Z. (2011). An activity-based genetic algorithm approach to multiproces-
sor scheduling. In 2011 Seventh international conference on natural computation (pp. 1048–
1052). Shanghai: IEEE.
10. Anandakumar, H., & Umamaheswari, K. (2017). Supervised machine learning techniques in
cognitive radio networks during cooperative spectrum handovers. Cluster Computing, 20(2),
1505–1515.
Chapter 12
An Algorithmic Approach to System
Identification in the Delta Domain Using
FAdFPA Algorithm

Souvik Ganguli , Gagandeep Kaur, Prasanta Sarkar, and S. Suman Rajest

12.1 Introduction

System identification relates to the field of modeling dynamic systems from experi-
mental data [1]. This process requires a huge amount of information regarding its
input/output which may not always be available. In the classical methods, a model
structure is first selected, and the unknown model parameters are identified by mini-
mizing an objective function. However, conventional techniques are typically based
on gradient descent techniques [2].
Metaheuristic algorithms and their hybridizations have already made inroads in
identification and control literature [3, 4]. Wiener and Hammerstein systems are two
commonly used prototypes used for the identification of systems [5]. Parameter
assessment of these systems is conducted in the discrete time based on the applica-
tion of the time domain shift operator and the complex domain z-transform using
various soft computing techniques [6–10]. A large volume of works also exists in
the continuous-time system using classical techniques [2]. Hammerstein and Wiener
model identification in a continuous domain is based on the application of the meta-
heuristic method approaches which are rarely being investigated.
There are several methods built on discrete time systems using the ability of
computers in system identification and control. Almost contemporaneously, there

S. Ganguli (*) · G. Kaur


Thapar Institute of Engineering & Technology, Patiala, Punjab, India
e-mail: souvik.ganguli@thapar.edu; gagandeep@thapar.edu
P. Sarkar
National Institute of Technical Teachers’ Training & Research,
Bidhannagar, West Bengal, India
e-mail: psarkar@nitttrkol.ac.in
S. S. Rajest
Vels Institute of Science, Technology & Advanced Studies, Chennai, Tamil Nadu, India

© Springer Nature Switzerland AG 2020 203


A. Haldorai et al. (eds.), Business Intelligence for Enterprise Internet of Things,
EAI/Springer Innovations in Communication and Computing,
https://doi.org/10.1007/978-3-030-44407-5_12
204 S. Ganguli et al.

has been an analogous endeavor in evolving methods in systems theory because the
physical signals are progressively by default. Modeling, control, and identification
using the delta operator is continuous based on the application of the delta operator
as comprehensive methodology whereby the systems and signals are structured in
the discrete domain and leads to convergence to its corresponding continuous-time
signals and systems at a high sampling frequency therefore unifying both the con-
tinuous- and discrete-time systems and signals [11].
Although, in the discrete-time domain, Hammerstein and Wiener model identifi-
cation with metaheuristic approaches is absolutely familiar, similar analyses for
continuous-time systems are rarely reported. Therefore, system classification with
hybrid metaheuristic techniques can be considered for unification of discrete and
continuous systems using the delta operator’s properties. In order to estimate the
unknown Hammerstein and Wiener model parameters in the delta domain, a hybrid
algorithm, namely, FAdFPA, proposed by Ganguli et al. [12] was used. The remain-
der of the chapter is constituted as follows. Section 2 discusses the problem of
identification in the delta domain. Section 3 gives an overview of the FAdFPA algo-
rithm. Section 4 highlights the results, while Sect. 5 infers the paper.

12.2 Statement of the Problem

Hammerstein and Wiener models are two common models used in dynamic system
identification. Ganguli et al. [13] formulated the simulation of the delta operator to
estimate the unknown structure and parameters of polynomial nonlinearity using
GWOCFA algorithm. Interested readers may go through their modeling equations
that this chapter avoids. The authors of this chapter focused on developing a new
heuristic approach to parameter calculation by minimizing the error resulting from
the difference between real and expected values as described in Eq. 12.1.

1 N
∑  y ( k ) − yˆ ( k )
2
J= (12.1)
N k =1 

where y(k) and ŷ ( k ) represent the actual and expected model performance
responses. Therefore, in the next section, the proposed hybrid technique used to
identify parameters in the delta domain is deliberated.

12.2.1 FAdFPA Algorithm

Previously, the authors developed a hybrid algorithm called FAdFPA to solve


unconstrained problems of optimization [12]. By using FA as a global optimizer, the
balance for exploration was achieved, whereas FPA was used to perform ­exploitation.
To improve upon the local search abilities of FPA, its switch probability formulation
has been made adaptive by the formula in Eq. 12.2
12 An Algorithmic Approach to System Identification in the Delta Domain Using… 205

Step1: Ignite the system with PRBS sequence adulterated with white noise.
Step2: Yield random solutions for parameters of the linear and nonlinear parts in
the specified search domain.
Step3: Calculate the fitness function J for all possible solutions produced in Step 2.
Step4: Using FAdFPA algorithm find the best possible solution and its position.
Step5: Compare the best position with the previous best position.
Step6: Update position for the new fitness function.
Step7: Repeat Step 4 until the maximum iteration or best fitness value is produced.

Fig. 12.1 Steps for identification algorithm using FAdFPA

p = pmax − ( pmax − pmin ) × ( t / T ) (12.2)

Here, pmaxand pmin are two user-defined parameters, considered as 0.9 and 0.4,
respectively, in the experiments as reported in the literature. “T” denotes the maxi-
mum number of iterations while “t” is the present iteration. The steps for system
identification algorithms using the abovementioned hybrid technique are shown in
Fig. 12.1 listed below.
Step 1: Ignite the system with PRBS sequence adulterated with white noise.
Step 2: Yield random solutions for parameters of the linear and nonlinear parts in the
specified search domain.
Step 3: Calculate the fitness function J for all possible solutions produced in Step 2.
Step 4: Using FAdFPA algorithm, find the best possible solution and its position.
Step 5: Compare the best position with the previous best position.
Step 6: Update position for the new fitness function.
Step 7: Repeat Step 4 until the maximum iteration or best fitness value is produced.

12.3 Results and Discussions

The transfer function plant model [2] in the continuous time is represented as

b0 ρ + b1 −ρ + 5
G(ρ ) = = (12.3)
ρ 4 + a1 ρ 3 + a2 ρ 2 + a3 ρ + a4 ρ 4 + 23 ρ 3 + 185 ρ 2 + 800 ρ + 2500

The corresponding delta model at 100 Hz sampling frequency is given by

b0δ + b1 −0.8δ + 4.5


G (δ ) = = (12.4)
δ 4 + a1δ 3 + a2δ 2 + a3δ + a4 δ 4 + 22δ 3 + 180δ 2 + 760δ + 2200

Two separate continuous nonlinearities, viz.,


206 S. Ganguli et al.

y ( k ) = c1 x ( k ) + c2 x 2 ( k ) + c3 x 3 ( k ) = x ( k ) + 0.5 x 2 ( k ) + 0.25 x 3 ( k ) (12.5)

and

x (k ) x (k )
y (k ) = = (12.6)
c1 + c2 x ( k )
2
0.10 + 0.90 x 2 ( k )

are taken up, respectively, for identifying Wiener and Hammerstein model parame-
ters in the delta domain. For the PRBS input signal, a sample size of 255 was con-
sidered. The input signal has been polluted with an unbiased signal, viz., white
Gaussian noise of SNR 50 dB. For each of the experiments, a population segment
of about 20 and a maximum number of iterations of a hundred are considered. The
parent algorithms FA and FPA and the latest heuristics such as MFO, MVO, SCA,
and SSA have been compared. For all the algorithms listed, standard parameter
values are considered for comparison with the hybrid technique.
As metaheuristic algorithms are fundamentally stochastic, they have to be exe-
cuted multiple times to generate meaningful numerical measures. Hence 20 test
runs are executed for every algorithm to get significant statistical results. Wilcoxon
rank-sum test [14] is also performed to validate the findings. The identified and
actual parameters for the Hammerstein and Wiener frameworks in the delta domains
are indicated in Tables 12.1 and 12.2, respectively. The results are compared with
sufficient number of heuristic techniques reported in the literature. The closer
parameter estimates to the actual values are marked with the help of bold letters.
From Tables 12.1 and 12.2, it is quite fair to say that the hybrid technique proves
better than the several metaheuristic algorithms in the delta domain. Table 12.3 also
provides the statistical measurements of the Wiener and Hammerstein model param-
eters in the delta domain. The best results obtained in each column of the table are
highlighted with bold letters.
The hybrid method yields the minimum fitness function as compared to other
algorithms. Since standard deviation with the hybrid approach turns out to be the
least, it can be inferred that the algorithm is more robust than those considered for
comparison. Moreover, to verify the validity of the results obtained, few additional
statistical tests need to be done, verifying that the results did not happen all of a
sudden. The non-parametric Wilcoxon rank-sum analysis is therefore conducted to
verify the significance of the obtained results, and the measured p-values are cited
as relevant metrics. A few selected p-values are reported in Table 12.4.
Table 12.4 suggests that the results of the hybrid method are meaningful for all
other algorithms. The convergence curve is drawn in Fig. 12.2 for the test system
used for Wiener model identification in the delta domain. The y-axis of the plot
represents the normalized fitness function while the x-axis depicts the number of
iterations. Further, the parent algorithms FA and FPA as well as other popular algo-
Table 12.1 Comparison of Wiener system model parameters
Types of values Algorithms b0 b1 a1 a2 a3 a4 c1 c2 c3
Actual −0.8 4.5 22 180 760 2200 1 0.5 0.25
estimated FAdFPA −0.7993 4.4856 21.9989 179.8492 759.8219 2206.2981 1.0018 0.5101 0.2494
FA −0.7572 4.3832 21.7958 172.1841 763.0585 2258.0174 1.0442 0.4592 0.2620
FPA −0.8344 4.5947 22.0104 175.9721 766.0472 2370.6045 1.0576 0.5189 0.2565
MFO −0.7265 4.7379 23.1151 169.4181 734.5583 2251.8777 0.9507 0.4694 0.2538
MVO −0.7271 4.7374 23.1263 169.4236 734.5486 2251.898 0.9422 0.4492 0.3143
SCA −0.7753 4.2673 21.7239 164.1472 741.4551 2312.0279 0.9608 0.4675 0.2636
SSA −0.8429 4.4526 22.2516 172.5593 699.8514 2309.6354 1.0326 0.5385 0.2485
12 An Algorithmic Approach to System Identification in the Delta Domain Using…
207
208

Table 12.2 Comparison of Hammerstein system model parameters


Types of values Algorithms c1 c2 b0 b1 a1 a2 a3 a4
Actual 0.1 0.9 −0.8 4.5 22 180 760 2200
estimated FAdFPA 0.9993 0.9023 −0.7943 4.4928 22.4527 179.8753 759.6725 2196.7754
FA 0.1009 0.9099 −0.8623 4.7801 20.6720 188.4102 763.1495 2031.5233
FPA 0.0909 0.8992 −0.8045 4.4909 23.1408 164.1544 825.5786 2031.3154
MFO 0.0992 0.8659 −0.7829 4.5010 21.4638 188.7328 804.1881 2237.4300
MVO 0.0994 0.8758 −0.7927 4.5210 21.4936 188.7438 805.1892 2237.4538
SCA 0.0983 0.9848 −0.7929 4.6235 20.1478 165.2356 793.3553 2275.4090
SSA 0.1063 0.9206 −0.8153 4.4029 20.0600 166.9278 808.3730 2076.1786
S. Ganguli et al.
12 An Algorithmic Approach to System Identification in the Delta Domain Using… 209

Table 12.3 Statistical analysis of the fitness function


Test systems Test methods Minimum Maximum Mean Standard deviation
Wiener model FAdFPA 0.0016 0.0017 0.0016 3.6635e-05
FA 0.0018 0.0020 0.0019 5.8361e-05
FPA 0.0019 0.0021 0.0020 7.4325e-05
MFO 0.0018 0.0020 0.0019 5.8296e-05
MVO 0.0018 0.0021 0.0019 5.8356e-05
SCA 0.0018 0.0019 0.0019 6.7876e-05
SSA 0.0018 0.0020 0.0019 5.5030e-05
Hammerstein model FAdFPA 0.0183 0.0186 0.0184 4.9132e-05
FA 0.0190 0.0196 0.0193 5.9423e-05
FPA 0.0190 0.0196 0.0193 7.7482e-05
MFO 0.0191 0.0197 0.0193 5.9376e-05
MVO 0.0191 0.0198 0.0194 5.9772e-05
SCA 0.0191 0.0195 0.0193 7.8523e-05
SSA 0.0190 0.0195 0.0193 9.7813e-05

Table 12.4 Selected significant p-values using non-parametric test


Test systems Proposed method FA FPA MFO SCA
Wiener system FAdFPA 7.9026E-05 5.6379E-05 1.2893E-06 2.9131E-06
Hammerstein system FAdFPA 4.9017E-05 5.3492E-05 1.1300E-02 1.2000E-02

Convergence plot of wiener model with FAdFPA in the delta domain


1

0.9

0.8
Normalized function value

0.7
FAdFPA
0.6 FA
FPA
0.5 MFO
MVO
0.4
SCA
SSA
0.3

0.2

0.1

0
0 10 20 30 40 50 60 70 80 90 100
No. of iterations

Fig. 12.2 Convergence curve of Wiener system with FAdFPA


210 S. Ganguli et al.

rithms like MFO, MVO, SCA, and SSA are used for comparison with the pro-
posed method.
It is obvious from Fig. 12.2 that the planned process congregates at a faster rate
as compared to the other algorithms considered. In addition, the proposed method
provides better accuracy in terms of the fitness function value.

12.4 Summary

The hybrid topology assimilating FA and FPA is used to recognize Hammerstein


and Wiener model parameters in the delta domain by minimizing the fitness func-
tion. Delta operator modeling unifies continuous and discrete-delta domain results.
The hybrid technique’s fitness value not only outperforms that obtained by some
latest metaheuristic algorithms but also the heuristics that it constitutes for the test
systems considered. Also, the Wilcoxon test verifies the significance of the hybrid
approach results. Similar to other algorithms, the hybrid approach shows superior
convergence characteristic in the delta domain. Therefore, in the future, the algo-
rithm can also be used to estimate fractional-order systems in the delta domain.

References

1. CLjung, L. (1987). System identification: Theory for the user. Upper Saddle River:
Prentice-hall.
2. Sinha, N. K., & Rao, G. P. (2012). Identification of continuous-time systems: Methodology and
computer implementation (Vol. 7). Dordrecht: Springer Science & Business Media.
3. Bhushan, B., & Singh, M. (2011). Adaptive control of DC motor using bacterial foraging
algorithm. Applied Soft Computing, 11(8), 4913–4920.
4. Xu, X., Wang, F., &Qian, F.(2011). Study on method of nonlinear system identification. In
International Conference on Intelligent Computation Technology and Automation (ICICTA)
(Vol. 1, pp. 944–947).
5. Al-Duwaish, H. N. (2000). A genetic approach to the identification of linear dynamical sys-
tems with static nonlinearities. International Journal of Systems Science, 31(3), 307–313.
6. Pal, P. S., Kar, R., Mandal, D., & Ghoshal, S. P. (2016). Identification of NARMAX
Hammerstein models with performance assessment using brain storm optimization algorithm.
International Journal of Adaptive Control and Signal Processing, 30(7), 1043–1070.
7. Sun, J., & Liu, X. (2013). A novel APSO-aided maximum likelihood identification method for
Hammerstein systems. Nonlinear Dynamics, 73(1–2), 449–462.
8. Raja, M. A. Z., Shah, A. A., Mehmood, A., Chaudhary, N. I., & Aslam, M. S. (2016). Bio-­
inspired computational heuristics for parameter estimation of nonlinear Hammerstein con-
trolled autoregressive system. Neural Computing and Applications, 29(12), 1455–1474.
(2018).
9. Cuevas, E., Díaz, P., Avalos, O., Zaldívar, D., & Pérez-Cisneros, M. (2018). Nonlinear sys-
tem identification based on ANFIS-Hammerstein model using gravitational search algorithm.
Applied Intelligence, 48(1), 182–203.
12 An Algorithmic Approach to System Identification in the Delta Domain Using… 211

10. Garnier, H., Mensler, M., & Richard, A. (2003). Continuous-time model identification from
sampled data: Implementation issues and performance evaluation. International Journal of
Control, 76(13), 1337–1357.
11. Middleton, R. H., & Goodwin, G. C. (1990). Digital control and estimation: A unified approach
(prentice hall information and system sciences series). Prentice Hall: Englewood Cliffs, NJ.
12. Ganguli, S., Kaur, G., & Sarkar, P. (2016). A novel hybrid metaheuristic algorithm to solve
unconstrained optimization problems. International Journal of Advanced Computational
Engineering and Networking, 4(5), 40–43.
13. Ganguli, S., Kaur, G., & Sarkar, P. (2019). A hybrid intelligent technique for model order
reduction in the delta domain: A unified approach. Soft Computing, 23(13), 4801–4814.
14. Wilcoxon, F., Katti, S. K., & Wilcox, R. A. (1970). Critical values and probability levels for
the Wilcoxon rank sum test and the Wilcoxon signed rank test. Selected tables in mathematical
statistics, 1, 171–259.
Chapter 13
An IoT-Based Controller Realization
for PV System Monitoring and Control

Jyoti Gupta, Manish Kumar Singla, Parag Nijhawan, Souvik Ganguli ,


and S. Suman Rajest

13.1 Introduction

In the recent era, the rapid increase in the consumption of electricity and issues
regarding environmental concerns are the main reasons behind the development of
renewable energy sources (RESs). The electricity generation from RES has been
derived for its economic benefits and reliability [1]. Micro-grid includes distributed
generation resources such as photovoltaic, wind, diesel generators, etc. The imple-
mentation of micro-grid in the distribution network was introduced in the early
years of 2000 and encouraged by various agencies and utilities. Micro-grid has
increased the operation of power electronics converter in the power system for effi-
cient and effective power quality [2]. The micro-grid in a distribution system can be
installed near the substation or at the end of the feeder. Micro-grid has an essential
and useful feature, i.e., it can be operated in two different modes, islanded and grid-­
connected mode, to give essential support during the time of grid failure or mainte-
nance of grid system by supplying constant power. Due to an increase in the
interconnection of micro-grid in the distribution system, it raises some problems
that include fluctuation in the voltage, steady-state over-voltage, increases in the
system loss, and issues related to the voltage regulation devices and protection;
therefore, appropriate allocation of micro-grid is desirable.
There are various types of micro-grids, namely, DC micro-grid, AC micro-grid,
and hybrid micro-grid, which combines both AC and DC micro-grids [3]. Solar
energy systems are dependent on two factors: temperature and irradiance. The max-

J. Gupta · M. K. Singla · P. Nijhawan · S. Ganguli (*)


Thapar Institute of Engineering and Technology, Patiala, India
e-mail: jgupta_phd19@thapar.edu; souvik.ganguli@thapar.edu
S. S. Rajest
Vels Institute of Science, Technology & Advanced Studies, Chennai, India

© Springer Nature Switzerland AG 2020 213


A. Haldorai et al. (eds.), Business Intelligence for Enterprise Internet of Things,
EAI/Springer Innovations in Communication and Computing,
https://doi.org/10.1007/978-3-030-44407-5_13
214 J. Gupta et al.

imum power point (MPP) is the point where maximum power output is produced by
PV array. The control scheme is developed using a neural network. A neural net-
work (NN) is designed to provide a gate pulse signal to the inverter circuit to
improve the performance of the system. The artificial neural network (ANN) is
based on a machine learning approach and contains the number of artificial neurons
to perform the specific task in the system [4]. To investigate its impact in different
conditions, the inverter control scheme is used to improve the efficiency of the
power transfer between the PV system and the grid. An IEEE 13-node test feeder
system is considered as a grid system and is commonly used to test features of a
standard distribution network of a power system, operating at 4.16 kV.
The main problem that arises in monitoring the output power of the photovoltaic
system is the accuracy and time duration for the detection of the fault and the appro-
priate solution to it. The best approach for dealing with such kind of issues is the
Internet of Things (IoT) [5, 6]. It is the new concept which has emerged recently and
has gained a lot of attention in a few years. It can be generally explained as an infor-
mation sharing environment where elements of the system are attached to a wireless
and wired network. These days, this concept is not only applicable in the field of
electronics but also in the area of home appliances, smart cars, industrial secu-
rity, etc.
For the individuals and companies associated with solar panels, the IoT makes it
possible to increase the MPPT reliability and performance of the system [7, 8]. For the
controlling or monitoring of the system, the proposed method is discussed in the chapter.

