Pandian a computer networks, big data and iot 2021

The wireless sensor system topology, the convenience of resources,and the energy consumption of nodes in different paths of the data collection treemay vary largely, thus affecting the g

Lecture Notes on Data Engineering

and Communications Technologies 66

Lecture Notes on Data Engineering and Communications Technologies

Volume 66

Series Editor

Fatos Xhafa, Technical University of Catalonia, Barcelona, Spain

The aim of the book series is to present cutting edge engineering approaches to datatechnologies and communications It will publish latest advances on the engineeringtask of building and deploying distributed, scalable and reliable data infrastructuresand communication systems.

The series will have a prominent applied focus on data technologies andcommunications with aim to promote the bridging from fundamental research ondata science and networking to data engineering and communications that lead toindustry products, business knowledge and standardisation

Indexed by SCOPUS, INSPEC, EI Compendex

All books published in the series are submitted for consideration in Web of Science

More information about this series athttp://www.springer.com/series/15362

A Pasumpon Pandian · Xavier Fernando · Syed Mohammed Shamsul Islam

Editors

Computer Networks, Big Data and IoT

Proceedings of ICCBI 2020

Syed Mohammed Shamsul Islam

Edith Cowan University (ECU)

Joondalup, WA, Australia

Xavier FernandoDepartment of Electrical and ComputerEngineering

Ryerson UniversityToronto, ON, Canada

ISSN 2367-4512 ISSN 2367-4520 (electronic)

Lecture Notes on Data Engineering and Communications Technologies

ISBN 978-981-16-0964-0 ISBN 978-981-16-0965-7 (eBook)

or dissimilar methodology now known or hereafter developed.

The use of general descriptive names, registered names, trademarks, service marks, etc in this publication does not imply, even in the absence of a specific statement, that such names are exempt from the relevant protective laws and regulations and therefore free for general use.

The publisher, the authors and the editors are safe to assume that the advice and information in this book are believed to be true and accurate at the date of publication Neither the publisher nor the authors or the editors give a warranty, expressed or implied, with respect to the material contained herein or for any errors or omissions that may have been made The publisher remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

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

We are honored to dedicate the proceedings

of ICCBI 2020 to all the participants and editors of ICCBI 2020.

It is with deep satisfaction that I write this Foreword to the proceedings of ICCBI

2020 held in Vaigai College of Engineering, Madurai, Tamil Nadu, on December15–16, 2020

This conference was bringing together researchers, academics and professionalsfrom all over the world, and experts in computer networks, big data and Internet ofthings

This conference particularly encouraged the interaction of research students anddeveloping academics with the more established academic community in an informalsetting to present and to discuss new and current work The papers contributed themost recent scientific knowledge known in the field of computer networks, bigdata and Internet of things Their contributions helped to make the conference asoutstanding as it has been The local organizing committee members and their helpersput much effort into ensuring the success of the day-to-day operation of the meeting

We hope that this program will further stimulate research in data communicationand computer networks, Internet of things, wireless communication, big data andcloud computing and also provide practitioners with better techniques, algorithms,and tools for deployment We feel honored and privileged to serve the best recentdevelopments to you through this exciting program

We thank all the guest editors, authors and participants for their contributions

Dr P SugumaranConference Chair, ICCBI 2020

vii

This conference proceedings volume contains the written versions of most of thecontributions presented during the conference of ICCBI 2020 The conferenceprovided a setting for discussing recent developments in a wide variety of topicsincluding computer networks, big data and Internet of things The conference hasbeen a good opportunity for participants coming from various destinations to presentand discuss topics in their respective research areas

This conference tends to collect the latest research results and applications oncomputer networks, big data and Internet of things It includes a selection of 74papers from 248 papers submitted to the conference from universities and industriesall over the world All of accepted papers were subjected to strict peer-reviewing

by 2–4 expert referees The papers have been selected for this volume because ofquality and the relevance to the conference

We would like to express our sincere appreciation to all the authors for theircontributions to this book We would like to extend our thanks to all the refereesfor their constructive comments on all papers; especially, we would like to thankthe organizing committee for their hard work Finally, we would like to thank theSpringer publications for producing this volume

Dr R Thiruchenthuran, Thiru S Kamalakannan, Thiru S Balasubramanian andThiru S Singaravelan, for the discussion, suggestion and cooperation to organizethe keynote speakers of this conference The organizers also wish to acknowledgespeakers and participants who attend this conference Many thanks are given to allpersons who help and support this conference ICCBI 2020 would like to acknowl-edge the contribution made to the organization by its many volunteers Memberscontribute their time, energy and knowledge at a local, regional and internationallevel.

We also thank all the chair persons and conference committee members for theirsupport

xi

Maximizing Network Lifetime in WSN Using Ant Colony

Algorıthm 1

M D Saranya, G Pradeepkumar, J L Mazher Iqbal,

B Maruthi Shankar, and K S Tamilselvan

Deep Ensemble Approach for Question Answer System 15

K P Moholkar and S H Patil

Information Sharing Over Social Media Analysis Using Centrality

Measure 25

K P Ashvitha, B Akshaya, S Thilagavathi, and M Rajendiran

AndroHealthCheck: A Malware Detection System for Android

Using Machine Learning 35

Prerna Agrawal and Bhushan Trivedi

Use of Machine Learning Services in Cloud 43

Chandrashekhar S Pawar, Amit Ganatra, Amit Nayak,

Dipak Ramoliya, and Rajesh Patel

An Experimental Analysis on Selfish Node Detection Techniques

for MANET Based on MSD and MBD-SNDT 53

V Ramesh and C Suresh Kumar

Metaheuristic-Enabled Shortest Path Selection for IoT-Based

Wireless Sensor Network 71

Subramonian Krishna Sarma

Improved Harris Hawks Optimization Algorithm for Workflow

Scheduling Challenge in Cloud–Edge Environment 87

Miodrag Zivkovic, Timea Bezdan, Ivana Strumberger,

Nebojsa Bacanin, and K Venkatachalam

xiii

xiv Contents

Generation of Random Binary Sequence Using Adaptive Row–

Column Approach and Synthetic Color Image 103

C Manikandan, N Raju, K Sai Siva Satwik, M Chandrasekar,

and V Elamaran

Blockchain: Application Domains, Research Issues and Challenges 115

Dipankar Debnath and Sarat Kr Chettri

A Study of Mobile Ad hoc Network and Its Performance

Optimization Algorithm 131

Vishal Polara and Jagdish M Rathod

Industrial IoT: Challenges and Mitigation Policies 143

Pankaj Kumar, Amit Singh, and Aritro Sengupta

Eclat_RPGrowth: Finding Rare Patterns Using Vertical Mining

and Rare Pattern Tree 161

Sunitha Vanamala, L Padma Sree, and S Durga Bhavani

Research Scholars transferring Scholarly Information through

Social Medias and Networks in the Selected State Universities

of Tamil Nadu 177

C Baskaran and P Pitchaipandi

Twitter-Based Disaster Management System Using Data Mining 193

V G Dhanya, Minu Susan