13.2 Micro-grid Model Description

An IEEE 13-node test feeder system is considered as a grid system and is com-
monly used to test features of a standard distribution network of a power system,
operating at 4.16 kV.

13.2.1 Photovoltaic Module

The photovoltaic system operates on the principle of the photovoltaic effect, that is, when
sunlight is radiated upon the semiconductor diode, there is a movement of electrons from
P-type to N-type side of the semiconductor which produces the current in the system [10].

The photovoltaic module is simulated in MATLAB/Simulink using a photovol-


taic array. To generate power of 341.65 kW, the SunPower SPR-445NX-WHT-D
model is used, selected from the module block in the PV array block. Also, there are
8 series-connected modules and 96 parallel-connected modules per string. The
module parameters and data are shown in Table 13.1. These values are calculated
standard temperature and irradiance, i.e., 25°C and 1000 W/m2 irradiance.
13 An IoT-Based Controller Realization for PV System Monitoring and Control 215

Table 13.1 Model parameters Name of the parameters Number/rating


of PV array
Maximum power (watt) 444.86
Cell per module 128
Light generated current Ip(A) 6.2167
Diode saturation current ID (A) 1.3552e-11
Shunt resistance Rsh (ohm) 508.2463
Series resistance Rse (ohm) 0.54861

13.2.2 DC-Link Capacitor and Inverter Circuit

A three-phase inverter circuit is connected across the DC link. DC-link capaci-


tors are employed to stabilize the DC-link voltage of the grid-connected inverter.
Due to temperature and irradiance variation to the total resistance within PV
cells leading to non-linear output efficiency, capacitance is needed to allow elec-
tronic control methods to maintain maximum output power. The three-phase
inverter converts the DC output voltage, i.e., 590V, from micro-grid into AC out-
put, i.e., 240V.

13.2.3 RL Filter, Transformer, and Load

RL filter is connected at the three-phase inverter circuit output, to reduce the


total harmonics distortion in the current. They prevent overcurrent condition in
the system. A three-phase coupling transformer is between the micro-grid sys-
tem and grid. Three-phase coupling transformer, star to the delta, is connected to
step-up the voltage from 240V to 4160V at 50Hz frequency. It has a winding
resistance of 0.01 per unit. The load is the distributed load of IEEE 13-node
feeder systems.

13.2.4 Point of Common Coupling and Grid

Point of common coupling is described by using circuit breakers between PV sys-


tem and grid system; it plays a vital role in the proposed method by isolating the PV
system from grid system, i.e., in the islanded mode at the time of grid blackout or
when grid is under maintenance and to supply power to particular load in case of
deficit energy produced by micro-grid. Grid is a standard IEEE 13-node test feeder
system, as displayed in Fig. 13.1.
216 J. Gupta et al.

Fig. 13.1 Block representation of micro-grid-connected grid system [9]

13.3 Proposed Control Features

Micro-grid is mainly operated in two modes, i.e., islanded and grid-connected. The
proper controlling action of micro-grid is an essential condition for steady and effi-
cient operation in every mode. The controller performs the following functions:
• Control of the power flow between the grid system and micro-grid.
• Proper synchronization of micro-grid with the grid system.
• Proper regulation of load sharing between the grid system and micro-grid.
• Regulation of the frequency and voltage for both operating modes, namely, grid-­
connected and islanded mode.
• Re-optimization of the operation cost of the micro-grid.
• While operating during the switching modes, proper handling of transients and
restoration of the desired condition are necessary.

13.3.1 MPPT Controller

Maximum power point tracking (MPPT) is the algorithm which extracts the maxi-
mum power from the photovoltaic system generated by photovoltaic cell or module
at the specific environmental condition. The specific voltage at which solar device
generates maximum power output is called maximum power point. There are various
types of techniques used for maximum power point tracking, out of which ­perturb and
observe algorithm is used in this work for determining the maximum power point of
13 An IoT-Based Controller Realization for PV System Monitoring and Control 217

the photovoltaic module. This technique uses only one sensor to detect the maximum
power point that is a voltage sensor. It also reduces the cost and complexity of the
system. This technique is quite easy as it has the least time complexity [11, 12].

13.3.2 Inverter Controllers


13.3.2.1 PI Controller

The inverter controller operates on the double closed-loop current algorithm. There are
two loops, outer and inner loop. The voltage control loop is referred to as the outer loop,
and its primary duty is to maintain the DC voltage of PV array. A current control loop is
referred to as the inner loop, and its primary duty is to keep the active and reactive current
signal of the grid (Id and Iq). For stabilizing the power factor at unity for efficient grid
interconnection, the voltage control loop provides the output of Id, which is the reference
of current, while keeping Iq value regulated to minimum, i.e., zero, using a PI controller
to minimize error. The inner control loop provides the voltage output Vd and Vq. The
tracking error can be reduced, and tracking speed can be increased by regulating the PI
controller in the inner current control loop. Forward-feed compensation output is included
in the loop to decrease the disadvantages to the perturbation of the voltage of grid.
The phase-locked loop (PLL) algorithm is applied to regulate the values of the
active power and reactive power injected into the IEEE 13-node feeder systems. The
advantage of a phase-locked loop is that it helps in locking and synchronizing the
frequency and phase angle output of the voltage with reference to the current of the
grid using transformations. The output of the current control loop and voltage con-
trol loop is feed to the converter block whose function is to sense the vector values
of current and voltage, respectively and convert theses values into the d-q reference
frame as DC quantities. The output of the converter block is then feed to over modu-
lation in order to increase the linear region of a three-­phase PWM modulator by
approximately 15%. The output is transferred to PWM three-level pulse generators
to produce the input pulse to the inverter.

13.3.2.2 Artificial Neural Network Controller

An artificial neural network is a massively parallel distributed processor consisting


of simple processing units, which has an inherent property to store knowledge of
experiments and make it available for use. In two respects it resembles the brain as
follows:
• The strengths of the interneuron connection, known as synaptic weights, are used
to store the knowledge gained.
• The network acquires knowledge from its environment through a learning
process.
• Mathematically, neuron modeling can be represented by the following Eqs. 13.1,
13.2 and 13.3:
218 J. Gupta et al.

M
uk = ∑wkj x j (13.1)
j =1

yk = ϕ ( uk + bk ) (13.2)

vk = uk + bk (13.3)

where xj denotes the inputs, wkj represents the synaptic weights of the neuron k,
uk is the combined output due to the inputs, bk is the bias, φ(.) is the activation
function, and yk is the output signal of the neuron. The use of bias has the effect
of applying an affine transformation to the linear combiner output uk in the mod-
eling process.
Levenberg-Marquardt algorithm is used to train the neural network, which is the
second-order optimization of the backpropagation algorithm. This algorithm typi-
cally requires less time and is robotic. This algorithm typically requires less time
and is robust, which are the advantages of this algorithm over the Gauss-Newton
algorithm and gradient descent method. The flow chart explaining the process of a
neural network is shown in Fig. 13.2.

13.3.2.3 Monitoring System

Mainly two main MPPT designing variables, perturbation period (Tp) and perturba-
tion magnitude (x), can be used as metrics. For the higher-resolution system, a
smaller metric has greater acceptability [13–15]. The primary constraint of the men-
tioned system is the speed at which it receives and regulates the desired information.
The constraints of the MPPT variable Tp are described in Eq. 13.4.

∈
− ln  
Tp >= 2 (13.4)
δ .ωn

where wn is the natural frequency of the defined inverter system, δ is the damping
factor of the system, ∈ is the controlling variable of the system, and mainly, 0.1 is
the assigned value. The constraints of variable (x) are described in Eq. 13.5.

1
x> . A.Kph. G .Tp (13.5)
µ

Here, μ is the static gain of the inverter, Kph is the PV material constant related
to the spectral-averaged responsiveness, G is the average of the irradiance slope,
and A is the combination of dependent variable irradiance [16, 17]. Figure 13.3
represents the block diagram of the IoT-based MPPT system of the modules.
13 An IoT-Based Controller Realization for PV System Monitoring and Control 219

Start

Data Base Formation

Training, Validation and


Testing Data set

Select NN Architecture

Train the Network using LM


Algorithm

MSE and R
Increase NN Size
NO goal met?

YES

Generate Simulink Deployment

Save Results

Stop

Fig. 13.2 Flow diagram of ANN

13.4 Results and Discussions

The system is simulated in Simulink for 0.75 s with the following parameters: fre-
quency of 50Hz at standard irradiance and temperature and 590V. The allocation of
PV system between the load bus 632 and 631 in the IEEE 13-node feeder system is
decided by the calculation of the central position in the grid using line segment data.
The inverter circuit is controlled by using a standard PI controller and neural
network. The power quality results using both the controllers are compared. The
power quality results include power output, voltage deviation, total harmonic distor-
tion, and phase angle deviation, which are observed as shown in Figs. 13.4 (a),
13.4(b), 13.5(a), 13.5(b), and 13.6, respectively.
220 J. Gupta et al.

Fig. 13.3 IoT-based


MPPT system [17]
Read Sensors
Algorithm

API Call
Switch Driving

(a) Power output across the load

Voltage using Ps
Voltage using nn

(b) Voltage deviation at different buses

Fig. 13.4 (a) Power output across the load. (b) Voltage deviation at different buses
13 An IoT-Based Controller Realization for PV System Monitoring and Control 221

Fig. 13.5 (a) FFT using PI and (b) FFT using NN

The output graphs of the power quality are divided into three half-sections:
1. The first part is from 0 s to 0.25 s; at this time, system load is met by both PV
system and grid, i.e., the system is operated at grid-connected PV mode.
2. The second part is from 0.25 s to 0.5 s; at this time, system load is met only by
grid, i.e., PV output is not sufficient to meet the required load.
3. The third part is from 0.5 s to 0.75 s; at this time, system load is achieved only
by PV system, i.e., PV output is sufficient to satisfy load, and the system is oper-
ated at islanded mode.
222 J. Gupta et al.

Fig. 13.6 Phase angle at different buses

13.5 Summary

It is presented that the interconnection of a photovoltaic system as a micro-grid


improves the reliability of electric supply and power quality by regulating the IoT-­
based inverter controller systems. The IoT-based ANN inverter control scheme for
the efficient and reliable power transfer between the grid system and micro-grid to
satisfy the desired load is being demonstrated in this work. The IoT-based ANN-­
based controller response has also been compared with IoT-based PI controller,
which justifies its technical feasibility.

References

1. Sadeghian, Hamidreza, Mir HadiAthari, & Zhifang Wang. (2017). Optimized solar photovol-
taic generation in a real local distribution network. In 2017 IEEE Power & Energy Society
Innovative Smart Grid Technologies Conference (ISGT) (pp. 1–5). IEEE.
2. Lasseter, R. H. (2002). Micro-grids. In 2002 IEEE Power engineering society winter meeting.
Conference Proceedings (Cat. No.02CH37309) 1 (pp. 305–308). IEEE.
3. Pham, D. H., Hunter, G., LiLi, & Zhu, J.. (2015). Advanced microgrid power control through
grid-connected inverters. In 2015 IEEE PES Asia-Pacific Power and Energy Engineering
Conference (APPEEC) (pp. 1–6). IEEE.
13 An IoT-Based Controller Realization for PV System Monitoring and Control 223

4. Singh, K., Swathi, P., &Ugender Reddy, M.. (2014). Performance analysis of PV inverter
in microgrid connected with PV system employing ANN control. In 2014 International
Conference on Green Computing Communication and Electrical Engineering (ICGCCEE)
(pp. 1–6). IEEE.
5. Chen, X., Sun, L., Zhu, H., Zhen, Y., & Chen, H.. (2012). Application of internet of things in
power-line monitoring. In 2012 International conference on cyber-enabled distributed com-
puting and knowledge discovery (pp. 423–426). IEEE.
6. Kang, B., Park, S., Lee, T., & Park, S.. (2015). IoT-based monitoring system using tri-level
context making model for smart home services.In 2015 IEEE International Conference on
Consumer Electronics (ICCE) (pp. 198–199). IEEE.
7. Woyte, A., Richter, M., Moser, D., Mau, S., Reich, N., & Jahn, U.. (2013). Monitoring of
photovoltaic systems: good practices and systematic analysis. In Proceedings of the 28th
European photovoltaic solar energy conference (pp. 3686–3694).
8. Belghith, O. B., & Sbita, L.. (2014). Remote GSM module monitoring and photovoltaic system
control.In 2014 First International Conference on Green Energy ICGE 2014 (pp. 188–192).
IEEE.
9. Kersting, W. H.(2009). Distribution feeder voltage regulation control. In 2009 IEEE rural elec-
tric power conference (pp. C1–C1). IEEE.
10. Villalva, M. G., Gazoli, J. R., & Filho, E. R. (2009). Comprehensive approach to modeling and
simulation of photovoltaic arrays. IEEE Transactions on power electronics, 24(5), 1198–1208.
11. Chin, C. S., Tan, M. K., Neelakantan, P., Chua, B. L., & Teo, K. T. K. (2011). Optimization of
partially shaded PV array using fuzzy MPPT. In 2011 IEEE colloquium on humanities, sci-
ence and engineering (pp. 481–486). Penang: IEEE.
12. Esram, T., & Chapman, P. L. (2007). Comparison of photovoltaic array maximum power point
tracking techniques. IEEE Transactions on Energy Conversion, 22(2), 439–449.
13. De Brito, M. A. G., Galotto, L., Sampaio, L. P., de Guilherme, A. e. M., & Canesin, C. A.
(2012). Evaluation of the main MPPT techniques for photovoltaic applications. IEEE
Transactions on Industrial Electronics, 60(3), 1156–1167.
14. Esram, T., & Chapman, P. L. (2007). Comparison of photovoltaic array maximum power point
tracking techniques. IEEE Transactions on Energy Conversion, 22(2), 439–449.
15. Beevi, S.Shanifa, J. M., & Vincent, G.. (2016). A high performance instantaneous resistance
maximum power point tracking algorithm. In 20167th India International Conference on
Power Electronics (IICPE) (pp. 1–5). IEEE.
16. Ma, J., Man, K. L., Ting, T. O., Zhang, N., Guan, S. U., & Wong, P. W. H. (2014). Estimation
and revision: A framework for maximum power point tracking on partially shaded photovol-
taic arrays. In 2014 International symposium on computer, consumer and control (pp. 162–
165). Taichung: IEEE.
17. Bardwell, M., Wong, J., Zhang, S., & Musilek, P. (2018). Design considerations for IoT-based
PV charge controllers. In 2018 IEEE world congress on services (SERVICES) (pp. 59–60).
San Francisco: IEEE.
Chapter 14
Development of an Efficient, Cheap,
and Flexible IoT-Based Wind Turbine
Emulator

Manish Kumar Singla, Jyoti Gupta, Parag Nijhawan, Souvik Ganguli ,


and S. Suman Rajest

14.1 Introduction

Energy is always the most crucial need for growth and development of a nation.
Due to the rapid growth of the world population, changes in technology, and
other political and economic scenarios, the energy demand is increased
immensely. As of now, though more than 80% of the energy is generated by
conventional sources of energy, it is gradually depleting. So, the development of
an efficient and reliable renewable system becomes the prime motive [1]. Wind
technology is rapidly improving in efficiency and scope. Energy researchers
have high future goals for wind energy development. The main difficulty for the
development of the wind energy system is the natural disruption of these
resources. If units are too large to meet demand, severe imbalances can endanger
system security. IoT has emerged as a smart and effective solution for monitor-
ing the renewable energy system. Internet of Things (IoT) can be explained as a
digital network that connects various elements of a particular system and regu-
lates the data according to the situation using advanced embedded systems,
including controllers, meters, and sensors. IoT brought the third revolution in
the field of technology [2]. In the near future, IoT is expected to be widespread
and will cover all aspects of human life, including the production and manage-
ment of renewable energy.
An advanced version of IoT is introduced named as the Internet of Energy (IoE),
which includes the arrangement of energy ecosystem and ICT. The implementation

M. K. Singla · J. Gupta · P. Nijhawan · S. Ganguli (*)


Thapar Institute of Engineering and Technology, Patiala, India
e-mail: jgupta_phd19@thapar.edu; souvik.ganguli@thapar.edu
S. S. Rajest
Vels Institute of Science, Technology & Advanced Studies, Chennai, India

© Springer Nature Switzerland AG 2020 225


A. Haldorai et al. (eds.), Business Intelligence for Enterprise Internet of Things,
EAI/Springer Innovations in Communication and Computing,
https://doi.org/10.1007/978-3-030-44407-5_14
226 M. K. Singla et al.

Fig. 14.1 Application of IoT in smart grid [4]

of IoT as a monitoring system in the smart grid is described in this study [3].
Nowadays, IoT is considered a vital part of the designing and execution of smart
building schemes and smart cities, as shown in Figure 14.1. With the advancement
in the field of IoT, a new theory is propounded defined as smart grid 2.0 which is
also known as the second generation of smart grids. It refers to the improved model
of the smart hybrid system with the IoT and is expected to be implemented in the
next few years [4].
In the new concept (smart grid 2.0), the power exchange between the hybrid
energy system and the power grid is regulated using the advanced-level smart
electric metering system. The amount of energy exchange, the plug and play
capability, and the vital information about the parameter between seller and buyer
could be shared using informatics infrastructure. Plug and play feature indicates
that a demand-side electricity source can inject power to the grid as easy as insert-
ing a plug into an outlet, i.e., ability to inject even a very minimal power genera-
tion into the power grid like vehicle-to-grid (V2G) power transfer. Such types of
consumers which have the ability to deliver power to another system are defined
as prosumers [5]. Still, now there are no technologies that can disconnect and
connect the distributed generation system with a grid at each desired moment. It
is sensible to study wind energy conversion system (WECS) with respect to
cybersecurity. The overall effect and significance of cybersecurity on the reliabil-
ity of WECS have been studied [6, 7].
14 Development of an Efficient, Cheap, and Flexible IoT-Based Wind Turbine Emulator 227

14.2 Wind Energy System with IoT

Wind energy is nowadays emerging as one of the most promising technologies.


However, the output of the wind energy conversion system (WECS) is dependent on
wind flow, which is sometimes erratic and at times unpredictable. By this time,
developing a control scheme and ensuring a reliable WECS has been a significant
area of focus. However, the efficacy of the control scheme can be assured by several
experiments with an actual wind turbine (WT). The actual wind turbine is ulti-
mately dependent on environmental conditions. Such dependency may cause indefi-
nite delays. Moreover, several worst cases for which system needs to be tested may
never occur, and thus performance and reliability of controllers for those cases are
always questionable. Another way is to set up a WT in the laboratory. But it is dif-
ficult to set up a WT in the laboratory because of challenges like space and con-
trolled environment, for example, wind tunnel.
A solution to these problems can be found in a wind turbine emulator. Wind
turbine emulator mimics the behavior of actual wind turbine under controlled man-
ner. Line diagram of wind turbine emulator and power conditioning unit is shown in
Fig. 14.2. Essentially it simulates the same operating pattern at hardware level in
real time similar to what an actual wind turbine does at given operating parameters
of wind speed and pitch angle. The term emulation is coined for simulation prac-
tices which involve hardware platform. It provides a fast-configurable testing plat-
form. Usually these are performed in real time. Emulator is similar to
hardware-in-the-loop simulation concepts. Wind turbine emulator can find applica-
tions in numerous fields. It provides a flexible testing platform for the study of
dynamic and steady-state behavior of wind turbine. Beginners can learn about
power/wind speed, torque/turbine speed, and power/turbine speed characteristics of
a wind turbine and can perform comparative studies about how changes in param-
eters of wind turbine affect the behavior. For other applications, this emulator can
be coupled to the generator (induction, PMSG, DFIG) followed by power electron-
ics in place of an actual wind turbine. So, researchers would not have to rely on
environmental conditions which are appropriate for driving the wind turbine at
some desired operating point. Since the operating point of wind emulator can be
controlled, researchers can simulate all possible scenarios of operation of a wind
turbine and accordingly can modify the power electronics and control algorithms.
Consequently, it will improve the product quality and reliability. The operating
point of wind emulator can be controlled using IoT. The block diagram of wind
operating system with IoT system is shown in Figure 14.3.
IoT topology in combination with ICT infrastructure allows the wind power pro-
ducers to regulate the output power at maximum efficiency and provides the accu-
rate maintenance reminder of each component at regular interval in order to avoid
any huge disaster. There are various algorithms which help in formulating the
desired schedule for maintenance like machine learning, fuzzy logic, neural net-
work, etc. For instance, on-time maintenance can reduce the index of levelized
228 M. K. Singla et al.

Fig. 14.2 Line diagram of wind turbine emulator and power conditioning unit

Wind AC-DC DC-AC Grid


power converter converter
generation

Load
IoT
system

Fig. 14.3 Block representation of wind operating system with IoT system
14 Development of an Efficient, Cheap, and Flexible IoT-Based Wind Turbine Emulator 229

energy costs (LCoE) for wind assets that denotes the net present value of the unit
power cost over the lifetime of the turbines [8].
In digital communication system, the beneficiary data of each element of system
is collected and processed using machine learning algorithm [9]. The main two
problems related with the IoT system are the delay in information transfer espe-
cially in offshore wind farms and limitation on the bandwidth for exchanging infor-
mation. Therefore, if essential information could be received and processed at a
faster rate, then corrective measures for the system at the time of failure can be
automated like shutting down of turbine system in case of turbulence. Therefore, we
need the deployment of more IoT systems with improved algorithm for monitoring
the wind power generation system [10].

14.3 Results

The wind emulator system developed is run at varying motor buck duties in order to
obtain the effect on the various output parameters such as generator voltage, genera-
tor current, generator speed, battery voltage, DC link voltage, and motor armature
current. The following results are also obtained using LabVIEW software. Their
variation with time is shown in Table 14.1.
Thus, all the output parameters are obtained in the graph with respect to time
which are shown in Figure 14.4 and are presented in two parts.
Speed is varying with change in the duty ratio while field and armature voltages
remain invariant. The graph shown in Fig. 14.4 provides the variation of the motor
output parameters versus the time. Thus, an efficient, economical, and flexible IoT-­
based wind turbine emulator is successfully developed in this chapter.

Table 14.1 Results of motor parameters


Motor buck duty (%) Field voltage (V) Armature voltage (V) Speed (rpm)
2 230 229 60
4 230 229 74
6 230 229 96
8 230 229 116
10 230 229 131
12 230 229 146
14 230 229 164
16 230 229 180
18 230 229 199
20 230 229 215
22 230 229 229
230 M. K. Singla et al.

Fig. 14.4 Variation of motor output parameters with time

14.4 Summary

The proposal for a highly efficient, cost-effective, and flexible IoT-based wind tur-
bine emulator is thus presented in this chapter. The monitoring can be performed
using a web-based service which would receive the wind turbine data and convert it
into useful information. The challenge however still lies with the feasibility of the
amount of data that can be accumulated from each and every element of a wind
power system in a real-time environment.
14 Development of an Efficient, Cheap, and Flexible IoT-Based Wind Turbine Emulator 231

References

1. Kalaiarasi, D., Anusha, A., Berslin Jeni, D., & Monisha, M.. (2016, April). Enhancement
of hybrid power systems using IoT. International Journal of Advanced Research Trends in
Engineering and Technology (IJARTET), 3(19).
2. Buyya, R., Calheiros, R. N., & Dastjerdi, A. V. (Eds.). (2016). Big data: principles and para-
digms. Cambridge: Morgan Kaufmann.
3. Byun, J., Kim, S., Sa, J., Kim, S., Shin, Y.-T., & Kim, J.-B. (2016). Smart city implementa-
tion models based on IoT technology. Advanced Science and Technology Letters, 129(41),
209–212.
4. Cao, J., & Yang, M. (2013). Energy internet—towards smart grid 2.0. In 2013 Fourth interna-
tional conference on networking and distributed computing (pp. 105–110). IEEE.
5. Ritzer, G. (2015). Automating prosumption: The decline of the prosumer and the rise of the
prosuming machines. Journal of Consumer Culture, 15(3), 407–424.
6. Zhang, Y., Xiang, Y., & Wang, L. (2016). Power system reliability assessment incorporating
cyber-attacks against wind farm energy management systems. IEEE Transactions on Smart
Grid, 8(5), 2343–2357.
7. Hatzivasilis, G., Fysarakis, K., Soultatos, O., Askoxylakis, I., Papaefstathiou, I., & Demetriou,
G. (2018). The industrial internet of things as an enabler for a circular economy Hy-LP: A
novel IIoT protocol, evaluated on a wind park’s SDN/NFV-enabled 5G industrial network.
Computer Communications, 119, 127–137.
8. Lotfi, H., & Khodaei, A. (2016). Levelized cost of energy calculations for microgrids. In 2016
IEEE Power and Energy Society General Meeting (PESGM) (pp. 1–5). IEEE.
9. Alhmoud, L., & Al-Zoubi, H. IoTapplications in wind energy conversion systems. Open
Engineering, 9(1), 490–499.
10. Moness, M., & Moustafa, A. M. (2015). A survey of cyber-physical advances and challenges
of wind energy conversion systems: prospects for internet of energy. IEEE Internet of Things
Journal, 3(2), 134–145.
Chapter 15
An Application of IoT to Develop Concept
of Smart Remote Monitoring System

Meera Sharma, Manish Kumar Singla, Parag Nijhawan, Souvik Ganguli ,


and S. Suman Rajest

15.1 Introduction

One of the main contributors of clean energy all over the world, currently, is the
solar photovoltaic system. A PV system’s power generation potential is the main
factor to be determined [1]. This potential may vary depending upon various con-
straints such as the technology employed as well as the location considered. The
expense of every kWh pay backs the return on investment on the electricity potential
[2]. Hence, before the installation of PV systems, certain measures need to be taken,
so as to have larger energy potentials. There may be chances of collapse and main-
tenance issues at the time of operation of the system despite the efforts made during
or before the installation of the PV systems. Such problems are occurring more
often where the installations are in remote locations. Therefore, to mitigate such
problems, a suitable approach is desired. The most accurate way to avoid such
issues is the frequent check that is almost impossible for a person to carry out. Even
problems involving concentration, giving full attention, and identification of solu-
tion cannot be encountered [3].
The best way to cater such a problem is to adopt the most emergent method
known as the “Internet of Things” (IoT) for remote evaluation of models. IoT helps
us to interact with objects used in our daily life in a much quicker and easy way with
the usage of communication devices following network protocols [4, 5].
IoT has a vast range of applications including infrastructure, industrial automa-
tion, healthcare support, home power management and the renewable energy frame-
work, traffic maintenance, automotive enterprise, micro-grids, and intelligent drive

M. Sharma · M. K. Singla · P. Nijhawan · S. Ganguli (*)


Thapar Institute of Engineering and Technology, Patiala, India
e-mail: souvik.ganguli@thapar.edu
S. S. Rajest
Vels Institute of Science, Technology & Advanced Studies, Chennai, India

© Springer Nature Switzerland AG 2020 233


A. Haldorai et al. (eds.), Business Intelligence for Enterprise Internet of Things,
EAI/Springer Innovations in Communication and Computing,
https://doi.org/10.1007/978-3-030-44407-5_15
234 M. Sharma et al.

systems, among others [5, 6]. Based on the advent of the IoT, solar PV systems are
the latest targets being focused upon. This is because of their increasing usage in the
current energy sector trends especially in energy distribution. Solar PV system’s
popularity would be a major breakthrough for the implementation of IoT systems in
combination with it, thereby giving an edge for the IoT service suppliers as well as
the consumers.

15.2 Photovoltaic Systems

A PV system consists of the arrangement of a PV module, power converters, and


storage devices. In essence it is the power harvester, which changes sunlight into
electricity [7]. This technique is quite different from the traditional process which
involves fossil fuels for power generation. Although power transmission and distri-
bution embraces similar traditional methods, PV arrays are formed by grouping PV
modules; such PV arrays are known as PV generators when arranged in series and
parallel configurations [8–10]. They are then installed in a manner such that they are
exposed to direct sunlight. The DC electricity generated from the PV generator is
changed into AC with the assistance of the inverters. This power can be self-­
consumed or distributed to the energy grid through the transmission network [11].
Nonetheless, the energy can be stored using the batteries instead of being trans-
ferred. The PV models are grouped into two forms, i.e., the off-grid and the on-grid
PV models, centered on various forms of functional components. An illustration in
Fig. 15.1 provides a layout of the operation of the photovoltaic system.

Fig. 15.1 Layout of a


typical photovoltaic system Sun
[12]

PV Array

AC Load

DC
DC Chopper
Smart Meter
AC
Battery
DC
Inverter

Electric Grid
15 An Application of IoT to Develop Concept of Smart Remote Monitoring System 235

15.2.1 IoT and Its Requirement in Photovoltaic Systems

Internet of Things, abbreviated as IoT, is a technology that is developed by grouping


together “wireless technologies, micro-electromechanical systems, and the Internet”
[3–6]. The mechanical/digital machines, computing devices and objects, solitary
identifiers, and other such analogous things coordinating together constitute
IoT. Because of this synchronization, which transmits the data across the network,
the distance between the operational systems and information technology is closed
without the aid of human-to-human and human-to-machine interaction.
The contemporary science and engineering systems cannot solve the most intri-
cate issues which IoT can solve. The operational behavior of various components of
the PV systems, which are used for generating power, varies. In short, a constant
generation of power is not achievable throughout because of the solar intensity
being weather dependent and time varying [12]. This has an indirect effect on the
working of other components of the device such as voltage levels of power convert-
ers, status of battery charging, and load energy demands. Some environmental con-
ditions, such as accumulation of dust, are also sometimes responsible for the poor
performance of the PV system. Nevertheless, these problems can lead to the col-
lapse of the whole system in longer terms. In order to maintain the operating data
log, humans face difficulties to monitoring since it requires visits to the plant site
time and time again. Henceforth, humans consume a lot of time in addressing these
failures because of system breakdown or bad performance [3]. To check the param-
eters of the system and store them in cloud, a continuous monitoring system together
with the PV system is to be equipped. This stored data will provide the performance
parameters and the causes of poor performance and will make troubleshooting and
maintenance operation much easier and faster. Therefore, the need for IoT is neces-
sary to optimize the device parameters with the option of remote control.

15.2.2 IoT-Based Photovoltaic System Architecture

The IoT architecture of a photovoltaic system consists of three distinct layers.


Figure 15.2 clearly displays the IoT photovoltaic system architecture. The initial
layer incorporates the PV model design ambience and is connected for user satis-
faction according to the required configurations. In this particular case, the Arduino
server is interlinked with the components of the PV device, creating the second
layer of the IoT architecture. Along with an Internet firewall option, using a router,
the web server can be inter-connected with the hardware projects of the PV scheme,
hence forming the gateway linkage in this second layer. The Arduino server is
majorly responsible for this integration. It carries out the main functions of con-
trolling, monitoring, and managing the PV scheme hardware constituents. The
server collects the data from the third and last layer known as the remote monitor-
ing and control layer. This information is transmitted to the storage devices that
236 M. Sharma et al.

Fig. 15.2 IoT-aided layout for photovoltaic system [12]

help generate periodic reports. Using an Android interface with cloud data storage
via a Wi-Fi network, these data can be drawn up in the form of visual graphs and
reports, and then the users can access it accordingly.

15.2.3  roposed Concept for IoT-Aided System


P
for Photovoltaic Monitoring

This research proposes a device that evaluates the condition of the PV framework
based on the IoT-centered network with the purpose of remotely controlling it. For
the transmission of sensor information, a mobile radio network is used. The remote
server data is sent through a GPRS module [13]. Figure 15.3 displays IoT technol-
ogy schematics for a solar power plant.
A three-layered schematic diagram having the bottom layer as the sensing layer
consists of current and voltage sensors, irradiance-measuring device (pyranome-
ter), and other sensors. The sensing layer also consists of a microcontroller-based
data processing which is acquired by the sensors. A wireless module is utilized to
communicate with the microcontroller in order to initialize and start transmitting
data to the server.
The second layer, known as the system layer, is where information logging is
done from the plant for real-time transmission and includes database storage as
15 An Application of IoT to Develop Concept of Smart Remote Monitoring System 237

Plant Monitoring Maintenance Data Analytics Historical Data analysis

Fault Monitoring Inspection Scheduling Generation Monitoring Other Decision Making

Application Layer

Internet Database
Network Layer

Microcontroller Wireless Module


Sensors

Sensing Layer

Fig. 15.3 IoT-based solar power plant [7]

well [14]. The application layer further uses this stored and processed data from
the network layer. Based on the collected data’s processing and storage, the web-
based services are hence designed smartly. In order to help in monitoring the
plant’s performance, graphical user interfaces are employed. With the console, the
decision-­making time is shortened, indicating the administrator with historical
data-based decision.
A remote monitoring system based on IoT makes it much easier to supervise the
solar power plant’s performance as a whole using a web-based technique as shown
in Fig. 15.4.

15.3 Summary

The main advantage of using IoT photovoltaic technology is that we can accurately
view the status of our property from the central control panel. Through connecting
the computer to the cloud network, we can even pinpoint where the problem lies
and allow technicians to repair it long before the entire system breaks down. The
network is less vulnerable to the production issues (due to power outages) and
potential security threats through the use of the Internet of Things. Through install-
ing an IoT solution directly and linking solar devices, we can monitor our solar
238 M. Sharma et al.

Fig. 15.4 Complete schematic of the smart remote monitoring system [7]

power system’s broadband even when there are thousands of devices connected to
the network. In addition to real-time business notifications, the solar industry’s
Internet of Things increases energy efficiency and profitability by gathering histori-
cal modeling data. This makes energy generation more efficient in terms of both
costs and logistics.

References

1. Cota, O. D., & Kumar, N. M. (2015). Solar energy: a solution for street lighting and water
pumping in rural areas of Nigeria. Proceedings of International Conference on Modelling,
Simulation and Control (ICMSC-2015), 2, 1073–1077.
2. Pezzotta, G., Pinto, R., Pirola, F., & Ouretani, M. Z. (2014). Balancing product-service pro-
vider’s performance and customer’s value: the service engineering methodology (SEEM). In
6th CIRP conference on industrial product-service systems (Vol. 16, pp. 50–55). Elsevier.
3. Internet of Things (IoT). IoT Agenda. http://internetofthingsagenda.techtarget.com/definition/
Internet-ofThings-IoT.
4. Atzori, L., Iera, A., & Morabito, G. (2010). The internet of things: A survey. Computer
Networks, 54(15), 2787–2805.
5. Zanella, A., Bui, N., Castellani, A., Vangelista, L., & Zorzi, M. (2014). Internet of things for
smart cities. IEEE Internet of Things Journal, 1(1), 22–32.
6. Bellavista, P., Cardone, G., Corradi, A., & Foschini, L. (2013). Convergence of MANET and
WSN in IoT urban scenarios. IEEE Sensors Journal, 13(10), 3558–3567.
7. Adhya, S., Saha, D., Das, A., Jana, J., & Saha, H. (2016). An IoT based smart solar photo-
voltaic remote monitoring and control unit. In 2016 2nd international conference on Control,
Instrumentation, Energy & Communication (CIEC) (pp. 432–436). IEEE.
8. Reddy, B. V., Sata, S. P., Reddy, S. K., & Babu, B. J. (2016). Solar energy analytics using
internet of things. International Journal of Applied Engineering Research, 11(7), 4803–4806.
15 An Application of IoT to Develop Concept of Smart Remote Monitoring System 239

9. Nadpurohit, B., Kulkarni, R., Matager, K., Devar, N., Karnawadi, R., & Carvalho, E. (2017).
Iot enabled smart solar pv system. International Journal of Innovative Research in Computer
and Communication Engineering, 5(6), 11324–11328.
10. Guamán, J., Guevara, D., Vargas, C., Ríos, A., & Nogales, R. (2017). Solar manager:
Acquisition, treatment and isolated photovoltaic system information visualization cloud plat-
form. International Journal of Renewable Energy Research, 7(1), 214–223.
11. Vignesh, R., & Samydurai, A. (2017). Automatic monitoring and lifetime detection of solar
panels using internet of things. International Journal of Innovative Research in Computer and
Communication Engineering, 5(4), 7014–7020.
12. Kim, J., Byun, J., Jeong, D., Choi, M.-i., Kang, B., & Park, S. (2015). An IoT-based home
energy management system over dynamic home area networks. International Journal of
Distributed Sensor Networks, 11(10), 828023.
13. Chen, X., Sun, L., Zhu, H., Zhen, Y., & Chen, H. (2012). Application of internet of things in
power-line monitoring. In 2012 International conference on cyber-enabled distributed com-
puting and knowledge discovery (pp. 423–426). IEEE.
14. Peijiang, C., & Xuehua, J. (2008). Design and Implementation of Remote monitoring system
based on GSM. In 2008 IEEE Pacific-Asia workshop on computational intelligence and indus-
trial application (Vol. 1, pp. 678–681). IEEE.
Chapter 16
Heat Maps for Human Group Activity
in Academic Blocks

Rajkumar Rajasekaran, Fiza Rasool, Sparsh Srivastava, Jolly Masih,


and S. Suman Rajest

16.1 Introduction

Sensing and identifying human group activities has garnered a rising research inter-
est in many practical applications like video surveillance and crime detection.
Multifarious algorithms have been designed and proposed for the recognition of
group activities or interactions. Though many of the already prevailing algorithms
have unearthed the wide-ranging features among the activities such as the full
movement data, they were unsuccessful in obtaining information about the temporal
motion (e.g., they failed to depict the time and place of a person) [1]. Hence, they
suffered from restrictions in case of detecting and identifying more convoluted
group activities. Though these other techniques include and study the temporal data
by using the chain models like HMM (hidden Markov models), they still suffer from
one disadvantage which is the requirement of enormous amounts of training data,
along with managing the variations in motion operation and behavior. Also, several
uncertainties also play a challenging concern in recognition of group activity. Due
to inherent variation in the motion of people in cluster activities, the precision and
correctness of recognition might be prominently influenced by the unsure changing
characteristic of group motion [2].
The ultimate goal of the system is optimization of targeted digital advertising. It
is the most effective form of advertising [3]. Most advertising done these days is
online, and almost all of it is targeted. Targeted advertising is a method by which

R. Rajasekaran (*) · F. Rasool · S. Srivastava


Vellore Institute of Technology, Vellore, India
J. Masih
Erasmus School of Economics, Rotterdam, Netherlands
S. S. Rajest
Vels Institute of Science, Technology & Advanced Studies, Chennai, India

© Springer Nature Switzerland AG 2020 241


A. Haldorai et al. (eds.), Business Intelligence for Enterprise Internet of Things,
EAI/Springer Innovations in Communication and Computing,
https://doi.org/10.1007/978-3-030-44407-5_16
242 R. Rajasekaran et al.

advertisers are able to provide more specific and relevant advertisements to users
based on data they have collected on the users from various sources – services like
Google AdWords and Facebook advertising. The reason why this type of advertis-
ing is extremely effective is because it utilizes data that is most pertinent to the user
or taker. Through this chapter, we put forth a very innovative heat map algorithm for
the recognition of human group activity in academic buildings and hostel blocks.
The main objectives of this chapter can be stated as the following points:
1. This chapter presents a completely new heat map (HM) aspect to characterize
group activities in academic buildings and hostel blocks. This proposed HM will
be able to efficiently fetch the temporal motion data of the crowd activities based
on the temporal and cultural factors [4].
2. This chapter proposes to implement a thermal diffusion method to generate the
heat map. In this technique, we will be able to efficiently address the motion
uncertainty from different people.
3. We also intend to perform a new SF (surface fitting) method which can also
identify the group activities. This presented process can essentially study the
features of the heat map and can provide us with recognition results effectively.

16.2 Literature Survey

Anandakumar and Umamaheswari [5], Arizona State University, USA, applied a


dynamic, actual-time heat map data capture, and for two military simulations, they
generated tools which include unmanned aerial vehicle sensor operator scenario and
ground-based combat scenario. Arulmurugan and Anandakumar [6] have developed
SURV or Smart Urban Visualization system. SURV is a new framework developed
by applying HTML, CSS, as well as Mapbox.js and CARTO.js JavaScript libraries
in order to generate dynamic visualizations and interactive maps. Python, the most
popular versatile language of these times, is also integrated and used with this sys-
tem due to its excellent performance in data processing. SURV offers an interface or
a web app where the end-user will be able to perform three-dimensional (3D) visu-
alizations along with data animation to deliver better data analysis.
Hirenkumar Gami, Miami University, USA, has only used a single PIR sensor
combined with machine learning algorithms and signal processing to clearly study
and predict the presence, proximity, and direction of the crowd movements on the
floors of a building. [7], PSUT, Jordan, presented the significance of visualization
techniques when implemented on huge real-time data sets from a certain corpora-
tion in Jordan as detailed tools for analysis of data. The name of the software uti-
lized to analyze and visualize this data is known as Tableau. [8] in his research paper
perceives the pairwise actions by fetching the causality features from the bitrajecto-
ries. Ni et al. further extended the cause and effect characteristics into three different
kinds. Among these, the first is individuals, the second is pairs, and the third is
groups. [9] discovers and studies crowd activities by presenting connected operating
16 Heat Maps for Human Group Activity in Academic Blocks 243

segmentations in order to depict and display the corresponding congruence among


the people. Cheng et al. propose to study the pattern for group activity by taking into
consideration the Gaussian parameters and creating trajectories that are determined
by the calculation from many people.

16.2.1 Detailed Working

The system is composed of two components – the hardware detectors and the soft-
ware backend. The hardware components have been prototyped using an Arduino
development board with PIR sensors. These are fixed at certain key positions in the
academic buildings, and based on the factors we considered, a number of people
visiting the academic buildings are stored in a software database. And using the
surveillance cameras in the academic buildings, the activities of students and staff
are continuously recorded. As discussed before, if the overall characteristics are
straight away being obtained from the motion data, most of the temporal data/infor-
mation will be lost. To avoid this loss of information, it is proposed that the trajec-
tory is represented as a sequence of heat sources. So, for converting the trajectory to
the series of heat sources, we need to firstly dissect the full video scene into tiny
patches which are non-overlaying. In the case wherein a trajectory goes falling into
a patch, this case will be regarded as a heat source. In this way, we can convert a
trajectory into several heat source series, and the thermal values pertaining to the
series of heat sources can be organized in an increasing order following the course
of the trajectory [10]. The obtained temporal data converted into some useful infor-
mation is effectively inserted.
Besides, as the trajectories of people might show huge deviations, straight away
utilizing the series of heat sources by way of characteristics shall be significantly
influenced by the motion fluctuation. Therefore, to reduce the fluctuation in motion,
a consecutive technique can be implemented by introducing a further process in the
form of thermal diffusion which disperses the heat from the series of heat sources.
This is called diffusion result of the heat map. Although this process has been used
in the past, this is the first time we are using it for human activity recognition in
academic buildings and hostel blocks. With our proposed heat map feature, we can
describe the activities’ information of motion with the help of 3D surfaces. But
again, the problem occurs in choosing an appropriate technique for implementing
recognition which uses this feature of HM. Therefore, we additionally proposed the
SF (surface fitting) method for recognition of activities [11]. In the proposed SF
technique, firstly, a bunch of prevailing classes of surfaces are recognized in order
to represent the diverse activities. Secondly, the various parallels concerning the
input surface HM and the standard surface classes can be easily determined. In the
final step, the standard surface class that almost matches will be selected, and its
equivalent human activity will be declared as the activity which is most recognized
for the inputted heat map. Therefore, the data that is generated from HMs is passed
244 R. Rajasekaran et al.

onto the digital advertising partners who then utilize this information to serve the
targeted advertising.

16.2.2 Hmb Method

The working of the heat map-based algorithm is shown below. Firstly, the input
trajectories are converted into a series of heat sources, followed by the process of
thermal diffusion to generate the HM characteristic which calculates the group
activity input [12]. After this, the surface fitting technique is useful for implement-
ing the activity recognition. The thermal diffusion and the transfer of heat source
series, along with the SF method, are the prime components of the presented algo-
rithm shown in Fig. 16.1.

16.3 Experimental Results

The experimental outcome for the proposed heat map-based algorithm is discussed
here. In the experiment, the HMB algorithm is compared with three other methods:
the WF-SVM algorithm, the GRAD algorithm, and the PGTB algorithm. Here, we
performed four individual experiments and the average of the results is calculated.
Further, we have determined miss and false alarm (FA) for the four different meth-
ods and then calculated their total as shown in Table 16.1.
From the above table, it is clear that the PGTB algorithm fails to yield good
results. Comparatively, the WF-SVM and the GRAD algorithms provide better
results and performance with the help of many features which are distinguishable.
When a comparison is made among these mentioned methods, the proposed HMB
algorithm shows the finest performance. And, as activities are detected, we can now
send the data to the advertising partner who can advertise based on the people’s
interests and activities (Fig. 16.2).

16.3.1  nalysis of Human Footfall in an Academic


A
Building (in Terms of Density)

From the above table, we can see that the density of human footfall is high toward
the morning of each day. This density in the early morning stays maximum for the
entire day except in the case of Monday, where 10 am–2 pm has increased density
over the morning 6 am–10am slot. For almost all the days, human density dwindles
as the day progresses with least amount of density at nighttime. One more notice-
able change takes place on Wednesday, 10 am–2 pm slot, where the density falls
16 Heat Maps for Human Group Activity in Academic Blocks 245

Group Activity Trajectories

Heat Source Transfer Series

Diffuse to Create Heat


Maps

Recognize Group Activity

Surface Filtering Recognition Result


Method

Fig. 16.1 Heat map method

Table 16.1 Heat map result


HMB WF-SVM PGTB GRAD
Miss 0.9% 11.8% 33.1% 10.7%
FA 1.2% 0.4% 4.2% 1.4%
Miss 5.6% 8.5% 8.1% 16.5%
FA 0.8% 2.1% 34.5% 0.9%
Miss 4.7% 8.8% 31.9% 13.9%
FA 0.8% 1.9% 0.1% 2.1%
Miss 3.2% 5.2% 34.8% 8.5%
FA 0.3% 0.3% 6.8% 0.7%
Miss 2.9% 4.7% 45.1% 9.9%
FA 1.3% 1.7% 0.02% 2.8%
Miss 5.2% 3.1% 59.8% 2.7%
FA 0.5% 2.0% 0.3% 1.1%
3.8% 6.9% 36.9% 11.2%

sharply as no classes are scheduled during this reserved slot meant for non-­academic
activities as shown in Table 16.2.
The following can be visualized better using the following heat map:
Thus, this heat map shown in Fig. 16.3 helps us to visualize that the density of
people in the academic building is maximum during the working hours which is
indicated by the bright red shades. As the day progresses, the pattern of the heat map
turns toward lighter pastel colors when almost all the students have returned to their
hostels and the footfall is negligible.
246 R. Rajasekaran et al.

Fig. 16.2 Human footfall in an academic block

Table 16.2 Day calculations 1


Days 6 am–10 am 10 am–2 pm 2 pm–6 pm 6 pm–10 pm 10 pm–6 am
Monday 0.64 0.7 0.51 0.32 0.0012
Tuesday 0.73 0.57 0.42 0.21 0.001
Wednesday 0.82 0.35 0.56 0.41 0.0015
Thursday 0.58 0.64 0.55 0.35 0.001
Friday 0.61 0.52 0.61 0.18 0.003

The bar graphs and the line graphs presented in Fig. 16.4 provide a better picture
in visualizing the footfall of the students in the building. These can prove to be
really helpful to target the temporal aspects of the situation and to tackle any mishap
as shown in Fig. 16.5 Table 16.3.

16.3.2 Visualization for Hostel Blocks

Here, we observe that the data collected through the sensors is at its maximum dur-
ing the nighttime between 10 pm and 6 am when nearly every student has returned
to the hostel post in time. The density from 6 am to 10 pm is around 0.5 as the
16 Heat Maps for Human Group Activity in Academic Blocks 247

Fig. 16.3 Heat map

classes start from 8 am. The 10 am–2 pm slot also sees a rise in density as most of
the students return to the hostel during lunch hours. Overall, the heat map for human
footfall at hostels can be plotted as follows in Fig. 16.6:
Here, a line graph shown in Fig. 16.7 and the bar plots provide a better insight to
the data collected by the PIR sensors. The overall human footfall through these
visualizations can prove to be extremely beneficial in the process of group activity
detection, analysis, and recognition as shown in Fig. 16.8.

16.4 Summary

Thus, through this chapter, we have tried to devise the HMB algorithm, which is an
algorithm based on heat maps, for the purpose of detecting and identifying group
activities, in order to avoid loss of life and property in case of a mishap. The use of
PIR sensors which are exceptional devices for detecting presence of humans, with
the help of a minor form-factor and strong-featured design, is the best choice for
248 R. Rajasekaran et al.

Fig. 16.4 Bar graphs and the line graphs

Fig. 16.5 Foot path comparison


16 Heat Maps for Human Group Activity in Academic Blocks 249

Table 16.3 Day calculations 2


Days 6 am–10 am 10 am–2 pm 2 pm–6 pm 6 pm–10 pm 10 pm–6 am
Monday 0.49 0.38 0.33 0.69 0.9988
Tuesday 0.51 0.41 0.38 0.78 0.999
Wednesday 0.52 0.68 0.36 0.66 0.9985
Thursday 0.48 0.54 0.32 0.68 0.999
Friday 0.46 0.56 0.37 0.82 0.997

Fig. 16.6 Human foot path

achieving cost-efficient surveillance. We, therefore, propose to generate the heat


map, for the purpose of demonstrating the group activities, and then consequently
implement the SF method for recognition of crowd activity. Based on the results, we
can effectively avoid any mishap based on the density and activity detection of the
people and hence can save many precious lives.
250 R. Rajasekaran et al.

Fig. 16.7 Human foot path densities

Fig.16.8 Overall analysis


16 Heat Maps for Human Group Activity in Academic Blocks 251

References

1. Chu, H., Lin, W., Wu, J., Zhou, X., Chen, Y., & Li, H. A new heat-map-based algorithm for
human group activity recognition. ACM.
2. Amresh, A., Femiani, J., Fairfield, J., & Fairfield, A. (2013, July). UAV sensor operator train-
ing enhancement through heat map analysis. In Information Visualisation (IV), 2013 17th
international conference. IEEE.
3. Ihaddadene, N., & Djeraba, C. (2008). Real-time crowd motion analysis. In 2008. ICPR 2008.
19th international conference on pattern recognition (pp. 1–4). IEEE.
4. Anandakumar, H., & Umamaheswari, K. (Mar. 2017). Supervised machine learning techniques
in cognitive radio networks during cooperative spectrum handovers. Cluster Computing, 20(2),
1505–1515.
5. Anandakumar, H., & Umamaheswari, K. (Oct. 2018). A bio-inspired swarm intelligence tech-
nique for social aware cognitive radio handovers. Computers & Electrical Engineering, 71,
925–937. https://doi.org/10.1016/j.compeleceng.2017.09.016.
6. Arulmurugan, R., & Anandakumar, H. (2018). Early detection of lung Cancer using
wavelet feature descriptor and feed forward Back propagation neural networks classi-
fier. Lecture Notes in Computational Vision and Biomechanics, 103–110. https://doi.
org/10.1007/978-3-319-71767-8_9.
7. Haldorai, A. R., & Murugan, S. Social aware cognitive radio networks. In Social net-
work analytics for contemporary business organizations (pp. 188–202). https://doi.
org/10.4018/978-1-5225-5097-6.ch010.
8. Arulmurugan, R., & Anandakumar, H. (2018). Region-based seed point cell segmentation and
detection for biomedical image analysis. International Journal of Biomedical Engineering and
Technology, 27(4), 273.
9. Suganya, M., & Anandakumar, H.. (2013, December). Handover based spectrum alloca-
tion in cognitive radio networks. In 2013 International Conference on Green Computing,
Communication and Conservation of Energy (ICGCE). https://doi.org/10.1109/
icgce.2013.6823431. https://doi.org/10.4018/978-1-5225-5246-8.ch012
10. Huacón, C. F., & Pelegrin, L. (2018). SURV: A system for massive urban data visualization. In
URTC 2017. IEEE MIT.
11. Gami, H. (2017). Movement direction and distance classification using asingle PIR sensor. In
IEEE sensors letters. IEEE.
12. Sharawi, L. I., & Sammour, G. Utilization of data visualization for knowledge discovery in
modern logistic service companies. In SENSET 2017. IEEE.
Chapter 17
Emphasizing on Space Complexity
in Enterprise Social Networks
for the Investigation of Link Prediction
Using Hybrid Approach

J. Gowri Thangam and A. Sankar

17.1 Introduction

Social Network is a structure comprised of a lot of actors. It provides a set of meth-


ods for analyzing the structure of whole social entities as well as a variety of theo-
ries explaining the patterns are pragmatic in these structures. By using those
structures, local and global patterns and visible entities can be identified and the
network dynamics can be examined.
Social Network Analysis (SNA) [1] is based on the assumption of the importance
of relationships among interacting units. In recent decades, online social networks
[2] have turned out to be progressively significant assets for individual interaction,
information process and social impact dispersion. Understanding and modeling the
mechanisms by which these networks evolve becomes a fundamental issue and
dynamic research area. Therefore, the conviction of SNA has pulled in significant
intrigue and interest from social network community. Quite a bit of this intrigue can
be credited to the engaging focal point of social network investigation on connec-
tions among social elements and ramifications of these connections.
It has become very significant in today’s scenario because of the accomplishment
of online person to person communication. This makes use of network theory [3] to
analyze social networks. SNA views social relationships in terms of network theory,
consisting of nodes, representing individual actors within the network, and ties or
edges that represent relationships between the individuals, such as friendship, kin-
ship, and organizations. These networks are often depicted in a social network dia-
gram or sociograph, where nodes are represented as points and ties are represented
as lines.

J. G. Thangam (*) · A. Sankar


Department of Computer Applications, PSG College of Technology,
Coimbatore, Tamil Nadu, India

© Springer Nature Switzerland AG 2020 253


A. Haldorai et al. (eds.), Business Intelligence for Enterprise Internet of Things,
EAI/Springer Innovations in Communication and Computing,
https://doi.org/10.1007/978-3-030-44407-5_17
254 J. G. Thangam and A. Sankar

Link Prediction [4] is a noteworthy issue in social network. It is the problem of


predicting the existence of link between two entities. Since the social network is
vast, it is very difficult to predict the link at the earliest. So there is real urge to con-
sider about the efficiency of the algorithm in terms of time and space. This tech-
nique is to reduce the dimensions so that link can be rapidly predicted.
This chapter is organized as follows: the related works is presented in next sec-
tion, while the preliminaries, the problem description, and proposed work are pre-
sented in Sects. 17.3, 17.4, and 17.5. In Sect. 17.6, an experimental study is
conducted on real data sets. The social network centrality measures are described in
Sect. 17.7. The complexity analysis is represented in Sect. 17.8. The evaluation
metrics are analyzed in Sect. 17.9. Finally, Sect. 17.10 is conclusion and the
future work.

17.2 Related Works

A Hybrid Active Learning Link Prediction (HALLP) [5] has used machine learning
techniques for binary classification of links. D. Sharma [6] has made a test examina-
tion of the connection forecast procedures. Michael Fire et al. [7] have used topo-
logical feature (friend’s measure) for predicting the links. A novel strategy has been
proposed by Naveen Gupta et al. [8], for identifying the missing links by consider-
ing the uncommon neighbors. Manu Kurakar et al. [9] have identified complex dis-
eases based on the sequence similarity in protein networks.
A link prediction model has been proposed by Wal et al. [10], for identifying the
links that might appear and disappear in future. A friend recommendation has been
suggested by traversing all paths of limited length by Alexis et al. [11]. A Social
Attribute Network (SAN) approach proposed by Neil Zhengiang Gong et al. [12],
have considered the performance of supervised and unsupervised algorithms. A
novel approach proposed by Aditya Krishna Menon et al. [13] has addressed the
class imbalance problem by optimizing the ranking loss. Daniel et al. [14] have
proposed a tensor-based method for predicting the future links temporally.
Jichang Zhao et al. [15] evaluated the performance of link prediction based on
local information by using sampling techniques. Ryan N. Lichtenwalter [16] pro-
posed a machine learning algorithm with class imbalance. An algorithm for param-
eter reduction of soft binary relation was proposed and investigated by Guangh Yu
[17]. Xiuqin Ma [18] proposed a parameter reduction algorithm of soft set for deci-
sion making.

17.3 Preliminaries

The notion of a social network and the methods of SNA have attracted considerable
interest and curiosity from the social and behavioral science community in recent
decades. It is used to quantify the relationships among social entities and on the pat-
17 Emphasizing on Space Complexity in Enterprise Social Networks… 255

terns and implications of these relationships. Milgram’s small-world phenomenon


is revealed in social network. It states that the average path length between any two
nodes should be shorter. The social network is represented as a graph.

17.3.1 Graph

A graph G is defined as G = (V, E), where V is a set of vertices and E is a set of


edges. The notion of vertices and nodes, links and edges is used interchangeably in
the literature. The adjacent matrix is used for representing any graph. The adjacency
matrix of a graph G with n vertices is an n ∗ n binary matrix given by A = [aij]
defined as aij= 1 if the ith and jth vertices are adjacent, that is, there is an edge con-
necting the ith and jth vertices, and aij= 0, otherwise, that is, if there is no edge con-
necting the ith and jth vertices.
The graph shown in Fig. 17.1 is a simple undirected graph with set of vertices as
{A, B, C, D, E, F} and set of edges as {AB, AC, BF, BD, BC, CE}. An edge from a
node A to node B is called a path. It is of length 1 and known to be length-1 path. In
a similar manner, the path from A node D through node B is of length 2 which is
referred as length-2 path.

17.3.2 Link Prediction Problem

Link Prediction Problem [19, 20] is identifying a potential or possible link among
vertices or nodes in a network. It is defined as given a snapshot of a social network,
where new interactions among its members that are likely to occur in the near future
can be inferred. Predicting such links is not an easy task. Many similarity index-­
based link prediction methods have been based on nature of computing them, as
local structure, global structure, and quasi local structure. Friendlink algorithm is
based on Quasi-local structure measure. It uses path of length more than 2.

Fig. 17.1 Graph


A B
D

C
F

E
256 J. G. Thangam and A. Sankar

17.4 Problem Description

The link prediction is an issue for foreseeing whether the link will exist in future or
not. The similarity score between the nodes is calculated for predicting the social
links. Since the social network data set is enormous, there is a need for dimensional-
ity reduction, without affecting the consistency of the data set. Soft set theory is
used for dimensionality reduction. The reduced dimension is used for predicting the
future links. The similarity score between the nodes are calculated for the reduced
dimension. Based on this similarity score, ranking of all other nodes has to be done
with respect to a particular node. The future link is more probable to occur, if the
similarity score between the nodes is higher.

17.4.1 Soft Set

Molodtsov [21] soft set theory is an overview of fuzzy set theory [22] that deals with
uncertainty in a nonparametric manner. Let U be an initial universe and E be the set
of possible parameters. The power set of U is denoted by P(U). The Soft set is
defined as follows:
Definition A pair (F, E) is called a soft set over U, where F is a mapping of E into
P(U), that is,

F , E → P (U ) (17.1)

In a soft set theory, the reduction of parameter is a vital problem. It deals with the
elimination of frequent occurring parameters in order to take an optimal decision
making, without affecting the novelty of the data set, so that the number of param-
eters is reduced to simply the calculations. Hence the soft set is used to discover the
similar links between the nodes. These similar links are removed from data set that
yields in reducing the dimension. Thereby the space as well time complexities will
have a significant improvement.

17.4.2 Friendlink Algorithm

Friendlink algorithm [23] finds the similarities between nodes in a sociograph using
the adjacency matrix. Rubin’s algorithm [24] is used to find the paths of length l.
Definition The similarity score sim(ux, uy) between two nodes ux and uy is defined
as counting of paths of varying length l from ux to uy is as given below:
17 Emphasizing on Space Complexity in Enterprise Social Networks… 257

l
1 pathsukx ,uy
sim ( ux ,uy ) = ∑ ∗ (17.2)
k =2 k − 1
∏ m =2 ( n − m )
k

where n is the number of vertices in a graph, l is the maximum length of a path from
1
ux and uy. is an attenuation factor that weighs paths according to their length.
k −1 k
pathsukx ,uy is the number of all length-l paths from ux to uy. ∏ ( n − m ) is the number
m =2
of all possible length-l paths from ux to uy. k and m denote current iteration. For
example, if the two nodes are similar, the similarity score is expected to be close to
1. On the other hand, if the two nodes are not similar, the similarity score is expected
to be close to 0.
The Friendlink algorithm comprises of two functionalities, one is for computing
the path of length-2 between two nonexistent nodes and other is for computing the
similarities between them. Then the similarity scores will be ranked. Finally the
node with the highest similarity score will be declared as the node which is more
probable to be communicated in future. Hence nodes can be recommended to the
target node according to the similarity scores.

17.5 Proposed Work

Link prediction is a pivotal problem in SNA in order to be acquainted with associa-


tions between nodes in any social communities. The proposed approach is explained
in Fig. 17.2.
The Social Network data set is obtained as edge list for a particular network. The
edge list is then converted into an adjacency representation. The soft set is applied
and the preprocessed data is obtained. The Soft set and Friendlink-based Link
Prediction algorithm (SFLP) uses the Friendlink algorithm for the path length of 2
and 3. The similarity score is calculated for the original data set using the length-2
path and length-3 path. Ultimately, the similarity score of a node against all other
nodes are computed and similarity scores are ranked. The node with higher similar-
ity score is the node to occur in future which is predicted. The pseudo code for
ranking the nodes is provided in Algorithm 1.
258 J. G. Thangam and A. Sankar

Algorithm 1: Soft Set and Friendlink-Based Link Prediction (SFLP)


Algorithm
Input: Adjacency Matrix.
Output: Ranked prediction list
1. Reduce the dimension using soft set.
For each node nodei.
Find all the links such that aij> 0
Remove aij if the link is redundant
  End For
2. Compute similarity score sim(nodei, nodej) using Friendlink algorithm.
For each node nodei
If aij! = 1
Find the similarity scores with all other nodes.
End if
  End For
3. Rank the nodes.

The algorithm is first organized to reduce the dimensions of the dataset. The reduced
data set is used for finding the similarity score between the nodes. Then all the
similarity scores are ranked, and the node with the maximum similarity score value
will be the node to be linked in future. The similarity scores obtained using SFLP is
compared with the similarity score of Friendlink Link Prediction (FLP) algorithm.
Those results are analyzed against the social network centrality measures.

17.6 Experimental Setup and Results

The experiments are conducted on a 2.50-GHz Intel Dual core PC with 4GB RAM
running Microsoft 7 ultimate. The SFLP and FLP algorithms are implemented using
MATLAB. An examination is made to assess the exhibition of the proposed method.
To evaluate the efficiency of the algorithm, it is examined using real world data
set. The data set is loaded from UCI Network Data Repository. The Zachary’s
Karate Club data set is a social network of friendships of a karate club at a US uni-
versity. Each node represents a member of the club and each edge represents a tie
between two members of the club. It represents the presence (1) or absence (0) of
ties among the members of the club. Hence, the work is carried out to predict the
indirect ties that exist between the members of the club in future. From this, the
friendships of the karate club in future will be predicted.
17 Emphasizing on Space Complexity in Enterprise Social Networks… 259

Social Network dataset

Preprocessing

Dimension Reduction
using Soft set
Pre-processed
Dataset

Reduced Dimension

Similarity Score Calculation using Friendlink algorithm

Ranking

Link Prediction Result

Fig. 17.2 The proposed approach

For finding the strong friendship between the nodes, the similarity score between
two nodes before dimensionality reduction (FLP) is calculated using Friendlink
algorithm. It considers all indirect links of length-2 and length-3 path. Out of all the
nodes, some nodes are extracted and are shown in the Table 17.1.
In this table the zero represents that there is no link between the nodes with the
path length of length-2 path and length-3 path and a nonzero value represents that
the nodes are having a probability of some value to get linked in future. For exam-
ple, the node1 is having indirect link with many other nodes like node24 and node26.
But it is found that node1 is having highest similarity score with node26 when com-
pared to all other remaining nodes. It is concluded that node1 will have higher prob-
ability to link with node26 in future. The graphical representation for the values of
260 J. G. Thangam and A. Sankar

Table 17.1 Similarity scores of Friendlink Link Prediction


Nodes 9 24 26 32
1 0 0.000504 0.031754 0
2 0.03125 0 0 0
3 0 0.063004 0.000504 0.03125
4 0 0 0 0
5 0 0 0 0
6 0 0 0 0
7 0 0 0 0
8 0 0 0 0
9 0 0.03125 0.001008 0.063004
10 0.032258 0.001512 0.000504 0.031754
11 0 0 0 0
12 0 0 0 0
13 0 0 0 0
14 0.032258 0.001512 0.000504 0.031754
15 0.064012 0.03125 0.001008 0.063004
16 0.064012 0.03125 0.001008 0.063004
17 0 0 0 0
18 0 0 0 0
19 0.064012 0.03125 0.001008 0.063004
20 0.032258 0.001512 0.000504 0.031754
21 0.064012 0.03125 0.001008 0.063004
22 0 0 0 0
23 0.064012 0.03125 0.001008 0.063004
24 0.031754 0 0 0
25 0.032258 0.0625 0 0.031754
26 0 0.000504 0 0
27 0 0.03125 0 0
28 0.032258 0.001512 0.000504 0.031754
29 0.032258 0 0.031754 0.031754
30 0.064012 0.03125 0.001008 0
31 0.064012 0.03125 0.001008 0.063004
32 0.064012 0.03125 0.001008 0.063004
33 0.032258 0.001512 0.000504 0.031754
34 0 0 0 0

node1 against all other remaining nodes is shown in Fig. 17.3a. Similarly, the node15
is having links with other nodes such as node9 and node32, and it is found that node9
is having higher similarity score. Therefore node15 is having higher probability to be
linked with node9. The graphical representation for the values of node15 against all
other remaining nodes is shown in Fig. 17.3b.
Since the data set is immense, the dimensionality reduction technique is applied
for extracting the friendship without affecting the members in the karate club. The
17 Emphasizing on Space Complexity in Enterprise Social Networks… 261

a) Similarity Score performance


0.035
Similarity Score 0.03
0.025
0.02
0.015
0.01
0.005
0
0 10 20 30 40
Node

b) Similarity Score performance


0.07
0.06
Similarity Score

0.05
0.04
0.03
0.02
0.01
0
0 10 20 30 40
Node

Fig. 17.3 Similarity scores of Friendlink Link Prediction

similarity score after applying Soft set-based FLP (SFLP) algorithm is calculated
and is shown in Table 17.2.
It also considers all indirect links of length-2 and length-3 path. For example,
with reduced dimension, the node1 is having indirect link with many other nodes
such as node26 and node30. But it is found that node1 is having highest similarity
score with node26 with respect to all other remaining nodes even after the dimen-
sionality reduction technique is employed. So, it is concluded that the node1 will
have a chance to be linked with node26. The graphical representation for the values
of node1 against all other remaining nodes is shown in Fig. 17.4a. Similarly, the
node15 is having links with other nodes such as node9 and node26, and it is found that
node9 is having higher similarity score. Therefore node15 is having higher probabil-
ity to be linked with node9. Hence it is observed that the node that is having an
indirect link of length-2 and length-3 path length will be similar even after the
262 J. G. Thangam and A. Sankar

Table 17.2 Similarity scores of Soft set-based Friendlink Link Prediction


Nodes 9 26 30 31
1 0 0.094254 0.09375 0.001008
2 0.000504 0.0625 0 0
3 0 0 0.064012 0.032258
4 0.001008 0.03125 0.03125 0.001008
5 0 0.000504 0 0
6 0 0.001512 0 0
7 0 0.001008 0 0
8 0 0 0 0
9 0.03125 0.032762 0.03125 0
10 0.03125 0.001008 0.001512 0.031754
11 0 0 0 0
12 0 0 0 0
13 0 0 0 0
14 0 0.001008 0.001512 0.031754
15 0.063508 0.032762 0.001008 0.03125
16 0.03125 0.032762 0.001008 0.063508
17 0 0 0 0
18 0 0 0 0
19 0.03125 0.032762 0.001008 0.063508
20 0.03125 0.001008 0.001512 0.031754
21 0.03125 0.032762 0.001008 0.063508
22 0 0 0 0
23 0.03125 0.032762 0.001008 0.063508
24 0 0 0 0.031754
25 0.03125 0 0.001512 0.031754
26 0 0 0 0
27 0 0 0 0
28 0.03125 0.001008 0.001512 0.031754
29 0.03125 0.001008 0.001512 0.031754
30 0.03125 0.032762 0.001008 0.063508
31 0.03125 0.032762 0.001008 0.063508
32 0.03125 0.032762 0.001008 0.063508
33 0.03125 0.001008 0.001512 0.031754
34 0 0 0 0

dimensionality reduction. The graphical representation for the values of node15


against all other remaining nodes is shown in Fig. 17.4b.
So, it is observed that the node that will occur in future is same for both the algo-
rithms. Finally the consistency of the data set is preserved. The running time for
executing the FLP algorithm is 0.1525 s and SFLP is 0.0147 s. It is found that the
approach with dimensionality reduction technique works significantly better.
17 Emphasizing on Space Complexity in Enterprise Social Networks… 263

a) Similarity Score performance


0.1
Similarity Score 0.08
0.06
0.04
0.02
0
0 10 20 30 40
Node

b) Similarity Score performance


0.07
0.06
Similarity Score

0.05
0.04
0.03
0.02
0.01
0
0 10 20 30 40
Node

Fig. 17.4 Similarity scores of Soft set Friendlink Link Prediction

17.7 Centrality Measures

In SNA, the graph theory is exceptionally essential to discover the “central” actors.
The graph theoretic thoughts measure an individual actor’s prominence quality
depending on centrality and prestige. An actor is said to be prominent if the ties of
the actor make the actor particularly noticeable to the other actors in the network.
Centrality is a measure of the importance of a node in a network. A prestigious actor
is characterized as one who is the object of broad ties, thus concentrating exclu-
sively on the actor as a recipient. Obviously, prestige is a more refined idea than
centrality and can’t generally be estimated. So, the centrality measures are consid-
ered to be significant. The commonly used centrality measures are degree, close-
ness, betweenness, eigenvector centrality, and clustering coefficient.
264 J. G. Thangam and A. Sankar

17.7.1 Degree Centrality

The degree centrality can be of, in-degree centrality, and out-degree centrality. The
in-degree centrality is defined as an actor who receives many ties, and considered as
prominent. The out-degree centrality is defined as an actor who disperses the infor-
mation quickly to many others in a network and is categorized as influential. The
degree centrality index [25] is given by

C D ( ni ) = d ( ni ) (17.3)

where CD is the degree centrality for the ith node denoted by ni, and d(ni) is the
degree of the node ni. This measure focuses on the most visible actors in the network.

17.7.2 Closeness Centrality

The closeness centrality is a measure of the degree to which an individual is close


to all other individuals in a network. The closeness centrality is computed by using
−1
 g 
CC ( ni ) =  ∑d ( ni ,n j )  (17.4)
 j =1 

where CC is the closeness centrality for the ith node denoted by ni, d(ni, nj) is the
g
length of any shortest path from ni to nj, ∑d ( n ,n )
j =1
i j is the total distance from all

other nodes to node i. This measure depends on the whole of the geodesic distances
from one another to all others. However, it is ambiguous for complicated graphs.

17.7.3 Betweenness Centrality

The betweenness centrality [26] is a measure of the extent to which a node is con-
nected to other nodes that are not connected to each other. The betweenness central-
ity index is computed by using

C B ( ni ) = ∑g jk ( ni ) / g jk (17.5)
j<k

where CB is the betweenness centrality for the ith node denoted by ni , gjk(ni) is the
number of geodesics linking two nodes j and k that contain the node i. This measure
17 Emphasizing on Space Complexity in Enterprise Social Networks… 265

estimates the degree of gatekeeping for every pair of actors in the network, concen-
trating on how much gatekeeping the subsequent actor accomplishes for the first.

17.8 Eigenvector Centrality

The eigenvector centrality is used to quantify the relative importance of a node in a


network. The eigenvector centrality is calculated using

1 i
CE ( ni ) = ∑aij ∗ x j (17.6)
λ n =1

where CE is the eigenvector centrality for the ith node denoted by ni, λ is a constant,
aij is equal to 1 if the ith and jth vertices are adjacent, that is, there is an edge connect-
ing the ith and jth vertices.aij is equal to 0 otherwise, that is, if there is no edge con-
necting the ith and jth vertices.

17.8.1 Clustering Coefficient

The clustering coefficient is the measure of the degree to which a node in a network
tends to cluster together. It is also known as local clustering coefficient. The cluster-
ing coefficient is calculated using

number of pairs of neighbors connectedbyedges


CuCo ( ni ) = (17.7)
number of paairs of neighbors

where CuCo is the clustering coefficient for the ith node denoted by ni. This measure
has become popular in Nature journal by Watts and Strogatz [27]. The centrality
measures are computed using the open source tools for the original data set and is
tabulated in Table 17.3.
It is observed that node34 and node1 are having high degree centrality and is con-
sidered as the prominent as well the most influential node in the Zachary’s Karate
Club data set. The node1 has high betweenness centrality which means that through
node1 only most of the nodes are connected to other nodes in a network. The node1
will be the gatekeeper for every pair of nodes in a network. The degree centrality of
node1 is same as that of the node that is having high betweenness centrality. So, the
degree centrality is directly proportional to the betweenness centrality.
The closeness centrality for the nodes such as node15 and node23 is higher, thereby
indicating that those nodes are very close to all other nodes in a network. The node34
is having higher eigenvector value which shows that the node is relatively important
266 J. G. Thangam and A. Sankar

Table 17.3 Centrality measure calculation of Friendlink Link Prediction


Degree Betweenness Closeness Eigenvector Clustering
Vertex Centrality Centrality Centrality Centrality Coefficient
1 8 0.852 2.200 0.221 0.143
2 5 0.042 2.467 0.165 0.300
3 8 0.423 2.033 0.277 0.143
4 5 0.030 2.600 0.143 0.350
5 1 0.000 3.933 0.007 0.000
6 2 0.000 3.900 0.008 0.500
7 4 0.523 2.967 0.041 0.083
8 4 0.000 2.633 0.134 0.500
9 5 0.249 2.033 0.263 0.250
10 1 0.000 2.800 0.076 0.000
13 2 0.000 3.133 0.060 0.500
14 1 0.000 2.800 0.076 0.000
15 2 0.000 2.600 0.144 0.500
16 2 0.000 2.600 0.144 0.500
17 2 0.000 3.900 0.008 0.500
19 2 0.000 2.600 0.144 0.500
20 1 0.000 2.800 0.076 0.000
21 2 0.000 2.600 0.144 0.500
23 2 0.000 2.600 0.144 0.500
24 4 0.191 2.600 0.128 0.083
25 4 0.048 2.300 0.156 0.250
26 2 0.000 2.867 0.067 0.500
27 1 0.000 3.433 0.028 0.000
28 4 0.113 2.167 0.169 0.083
29 3 0.017 2.300 0.162 0.167
30 4 0.201 2.467 0.170 0.167
31 4 0.073 2.233 0.215 0.250
32 6 0.421 1.967 0.245 0.133
33 12 0.708 1.867 0.412 0.091
34 16 1.000 1.833 0.456 0.054

Table 17.4 Topological properties of the real data set for Friendlink Link Prediction
UE EWD TE SL V GD
78 0 78 0 34 0.069518

with respect to other nodes in a network. The clustering coefficient for some of the
nodes like node15, node8, node19 that have higher values shows the property of one’s
friends are also friends of each other. Also, degree centrality is inversely propor-
tional to the closeness centrality. It is concluded that nodes having low degree and
betweenness centrality can be removed. The density of the graph G is the ratio of
17 Emphasizing on Space Complexity in Enterprise Social Networks… 267

Table 17.5 Centrality measure calculation of Soft set Friendlink Link Prediction
Degree Betweenness Closeness Eigenvector Clustering
Vertex Centrality Centrality Centrality Centrality Coefficient
1 16 1.000 1.758 0.356 0.075
2 9 0.123 2.061 0.266 0.167
3 10 0.328 1.788 0.317 0.122
4 6 0.027 2.152 0.211 0.333
5 3 0.001 2.636 0.076 0.333
6 4 0.069 2.606 0.079 0.250
7 4 0.069 2.606 0.079 0.250
8 4 0.000 2.273 0.171 0.500
9 5 0.128 1.939 0.227 0.250
10 2 0.002 2.303 0.103 0.000
11 3 0.001 2.636 0.076 0.333
12 1 0.000 2.727 0.053 0.000
13 2 0.000 2.697 0.084 0.500
14 5 0.105 1.939 0.226 0.300
15 2 0.000 2.697 0.101 0.500
16 2 0.000 2.697 0.101 0.500
17 2 0.000 3.515 0.024 0.500
18 2 0.000 2.667 0.092 0.500
19 2 0.000 2.697 0.101 0.500
20 3 0.074 2.000 0.148 0.167
21 2 0.000 2.697 0.101 0.500
22 2 0.000 2.667 0.092 0.500
23 2 0.000 2.697 0.101 0.500
24 5 0.040 2.545 0.150 0.200
25 3 0.005 2.667 0.057 0.167
26 3 0.009 2.667 0.059 0.167
27 2 0.000 2.758 0.076 0.500
28 4 0.051 2.182 0.133 0.083
29 3 0.004 2.212 0.131 0.167
30 4 0.007 2.606 0.135 0.333
31 4 0.033 2.182 0.175 0.250
32 6 0.316 1.848 0.191 0.100
33 12 0.332 1.939 0.309 0.098
34 17 0.695 1.818 0.374 0.055

edges in G to the maximum number of edges. The graph density is computed for the
original data set and shown in Table 17.4. The topological properties are Unique
Edges (UE), Edges With Duplicates (EWD), Total Edges (TE), Self-Loops (SL),
Vertices (V), and Graph Density (GD).
The centrality measures are computed for the data set for which the soft set the-
ory is applied and finally the reduced dimension data set is obtained. The centrality
measure is tabulated in Table 17.5.
268 J. G. Thangam and A. Sankar

Table 17.6 Topological properties of the real data set for Soft set Friendlink Link Prediction
UE EWD TE SL V GD
60 0 60 0 34 0.064516

The degree for node34 and node1 is higher, so the original data set node34 and
node1 are prominent and influential nodes. The betweenness centrality for the node1
is higher which is same as that of the centrality measures obtained for the original
data set. The remaining metrics are also the same. Hence, the centrality measures
for both original data set and reduced dimension are compared. It is agreed that the
node having low degree and betweenness centrality is removed without affecting
the consistency of the data set, so it improves the memory space utilization. The
graph density is computed and shown in Table 17.6.
It is observed that the graph density is compressed for the reduced dimension.

17.9 Theoretical Analysis

The proposed work comprises of three tasks, the first task is to construct a reduced
data set and the time complexity for constructing is Ο(n + e), where n is the nodes
and e is the number of edges, always n < e. The second task is to find similarity
score using Friendlink algorithm and the time complexity is Ο(n ∗ a), where n is the
number of nodes and a is the average degree of the network. So, the time complexity
of these two tasks would be Ο(max(n + e, n ∗ a)), since the average degree is less
than the number of edges, the time complexity would be Ο(n + e). The third task is
to rank all the similarity scores and the time complexity is Ο(n log n). Hence, the
total time complexity of the proposed algorithm would be Ο(max(n + e, n log n)),
since the number of nodes is comparatively less than the number of edges, the time
complexity would be Ο(n log n) and the space complexity would be Ο(n1 ∗ a1),
where a1 is the average node degree and n1 is number of nodes after dimensionality
reduction which is strictly less than the number of nodes present in the original data
set. The proposed technique significantly improves the performance of link predic-
tion compared to Katz [28] index, since it needs the inverse matrix representation
for further computation. For implementation, the adjacent list representation is used
for storing the adjacent nodes in a graph.

17.10 Empirical Analysis

Most standard evaluation metric is Mean Average Precision (MAP), which provides
a single-figure quality measure. It is defined by
17 Emphasizing on Space Complexity in Enterprise Social Networks… 269

i
MAP = ∑AveP ( n ) (17.8)
n =1

where n is the total number of nodes, i denotes the iteration, AveP(n) is the average
precision value. It has been shown to have especially good discrimination and sta-
bility among the various metrics. So, the link prediction accuracy is evaluated using
MAP for both algorithms. The MAP value for SFLP is 0.38853 and FLP is 0.07865.
The empirical analysis shows that MAP value is higher for SFLP than FLP
algorithm.

17.11 Conclusion and Future Work

Social networks play a significant role in all domains, in which the data set is very
large. This chapter concentrates on soft set theory that has been applied for dimen-
sionality reduction. The Friendlink algorithm is used for finding the similarity score
of a node with other nodes. Both the Soft set theory and Friendlink algorithm are
applied for the real data set in order to predict the node that will occur in future.
Then the consistency of the data is verified against the various social network met-
rics. The experimental results show that the approach works better in terms of space.
Further improvement might be predicting the links that can be applied to various
domains like community detection, name disambiguation, and health care.

References

1. Wasserman, S., & Faust, K. (1994). Social network analysis: Methods and applications (Vol.
8). Cambridge: Cambridge University Press.
2. Song, Han Hee, Tae Won Cho, Vacha Dave, Yin Zhang, and Lili Qiu. (2009) Scalable proxim-
ity estimation and link prediction in online social networks. In Proceedings of the 9th ACM
SIGCOMM conference on Internet measurement conference, ACM (pp. 322–335).
3. Aldous, J. M., & Wilson, R. J.. (2003) Graphs and applications: An introductory approach
(Vol. 1). Springer.
4. Liben-Nowell, D., & Kleinberg, J. (2007). The link-prediction problem for social networks.
Journal of the Association for Information Science and Technology, 58(7), 1019–1031.
5. Chen, K.-J., Han, J., & Li, Y. (2014). HALLP: A hybrid active learning approach to link predic-
tion task. JCP, 9(3), 551–556.
6. Sharma, D., Sharma, U., & Khatr, S. K. (2014). An experimental comparison of the link pre-
diction techniques in social networks. International Journal of Modeling and Optimization,
4(1), 21.
7. Michael, F., Tenenboim-Chekina, L., Puzis, R., Lesser, O., Rokach, L., & Elovici, Y. (2013).
Computationally efficient link prediction in a variety of social networks. ACM Transactions on
Intelligent Systems and Technology (TIST), 5(1), 10.
8. Gupta, N., & Singh, A.. (2014). A novel strategy for link prediction in social networks. In
Proceedings of the 2014 CoNEXT on Student Workshop, ACM (pp. 12–14).
270 J. G. Thangam and A. Sankar

9. Kurakar, M., & Izudheen, S. (2014). Link prediction in protein-protein networks. Survey:
International Journal of Computer Trends & Technology, 1(9), 164–168.
10. Almansoori, Wadhah, Shang Gao, Tamer M. Jarada, Reda Alhajj, & Jon Rokne. (2011) Link
prediction and classification in social networks and its application in healthcare. In Information
Reuse and Integration (IRI), 2011 IEEE International Conference on, IEEE, pp. 12–14.
11. Papadimitriou, A., Symeonidis, P., & Manolopoulos, Y. (2012). Fast and accurate link predic-
tion in social networking systems. Journal of Systems and Software, 85(9), 2119–2132.
12. Gong, N. Z., Talwalkar, A., Mackey, L., Huang, L., Shin, E. C. R., Stefanov, E., & Song, D.
(2012). Jointly predicting links and inferring attributes using a social-attribute network (san).
ACM, 1–9.
13. Menon, A. K., & Elkan, C. (2011). Link prediction via matrix factorization. In Joint European
conference on machine learning and knowledge discovery in databases (pp. 437–452). Berlin
Heidelberg: Springer.
14. Dunlavy, D. M., Kolda, T. G., & Acar, E. (2011). Temporal link prediction using matrix and
tensor factorizations. ACM Transactions on Knowledge Discovery from Data (TKDD), 5(2),
10.
15. Zhao, J., Xu, F., Dong, L., Liang, X., & Xu, K. (2012). Performance of local information-based
link prediction: A sampling perspective. Journal of Physics A Mathematical and Theoretical,
45(34), 345001.
16. Lichtenwalter, R. N., Lussier, J. T., & Chawla, N. V. (2010). New perspectives and meth-
ods in link prediction. Proceedings of the 16th ACM SIGKDD international conference on
Knowledge discovery and data mining ACM, 243–252.
17. Yu, G. (2015). On the parameter reduction of soft binary relations. Annals of Fuzzy Athematics
and Informatics, 1–10.
18. Ma, X., & Qin, H. (2014). Application of a new efficient Normal parameter reduction algo-
rithm of soft sets in online shopping. World Academy of Science, Engineering and Technology,
International Journal of Computer, Electrical, Automation, Control and Information
Engineering, 8(7), 1174–1176.
19. Liben-Nowell, D., & Kleinberg, J. The link-prediction problem for social networks. Journal
of the American Society for Information Science and Technology, 58(7), 1019–1031.
20. Yu, C., Zhao, X., Lu, A., & Lin, X. (2017). Similarity-based link prediction in social networks:
A path and node combined approach. Journal of Information Science, 43(5), 683–695.
21. Molodtsov, D. (1999). Soft set theory—First results. Computers & Mathematics with
Applications, 37(4–5), 19–31.
22. Zimmermann, H.-J. (2011). Fuzzy set theory—And its applications. Springer.
23. Papadimitriou, A., Symeonidis, P., & Manolopoulos, Y. (2011). Friendlink: link prediction in
social networks via bounded local path traversal. In Computational Aspects of Social Networks
(CASoN) (pp. 66–71). 2011 International Conference on, IEEE.
24. Rubin, F. R. A. N. K. (1978). Enumerating all simple paths in a graph. IEEE Transactions on
Circuits and Systems, 25(8), 641–642.
25. Baagyere, E. Y., Qin, Z., Xiong, H., & Zhiguang, Q. (2016). The structural properties of
online social networks and their application areas. IAENG International Journal of Computer
Science, 43(2).
26. Freeman, L. C. (1977) A set of measures of centrality based on: (1977): 35- betweenness.
Sociometry 41.
27. Watts, D. J., & Strogatz, S. H. (1998). Collective dynamics of ‘small-world’ networks. Nature,
393(6684), 440.
28. Katz, L. (1953). A new status index derived from sociometric analysis. Psychometrika, 18(1),
39–43.
Chapter 18
Overview on Deep Neural Networks:
Architecture, Application and Rising
Analysis Trends

V. Niranjani and N. Saravana Selvam

18.1 Introduction

Alongside Big Data [1], the subject of deep education has become a dominant sub-
ject for the industry and research fields for the growth of different intelligent sys-
tems, Cloud Computing [2] and the Internet of Things (IoT) [3]. The ability to
estimate and reduce comprehensive, complicated data sets into extremely precise,
transformative output has demonstrated an important potential [4, 5]. Conversely,
deep neural networks (DNN) architectures may be applied to all kinds of informa-
tion—whether numerical, visual, texts, audio, or a certain mixture—as opposed to
complicated hard-coded programs for one inflexible job only. In addition, sophisti-
cated, often open-source and widely accessible advanced, deep learning platforms
become increasingly accessible. Moreover, significant businesses such as Amazon,
Flipkart, Microsoft, Apple, and others are investing strongly in profound learning
technology.
Supervised classification tasks have outstripped human capacities in fields like
manuscript and picture recognition, one in each of the key fields for deep learning
applications [6]. Furthermore, unattended data set learning without specific labels
have shown the possibility of extraction by means of the clustering and statistical
analysis of unpredicted assessment and business value. Most potentially attractive
yet is strengthened learning, providing feedback from a linked setting that offers the
opportunity for deep learning without human oversight. The field of robotics and
computer view was widely used for this sort of profound learning.

V. Niranjani (*)
Sri Eshwar College of Engineering, Coimbatore, India
N. S. Selvam
PSR Engineering College, Sivakasi, India

© Springer Nature Switzerland AG 2020 271


A. Haldorai et al. (eds.), Business Intelligence for Enterprise Internet of Things,
EAI/Springer Innovations in Communication and Computing,
https://doi.org/10.1007/978-3-030-44407-5_18
272 V. Niranjani and N. S. Selvam

As IoT and smart-world systems are steadily growing, powered by CPS (cyber-
physical system) technologies in which networking in all devices and capable of
communicated sensitive information and physical items monitor, bigger and bigger
datasets are accessible for profound learning which can have a material effect on our
everyday life [6, 7–9]. For example, smart phones have provided the ability to create
rich video, audio, and text data through integrated GPS modules from various media
apps and integrated sensors, as well as massively locating population movement.
All such apps are clearly generated, alone, or together, by unprecedented Big Data.
Intelligent cities are designed to allow resource management in almost all fields like
communications, electricity, transportation, emergency, and other utilities to build-
ing. Intelligent transport systems will interconnect auto-drive cars and infrastruc-
ture networks to revolutionize day-to-day mass transit, eliminate accidents, and
facilitate the storage of the secondary grid.

18.2 Overview of Deep Neural Networks

Machine learning (ML) includes a wide range of algorithms, which cannot all be
categorized as deep learning. Specific algorithms like Bayesian algorithms are
restricted in implementation and in the capacity to learn massive complicated infor-
mation representations, including statistical processes such as linear-regression or
decision-making trees. The artificial intelligence field is primarily when machines
can perform functions that typically require human intelligence [10]. It includes
machine learning, where machines can obtain abilities through experience and
without participation from human beings [11]. Deep learning is a component of
machine learning that is used to learn from big quantities of information by artificial
neural networks, human brain-inspired algorithms. Similar to how we learn from
experience, each time the deep learning algorithm tweaks a job a bit in order to
enhance the result. We are talking about ‘profound learning’ because neural net-
works have different (profound) levels that allow learning to take place [12].
The quantity of information we produce each day—currently estimated at 2.6
quintillion octets—is astonishing and is the resource that enables profound learning.
Since profound learning algorithms require a ton of information to draw on, this rise
in information development is one reason why in the latest years profound learner
skills have grown. In relation to further development of information and the prolif-
eration of artificial intelligence (AI), deep learning algorithms profit from today’s
greater computing power. AI as the service has offered smaller organizations access,
especially the AI algorithms needed for deep learning without a big one, to artificial
intelligence technology shown in Fig. 18.1.
18 Overview on Deep Neural Networks: Architecture, Application and Rising… 273

Fig. 18.1 Diagram of ML vs. DL

18.2.1 Deep-Neural Network Architecture

Three significant layers are input layer, hidden layers, and output layers for deep-­
learning architectures. The number of layers concealed determines the architectural
depth. Various nonlinear features can be learned depending on the sort of concealed
layers used. Equations (18.1) and (18.2) are straightforward ANN patterns. In Eq.
(18.1) the nonlinearity observed within the information (concealed layer) is where
the activation function determines the nonlinearity features of the model. The same
applies to Eq. (18.2) (Output layer) which translates the feature of nonlinearity into
a forecast. An easy model for deep learning has several hidden layers.

AC1 = Act ( Z1A + C1 )


(18.1)

Ŷ = Z 2 AC1 + C2 (18.2)

18.2.2 Activation Function

The hidden layer activation function helps to map the nonlinearity connection
between input and output. In concealed layers, most used activation functions are
sigmoid and hyperbolic (Tanh). Specific activation functions are not applicable. For
datasets, various activation functions must be assessed. In this document, we use the
rectified linear activation function (also known as “relu”) (see Eq. (18.3)), since the
techniques of the gradation optimization facilitate model training [13].
274 V. Niranjani and N. S. Selvam

0, x < 0
f ( x) = {
x, x ≥ 0 (18.3)

18.2.3 Hidden Layer

Equation (18.1) is a straightforward concealed layer. A number of parameters are


used to define the complexity of each hidden layer. The number of parameters that
are the number of concealed units and the regularization parameter in L2 are gov-
erned by hyper parameters. The number of hidden parameters in each layer (i.e., the
weights) and the regularization of L2 reduce the extent of the parameters in order to
avoid over fitting [13]. To ensure a good model fit, tuning hyper parameters is
essential.

18.2.4 Outer Layer

Typically, the output comprises inactivation of an identity (also known as linear


activation) for regression issues. Activation of identity allows for adverse projec-
tions. As the demand for cooling and heating energy is non-negative reaction values
with a minimum “zero,” identification and activation for this implementation is not
appropriate. The linear activation function is used rather than rectified in Fig. 18.2.

Fig. 18.2 Diagram of deep learning architecture


18 Overview on Deep Neural Networks: Architecture, Application and Rising… 275

18.3 Deep-Neural Network Architecture

18.3.1 Supervised Learning

Supervised learning (SL) is named because of the need for consistent marking of the
studied data and for tracking or classifying the study results as accurate or incorrect.
Supervised learning(SL) is, in particular, used as a mechanism of prediction where
some of the data are learned (also known as training-set), a second part to validate a
model which are already trained (cross-validation), and the remaining data are used
to the predict the accuracy and effectiveness. While accuracy is a significant mea-
sure, the ability of a trained model to generalize to new information is used through
other statistically significant processes, such as precision, recall, and F1 scores.
Classification and regression are the two main tasks for monitored learning.

18.3.2 Unsupervised Learning

In unsupervised learning, datasets are not marked in any manner that identifies a
right or improper outcome, given as input for machine learning. Rather, the out-
come may reach a larger required objective, the capacity to find something easy to
understand by humans or a complicated use of a statistical function to obtain the
required value may be assessed. For example, the clustering algorithm may cluster-­
data or fugitive groups, but it cannot be simple to tell if the clustering is actually
correct without an appropriate visual representation. Similarly, the density estimate
provides only for an estimate to be used for compression or a decrease in dimension
that is either relevant or irrelevant to a dataset or to encode information effectively.
However, it may still be necessary to find the extraction ability of compressed
images accurately to establish the adequacy of implementation.

18.3.3 Reinforcement Learning

Reinforcement Learning is an intermediate between monitored and uncontrolled


learning, although information is not specifically labeled, there is a reward for the
performance in each action. In particular, the architecture of this learning is
enhanced learning that interacts directly with the setting, thus providing a particular
reward to a changing setting. In order to enlarge the benefit to all the state transi-
tions, the enhanced education scheme aims to learn the best measures to be taken at
every state. This can be done for an endless moment or can be implemented in meet-
ings to maximize the results of each session through a perception-action-learning
loop. Furthermore, the feedback can either be provided straight from the setting or,
276 V. Niranjani and N. S. Selvam

Table 18.1 Different categorization of deep learning


Supervised learning Unsupervised learning Reinforcement learning
Definition Both predictors and Training package has In every assignment they can
predictions exist for only data set predictors. produce cutting-edge outcomes.
the training set.
Algorithm SVM, Linear Dimensionality Q-Learning, SARSA, DQN
regression and logistic reduction, K-Means,
regression, Naive Clustering
Bayes
Uses Forecasting Pre-process the data, Warehouses, Inventory
pre-train supervised management, delivery
learning algorithms. management, Power system,
Financial systems.

as a consequence of some calculation or function, by way of numerical counter in


an interactive setting shown Table 18.1.
The search policy and approximation function can be divided between the two
main means of enhancement. Policy search can be performed using gradient-based
(back propagation) or gradient-free (evolutionary) techniques to search for an ideal
policy directly. Value function is a technique through which an evaluation of the
expected return of a particular state and the selection of an ideal strategy to select an
action to maximize the desired value are required for each state action.

18.4 Applications of Deep Neural Networks

Here we explore the primary applications of deep learning. An important body of


job has continuously evolved over the past few years towards the implementation of
deep learning. In particular, the main advances were in applying deep neural net-
works to analyze the multimedia data including image, video, audio, and NLP that
resulted in significant state-of-the-art leaps for all systems. In fact, ML is primarily
worried about information fitting, whose main uses are discrimination, prediction,
and optimization. In addition, the progress made in Big Data and the Cloud
Computing sector has created a chance for machine learning to flourish, which
enables the information collection, dissemination, and computational model perfor-
mances. The presence of the information and the nature of their potential immedi-
ately required more precise, widespread, and effective processes of learning.

18.4.1 Fraud Detection

The financial and banking industry is an area that benefits from deep learning and is
affected by the job of detecting fraud with digital cash transactions. Fraud detection
is performed on the basis of identification of patterns and credit scores for client
18 Overview on Deep Neural Networks: Architecture, Application and Rising… 277

operations, anomalous conduct, and outlines. Auto-encoders are created for credit-­
card-­fraud in Tensor flow and Keras that save billions of USD of financial institution
recapture and insurance price. Machine learning methods and neural networks are
used for fraud detection. Mainly machine learning is used to highlight cases of fraud
which require human deliberation; deep learning is intended to minimize such
efforts.

18.4.2 Diagnostics Medical

Highly affected by advances in image analysis, the fast improvements in deep learn-
ing have greatly benefited medical diagnostics. Considerable research has been per-
formed to improve the detection of pictures of MRI, tumors, CT scans, illnesses,
and other abnormalities. Furthermore, the devices of IoT for medical application are
providing independent patient surveillance and to extract helpful medical popula-
tion information.

18.4.3 Self-Driving Car

Deep learning gives life to self-employed movement. One million set of information
are supplied in a scheme for model building, machine training, and ensuring that
outcomes are tested in secure setting. Autonomous car developers are concerned
with handling scenarios that are unprecedented. Typical to deep learning algorithms,
a periodic cycle of testing and execution ensures secure driving and increasing
exposure to millions of situations. Dash cams, geo-mapping, and sensor data help to
generate brief and advanced models for navigating through traffic, identifying paths,
signage, pedestrian-only routings, etc.

18.5 Summary

Deep learning is a technology that continues to evolve and has been implemented to
great effect in a multitude of applications and domains. Deep learning methods are
practical for us to fix a lot of issues. Although full-scale implementation of DL tech-
nology in industry is ongoing, calculated steps must be taken to ensure proper appli-
cation of deep learning, as the subversion of deep learning models can lead to
significant loss of financial value, confidence, or even extreme existence. We ana-
lyzed in detail deep learning architectures based on learning mechanisms (super-
vised, unsupervised, and reinforced) and target output models. In addition, in deep
learning science, we have thoroughly investigated the state of the art. Such areas
include the processing of multimedia (text, audio, and video), automated systems,
278 V. Niranjani and N. S. Selvam

medical diagnostics, financial applications, and security analysis. We hope that this
work offers a useful reference for both scientists and computer science profession-
als in considering deep learning methods, and application, and causes interest in
fields that urgently need further consideration.

References

1. Liang, F., Yu, W., An, D., Yang, Q., Fu, X., & Zhao, W. (2018). A survey on big data market:
Pricing, trading and protection. IEEE Access, 6, 15132–15154.
2. Huang, H. H., & Liu, H. (2014, October). Big data machine learning and graph analytics:
Current state and future challenges. In Proceedings of IEEE international conference on Big
Data (Big Data) (pp. 16–17).
3. Yu, W., et al. (2017). A survey on the edge computing for the internet of things. IEEE Access,
6, 6900–6919.
4. Chen, X. W., & Lin, X. (2014). Big data deep learning: Challenges and perspectives. IEEE
Access, 2, 514–525.
5. Nguyen, N. D., Nguyen, T., & Nahavandi, S. (2017). System design perspective for human-­
level agents using deep reinforcement learning: A survey. IEEE Access, 5, 27091–27102.
6. Nielsen, M.. (2017). Neural networks and deep learning. [Online]. Available: http://neuralnet-
worksanddeeplearning.com/.
7. Xu, H., Lin, J., & Yu, W. (2017). Smart transportation systems: Architecture, enabling tech-
nologies, and open issues (pp. 23–49). Singapore: Springer.
8. Wu, J., & Zhao, W. (2016, November). Design and realization of internet: From net of things
to internet of things. ACM Transactions on Cyber-Physical Systems, 1(1):1–2.
9. Ekedebe, N., Lu, C., & Yu, W. (2015, June). Towards experimental evaluation of intelligent
transportation system safety and traffic efficiency. In Proceedings of International Conference
on Communications (ICC) (pp. 3757–3762).
10. Shi, W., Cao, J., Zhang, Q., Li, Y., & Xu, L. (2016, October). Edge computing: Vision and
challenges. IEEE Internet of Things Journal, 3(5): 637–646.
11. Arulkumaran, K., Deisenroth, M. P., Brundage, M., & Bharath, A. A. (2017, November).
Deep reinforcement learning: A brief survey. IEEE Signal Processing Magazine, 34, no. 6,
pp. 26–38.
12. Fadlullah, Z. M. et al. (2017). State-of-the-art deep learning: Evolving machine intelligence
toward tomorrow’s intelligent network traffic control systems. IEEE Communications Surveys
and Tutorials, 19(4): 2432–2455, 4th Quart.
13. Goodfellow, I., Bengio, Y., Courville, A. (2016). Deep Learn. Google Scholar
Index

A multi-class model, 108


Access-level attacks, 84 negative binomial regression
AdaBoost algorithm, 106 models, 106
Adaptive FPSO (AFPSO), 179–180 network communication data, 109
Advanced machine learning algorithms network traffic analysis, 108
activities, 118 non-linear pattern, 106
application physical sensor devices, 107
architecture diagram, 108 state-of-the-art techniques, 105
artificial intelligence, 109 stress behavior, 109
artificial neural network algorithm, 106 supervised learning, 103
automobile insurance agency, 106 time-series data, 108
big data, 109 unsupervised learning, 103
business enterprise, 107 web services, 109
classification and prediction BPM, 101
models, 106 business processes, 100, 101
classification model, 109 embedding computing devices, 99
cloud-based architecture, 110 enhancing customer value, 117
cluster-based negative binomial enterprise forecasting, 101–103
regression method, 106 enterprise IoT (E-IoT), 99, 100
clustering algorithms, 106 enterprise supply chain, 115, 116
data analytics, 103 ERP software, 100
dataset, 104, 105 heterogenous business data, 100
decision making and planning, 110 human intervention, 99
decision support system, 109 inter-connection, 99
development, 104 real-time stream analytics, 100
domains, 105, 107 sensors and actuators, 99
E-IoT, 108 technologies, 99
healthcare insurance agencies, 106 Advanced message queuing protocol
hidden Markov method, 109 (AMQP), 85
integrated system model, 109 Adversary location attack, 84
Internet, 110 Agent-based technologies, 82
Internet of Things, 110, 111 ANN controller, 217
IoMT, 110 Ant colony optimization (ACO) algorithm
IoT sensors, 105, 107 ABC algorithm, 174
irrigation field, 109 aim, 174

© Springer Nature Switzerland AG 2020 279


A. Haldorai et al. (eds.), Business Intelligence for Enterprise Internet of Things,
EAI/Springer Innovations in Communication and Computing,
https://doi.org/10.1007/978-3-030-44407-5
280 Index

Ant colony optimization (ACO) algorithm (cont.) definition, 151


AS (see Ant system (AS)) dynamic networks, 166
bin-packing algorithm, 174 healthcare, 140
inclusive behavior, 174 huge data handling, 153
path forming, 174 innovative approach, 139
stagnation, 175 IoT research
Ant system (AS) advancement, 164
complexity, 175 communication protocols, 164
elitist, 177 devices, 165
extended versions, 175 information gathering, 164
max-min, 177 infrastructure, 165
rank-based, 177 issues, 165
travelling algorithm steps, 175 low-power-consumption devices, 165
travelling salesman, 175 outdoor-computing facility, 164
Apache Spark cluster, 109 smart objects, 165
Application layer, 237 standardized protocols, 165
Application layer threats, 39 traffic management, 164
Apriori algorithm, 144 web services, 165
Arduino server, 235 machine learning, 154
Artificial intelligence, 153 management, 140
Artificial intelligence field, 272 opportunistic data distribution, 162–163
Artificial neural network algorithm, 106 Big Data methods, 74
Artificial neural networks (ANN), 40, 117, Binarization, 187
143, 189, 214 Biomarkers (BM), 145
Association rule mining, 143, 144 Bolster vector relapse, 147
Associative memory algorithms, 157, 158 Buffer-overflow attacks, 39
Asymmetric travelling salesman problem Business domain model, 59
(ATSP), 175 Business forecast, 101
Attribute-based access control (ABAC), 44 Business process management (BPM), 27, 124
AutoFOCUS31, 9 Business process modeling (BPM), 101

B C
Backpropagation neural network Capability-based access control (CAC), 44
(BPNN), 189–193 Cardiovascular autonomic neuropathy
Bare bones particle swarm optimization (CAN), 147
(BBPSO), 179 Centrality measures
Base stations, 29 betweenness centrality, 264
Bat algorithm, 173 closeness centrality, 264
Battery class, 18 clustering coefficient, 265–268
Bayesian algorithms, 272 degree centrality, 264
Bayesian cognitive learning methods, 156, 157 eigenvector, 265
Bayesian network, 109 graph theory, 263
Big data, 9 Chaotic PSO (CPSO), 179, 180
business use cases Charlson Comorbidity Index score, 146
customer experience, 152 Cloud-based services, 89
new product establishment, 152 Cloud computing (CC), 28, 47, 48, 271
predictions mechanical failures, 152 advantage, 5
security breaches, 152 data centers, 160
challenges, 158 decision-making process, 6
cloud computing, 151 industrial operation, 5
components, 139 internet services, 173
data processing, 158 IoT devices, 160
Index 281

mapping tasks, 172 Data flow diagram (DFD), 90


memory and storage units, 170 Data integrity, 88
service frameworks, 169 Daubechies wavelet function, 189
tool, 23 DCSim, 23
traffic engineering, 171 Decentralized decision-making process, 4
VM placement, 169, 170 Decision trees (DT), 62, 117, 143
Cloudlets, 160 Dedicated short-range communication
Cloud-to-cloud, 130 (DSRC), 29
Clustering, 144 Deep learning (DL), 33, 153
CoAP protocols, 12 Deep neural network (DNN), 45
Cognitive computing activation function, 273
process design, 155 applications
tools, 155 deep learning, 276
unstructured information, 153 diagnostics medical, 277
Cognitive learning (CL) fraud detection, 276
definition, 153 self-driving car, 277
learning technologies, 153 architecture, 273
machine learning algorithms, 154 Big Data, 272
role, 153 data set learning, 271
usage, 153 development, 272
Cognitive-machine learning hidden layer, 274
algorithms industry and research fields, 271
artificial intelligence, 155 intelligent transport systems, 272
associative memory, 157, 158 IoT and smart-world systems, 272
Bayesian learning methods, 156, 157 kinds of information, 271
HCA-GA, 157 linear-regression/decision-making
ordered version space search, 156 trees, 272
PSO, 156 outer layer, 274
decision-making, 155 quantity, 272
implementation, 155 reinforcement learning, 275, 276
intelligent personal assistants, 155 robotics and computer view, 271
principles, 155 SL, 275
Community brokers, 163 smart phones, 272
Community detection methodology, 163 supervised classification tasks, 271
Community-related algorithms, 163 unsupervised learning, 275
Constrained application protocol (CoAP), 7, Deep reinforcement learning, 42, 43
37, 45, 85 Delta operator, 204, 210
Contagion effect, 4 Denial of service (DoS), 84
ContentPlace method, 164 Denoising, 188
Context-aware access control (CWAC), 44 Device management, 133
Context-aware applications, 160 Device-to-cloud, 130
Context-aware computing, 160, 161 Device-to-device (D2D), 29, 130
Context filtering, 160 Device-to-gateway, 130
Continuous-and discrete-time systems, 204 Diabetes mellitus (DM)
Contract decoupling, 124 anti-diabetic medications, 141
Conventional algorithms, 170 Big Data analytics, 141
Covariance matrix adaptation evolution BM, 145
(CMA-ES), 198 components, 144
definition, 140
descriptive/predictive analytics, 141
D diagnosis, 144
Data-centric networks, 35 endocrine issue, 140
Data distribution services (DSS), 85 Hadoop MapReduce, 141
282 Index

Diabetes mellitus (DM) (cont.) challenges, 124, 125


KDD, 142 characteristics, 123, 124
pathophysiologies, 140 client-centric enterprise, 126
prevention and management, 140 cloud data technology, 126
types, 140 cloud platforms, 125
Diabetes Preprocessing Research Activity data streaming, 128
(DiScRi), 147 device layer, 129
Diabetic fundus image recuperation devices and machines, 124
(DFIR), 147 efficiency, 126
Diabetic nephropathy, 146 functioning capacity, 129
Diabetic retinopathy (DR), 147 gateways, 128
Digital communication system, 229 heterogeneous device, 129
Distributed denial of service (DDoS), 33 industrial architecture, 123
Distributed DoS (DDoS), 47, 48 industrial missions and visions, 123
Distributed Frank-Wolfe (dFW) algorithm, 45 Internet, 126
DM prediction internet-based communication, 125
high-dimensional datasets, 146 key attributes, 126, 127
hyperglycemia, 146–147 legacy networking devices, 127
LDA–MWSVM, 145 low-power and legacy devices, 128
Domain-based architectures, 84 middleware, 129
Dyna-Q model, 45 monetization models
application development, 133
device management, 133
E enable flexible, 132
Ease of development, 125 license and entitlement management,
Ease of testing, 125 132, 133
Eclipse framework, 9 product-based monetizing
Econometric model, 103 approach, 132
Edge processing, 159, 160 service-based approach, 132
EdgeBroker class, 17 software upgrades, 133
EdgeCloudSim, 11 network device, 129
EdgeDataCenter Class, 16 networking and hardware vendors, 129
Edgedevice, 16 opportunity, 124
EdgeDevice class, 17 organizational processing, 125
EdgeLet class, 17 procedures, 125
Edgelet model, 16 process, 129
Electromagnetic (EM), 145 processes and services, 123
Elitist ant system, 177 sector, 124, 125
Elliptic curve cryptography (ECC), 88, 89 secure communication, 129
Elliptic curve integrated encryption scheme sensors, 128
(ECIES), 90 smart cold chain model, 134, 135
Embedded frameworks, 28 smart security, 135, 136
Emulation, 227 standard-centered wireless, 127
Energy exchange, 226 technology architectures, 124
Energy management, 171 tiers, 127
Enterprise architecture (EA) traditional business, 124
application, 124, 125 transfer control instructions, 125
basic features, 126 trends, 124, 125
big data technology, 126 waste management, 134
business model wired network paradigms, 127
“as-a-service” enterprise model, 130 Enterprise IoT (E-IoT)
clients, 130 agent-based technologies, 82
subscription-centered framework, 130 applications, 81, 83
types, 130–132 CCTVs, 82
Index 283

customized/generic services, 81 Functional-centered selection method, 36


data points and data, 82 Fuzzy C-means (FCM) techniques, 46
devices, 82
electronic revenue, 82
elements of security, 87, 88 G
emerging technology, 81 Gartner Press, 81
growth, 83 Genetic algorithms (GA), 143
hardware components, 82 actual-parameter, 195
human threats, 82 adaptability, 195
lightweight secure measures, 88–90 CMA-ES, 198
natural threats, 82 constrained optimization, 199
power management applications, 83 DAG tasks, 199
security and privacy, 83 elements, 196
security practices, 94 G3 system, 198
security threats and attacks objective value, 199, 200
access-level attacks, 84 operator’s role, 197
adversary location attack, 84 parameters, 199, 200
data, 83 practical optimization issues, 195
devices, 83 SPX, 198
host-based attack, 84 structure, 196
information-level attacks, 83 task scheduling, 197, 199
layered architecture and associated techniques, 197
attacks, 84–86 UNDX, 198
phases, 86, 87 GRAD algorithms, 244
sensor technology, 81 GreenCloud, 11, 23
sensors, 82 Grid-connected PV system, 221
use cases, 82 Group activity recognition
user’s security and privacy, 83 HMM, 241
Enterprise resource planning (ERP) multifarious algorithms, 241
software, 100 targeted advertising, 241
Enterprise supply chain, 115, 116 GWOCFA algorithm, 204
Euclidean distances, 198
Extensible access control mark-up language
(XACML), 45 H
Extensible message and presence protocol Hadoop MapReduce, 141
(XMPP), 85 Hammerstein and Wiener model identification
Extreme learning machine (ELM) algorithms, 46 Delta domain, 204
FAdFPA (see Flower pollination algorithm
(FAdFPA))
F metaheuristic method approaches, 203, 204
Fault tolerance, 125 polynomial nonlinearity, 204
Fitness element, 197 Hammerstein and Wiener model
Fitness function, 197 parameters, 210
Flower pollination algorithm (FAdFPA) Handwriting identifications, 187
convergence curve, 206, 209 Handwritten character recognition (HCR),
fitness function, 206 186, 187
hybrid technique, 205 Hardware components, 243
metaheuristic algorithms, 206 Hardware-in-the-loop simulation, 227
optimization, 204 Heat map (HM)
PRBS input signal, 206 analysis, 250
p-values, 206, 209 characteristics, 244
user-defined parameters, 205 day calculations, 244, 246, 249
Fog computing (FC), 3, 28, 159 density of people, academic building, 245
Friendlink algorithm, 255–257, 259, 268 foot path comparison, 248
284 Index

Heat map (HM) (cont.) Inter-connection, 29


GRAD algorithms, 244 Internal domain operability, 125
HMB algorithm, 244 International Telecommunication Union
human activity recognition, 243 (ITU-T), 81
human foot path Internet of Energy (IoE), 225
densities, 250 Internet of Intelligent Things (IoIT), 70
hostels, 247, 249 Internet of Manufacturing Things (IoMT), 110
human footfall, academic block, 246 Internet of Things (IoT), 271
PGTB algorithm, 244 application engineers and developers, 8
result, 245 application layer threats, 39
SF method, 243–245 applications, 28, 233
standard surface classes, 243 architecture, 7, 8
WF-SVM algorithms, 244 Big Data technology, 1, 114
Heat sources, 243 challenges, 10, 11
Heterogeneity, 5, 29, 49 authentication and identification, 73
Heterogeneous business applications, 94 connectivity, 74
Heterogeneous network, 113 consumer awareness, 77
Hidden Markov models (HMM), 241 data capturing capabilities, 75
Hill climbing-GA (HCA-GA), 157 data security, 76
HMB algorithm, 244 delivering value, 77
Host-based attack, 84 handling unstructured data, 74
Human crowd, 242 integration, 74
Hybrid Active Learning Link Prediction intelligent analytics, 75, 76
(HALLP), 254 interoperability and compatibility, 73
Hyperglycemia privacy issues, 76
dataset, 146 commercialization and technological
diabetic nephropathy, 146 advancements, 28
insulin resistance, 146 communication competencies, 35
morbidity and mortality, 146 communication protocol, 13, 14
negative effects, 146 contributions, 11
random forest, 146 cross-sensing organization procedures, 27
Hypertext Transfer Protocol (HTTP), 7 data-centric networks, 35
Hypoglycemia, 147 data collection, 113
data evaluation systems, 6
data heterogeneity, 28
I data security issues, 112
IEEE 13-node test feeder system, 214 data sensing, 27
DC-link capacitors, 215 developers, 27
photovoltaic module, 214 devices, 28
point of common coupling, 215 distributed and interlinked network, 28
RL filter, 215 diversities, 27
transformer and load, 215 domains, 28
iFogSim, 11 edge computing, 7, 11, 12
Industrial architecture, 123 factors, 9
Industrial IoT (IIoT), 159 heterogeneous network, 113
Inexpensive and less-energy information ICT, 1
transfer, 30 industrial application
Information-level attacks, 83 centralized processing, 6
Infrastructure as a service (IaaS), 169 cloud computing, 5
Instance-based learning (IBL), 143 IoTs and human, 6
Intellectual property, 86 Pi and UDOO boards, 6
Intelligent learning, 153 system dynamics, 5
Intelligent systems, 271 transformative implication, 6
Intelligent urban environments, 30 WSNs and RFIDs Networks, 5
Index 285

integration, 113 EdgeLet class, 17


intelligent devices, 28 MEL class, 17
intelligent predictive analytics, 114 mobility class, 18
intelligent sensors, 27 policies class, 18
IP-and TCP-centered information user interface class, 18
transfer, 35 cloudsim, 15
lack of awareness, 114 design and implementation, 15, 16
lot of services and applications, 35 healthcare sectors, 20, 21
ML, 28 RSU controllers, 21, 22
modelling ecosystems, 24 smart building systems, 21
modern manufacturing, 2–4 IoTSim-Edge simulators, 11
multiple-layer threats, 39, 40
network connection, 112, 113
networking layer threats, 38 J
network simulators, 22–24 Java Context Awareness Framework
organizational automations, 1 (JCAF), 160
organizational procedure, 27
physical threats, 35–37
privacy and security concerns, 34 K
PV systems (see Photovoltaic (PV) system) k-clique algorithm, 164
reliability theory, 2–4 k-nearest neighbors (k-NN), 143
remote evaluation of models, 233 Knowledge discovery in databases
resource-based, 35 (KDD), 142
SDNs, 35 Korea Transportation Safety Authority, 106
services and applications, 28
survey scope and research contributions, 34
surveys, 34 L
technological vendors, 1 LabVIEW software, 229
transportation layer threats, 38 Large-scale deployment, 29
Internet of Vehicles (IoV) network, 90 League Championship Algorithm (LCA), 173
Internet Technology (IT), 2 Levelized energy costs (LCoE), 227–229
Inverse sign test (IST), 145 Levenberg-Marquardt algorithm, 218
Inverter circuit, 219 Lightweight implementation, 125
Inverter controller Linear Discriminant Analysis (LDA), 146
ANN, 217, 218 Linear regression, 41
monitoring system, 218 Local computing, 159
PI, 217
IoT-based inverter controller systems, 222
IoT-based solar power plant, 237 M
IoTs battery drainage Machine learning (ML), 28, 32, 33, 272
architectural computing, 14, 15 access control, 44
computation and event processing, 19 advantages, 71
description, 14 AI, 70
sim edge (see IoTs Sim edge model) architectural layers, 44
sim edge architecture, 15 artificial intelligence industry, 56
IoTs ecosystem, 10 asset tracking, 69
IoTs sim edge architecture, 15 authentication and access, 44
IoTs Sim edge model authentication processes, 45
architecture, 15 automatic fulfillment, 69, 70
classes business activities, 58
battery class, 18 business domain model, 59
EdgeBroker class, 17 compliance monitoring, 67
EdgeDataCenter Class, 16 consumer lifetime valuation, 57
EdgeDevice class, 17 consumers, 58
286 Index

Machine learning (ML) (cont.) MEL class, 17


data-driven approach, 154 Message queuing telemetry transport protocol
DDoS, 47, 48 (MQTTP), 85
decision-aided system, 44 Metaheuristic algorithms, 203
decision tree, 62 Metaheuristic techniques, 173
deep reinforcement learning, 42, 43 Micro-grid
definition, 142, 153 feature, 213
domains, 57 IEEE 13-node, 214
DoS, 47, 48 inverter circuit, 215
economics, 66 inverter controller, 217–219
features, 154 modes, 216
frequency channels, 45 MPPT, 216
improved accuracy rate, 71 power system, 213
improved customer satisfaction, 72 types, 213
increased operational efficiency, 72 Microsoft Security Development Lifecycle
industrial applications, 63, 65 (SDL), 91
industry, 154 Minimum and maximum ant system, 177
information transformation, 57 ML-centered methods, 48
Internet, 55 MLP network models, 193
IoT data, 58 MobiAd, 162
IoT networking system, 43 Mobile edge computing (MEC), 160
life cycle, 154 Morlet Wavelet Support Vector Machine
lot of applications, 44 (MWSVM), 146
malicious programs, 49–51 MPPT controller, 216
methods, 56 Multi-class model, 108
mitigation and detection, 46 Multifarious algorithms, 241
naïve Bayes (NB), 43 Multilayer perceptron neural network
new innovative IoT emerging business human brain, 190
models, 65 performance analysis
objectives, 154 accuracy, 191
predictive analysis and maintenance, 71 sensitivity, 191
preventative maintenance, 67, 68 specificity, 192
regression, 43 Multiple linear regression algorithm, 109
remote diagnostics, 68 Multiple-layer threats, 39, 40
segments, 44
sellers, 58
semi-supervised learning, 42 N
shaping experiences to individuals, 65 Nash equilibrium (NE), 45, 50
SL, 60, 61 National Environment Agency (NEA), 134
supervised learning algorithms, 43 Nature-inspired algorithms, 172
supervised machine learning, 40 Network-operated intelligent bots, 160
SVMs, 43 Network simulators, 22–24
tasks, 142 Network virtualization, 159
unsupervised learning, 62, 63 Neural network (NN), 116, 187, 214
unsupervised machine learning, 41 Next-generation enterprise
Wi-Fi signals, 45 Big Data, 5
Makespan, 199, 201 decentralized decision-making process, 4
Man-in-the-middle (MITM), 84 dynamics and flat structures, 4
Maturity onset diabetes of the young heterogeneity, 5
(MODY), 140 Non-conventional algorithms, 170
Maximum power point (MPP), 213–214 Non-parametric Wilcoxon rank-sum analysis, 206
Maximum power point tracking (MPPT), Nonproliferative (NPDR), 147
214, 216 Novel performance models, 9
Mean Average Precision (MAP), 268 NP-hard issues, 172
Index 287

O components, 235
OMNeT++ system, 23 DC electricity, 234
OneMax function, 200 environmental conditions, 235
Open authorization (OAuth), 45 installation, 233
Open-source platform, 151 IoT, 233, 234
Open Web Application Security Project layout, 234
(OWASP), 39, 50, 93 models, 234
Opportunistic data distribution parameters, 235
challenges, 162 power generation potential, 233
interaction pattern, 162 power harvester, 234
mobile devices, 162 PV arrays, 234
short-range communication, 162 remote monitoring and control layer, 235
Opportunistic networks (ONs) remote server data, 236
data distribution, 162 sensing layer, 236
community brokers, 163 solar power plant, 237
community detection, 164 system layer, 236
community-related algorithms, 163 Photovoltaic system, 214
ContentPlace method, 164 Physical threats, 35–37
gossiping, 163 PIR sensors, 242, 243, 247
socio-aware overlay algorithm, 163 Platform as a service (PaaS), 169
subscribe/unsubscribe approach, 163 Policies classes, 18
DTNs, 162 Policy-based access control (PBAC), 44
smart city application, 163 Polynomial regression, 41
smart city sensors, 163 Power quality, 221
topology, 162 Power-use-centered selection method, 36
Optical character recognition (OCR) Predictive analytics, 148
applications, 185 Predictive markers
digital image format, 185 assessment, 145
Optimization algorithms, 170, 171 EM and IST, 145
Organization-based access control (OAC), 44 Priority-aware VM allocation (PAVA), 173
Product-as-a-service (PaaS), 132
Product-based monetizing approach, 132
P Proliferative (PDR), 147
Particle swarm optimization (PSO), 156 Public key infrastructure (PKI), 89
AFPSO, 180 PV arrays, 234
applications, 180 PV generators, 234
BBPSO, 179 Python, 242
bird flocking, 177
cognitive element, 179
comparison, 180 Q
CPSO, 179 Quality of services (QoS), 12, 37
particles movement, 178 Quantum-behaved Particle Swarm
pseudo-code, 178, 179 Optimization–Simplex method
QPSO, 179 (QPSO), 179
TVAC, 180 Quasi-Newton method, 198
Particle swarm optimization method, 106–107
Performance Management Work Tool (PMWT), 9
Perturbation period (Tp), 218 R
PGTB algorithm, 244 Radial basis function (RBF), 189–193
Phase-locked loop (PLL) algorithm, 217 Radio-frequency identification devices
Photovoltaic (PV) system ML, 32, 33
application layer, 237 network features, 29–31
architecture, 235, 236 privacy issue, 31
Arduino server, 235 privacy resolutions, 31, 32
288 Index

Random candidate selection (RCS), 189–193 Smart sensors, 159


Random neural network (RNN), 48 Smart Urban Visualization system
Rank-based ant system, 177 (SURV), 242
Raspberry Pi 3, 93 Smart wearables, 159
Recurrent neural network (RNN), 116 Social Network Analysis (SNA)
Reinforcement learning (RL), 42, 144, 275, 276 dataset, 258
Reliability theory, 2–4 efficiency, 254, 258
Remote monitoring and control layer, 235 elements and ramifications, 253
Renewable energy sources (RESs), 213 empirical analysis, 268
Resource management, 171 experiments, 258
Resource utilization, 159 Friendlink link prediction, 260
Role-based access control (RBAC), 44 graph, 255
Rubin’s algorithm, 256 link prediction model, 254
Link Prediction Problem, 255
MATLAB, 258
S mechanisms, 253
SamBot, 109 network theory, 253
Sanskrit character recognition nodes, 259
ANN, 186 problem description
classification Friendlink algorithm, 256, 257
BPNN with RBF, 190 social links, 256
multilayer perceptron neural network, 190 soft set theory, 256
RCS with BPNN, 189, 190 proposed approach, 259
handwritten, 186 sampling techniques, 254
HCR, 186 SFLP, 261, 262
NNs, 186 social entities, 253
OCR, 186 supervised and unsupervised
process algorithms, 254
feature extraction, 189 theoretical analysis, 268
image acquisition, 187 Socio-aware overlay algorithm, 163
pre-processing, 187, 188 Soft set and Friendlink-based Link Prediction
segmentation, 188 algorithm (SFLP), 257, 261, 262
stages, 187 Soft set theory, 254, 256, 257, 267, 269
traditional classifiers, 192 Software architecture, 9
Scalability, 124 Software database, 243
Segmentation, 188 Software as a service (SaaS), 130, 169
Semi-supervised learning, 42, 147 Software-defined networks (SDNs), 28, 35
Sensing layer, 236 Solar photovoltaic system, 233
Sensing/perception layer, 85 Solar power plant, 236
Sensitivity, 191 Stagnation, 175
SensorLogic, 132, 133 Standard communication protocols, 166
Sensors, 82 STRIDE framework methodology, 91, 92
Service-based approach, 132 Subscribe/unsubscribe approach, 163
Simplex crossover (SPX), 198 Supervised learning (SL), 60, 61, 142,
Simulators, 11 143, 275
Smart building systems, 21 Supervised machine learning, 40
Smart city, 158 Supervised machine learning
Smart cold chain model, 134, 135 algorithms, 103
Smart grid, 226 Support vector machine (SVM), 116, 141, 143
Smart networking, 30 Supported vector regression (SVR), 41
Smart remote monitoring system Surface fitting (SF) method, 243
IoT, 237 Surveillance cameras, 243
PV system (see Photovoltaic (PV) system) System Dynamics (SD), 3
web-based technique, 237, 238 System layer, 236
Index 289

T V
Tableau, 242 Vehicle-to-grid (V2G), 226
Targeted advertising, 241 Verification of success (VoS), 90
Target-of-evaluation (ToE), 90 Virtual machine (VM) placement
Temporal motion data, 241 aim, 172
Thermal diffusion method, 243, 244 cloud computing, 172, 173
Threat modeling (TM) migration, 173
Kali Linux-based ethical hacking, nature-inspired algorithms, 172
92–94 objects, 172
lifecycle, 90 optimization, 173
Microsoft SDL, 91 VM migration, 173
SDL, 90 Virtual machine (VM) scheduling, 169
STRIDE framework methodology, Virtualization, 169
91, 92
TMT, 92
Threat modeling tool (TMT), 92 W
Time-series models, 103 Waste management, 134
Traditional classifiers, 192 Web-service-related middleware
Traditional network, 162 ESCAPE, 160
Transmission control protocol WF-SVM algorithms, 244
(TCP), 85 Wide area networks (WANs), 160
Transportation layer threats, 38 Wind energy conversion system (WECS)
Travelling salesman problem cybersecurity, 226
(TSP), 175 wind flow, 227
WT, 227
Wind energy system, 225
U Wind technology, 225
UCI Network Data Repository, 258 Wind turbine (WT), 227
Ultra-reliable and low-latency communication Wind turbine emulator
(URLLC), 30 generators, 227
Unimodal normal distribution crossover hardware-in-the-loop simulation, 227
(UNDX), 198 IoT-based, 229, 230
Unsupervised learning, 62, 63, 143, 275 line diagram, 227, 228
Unsupervised machine learning, 41 motor buck duties, 229
Unsupervised machine learning quality and reliability, 227
algorithms, 103–104 Wireless multimedia system (WMS), 50
Usage control-based access control
(UCAC), 44
User datagram protocol (UDP), 85 Z
User interface classes, 18 Zachary’s Karate Club data set, 258
User-managed access (UMA), 45 Zhang-Suen thinning algorithm, 188

View publication stats

You might also like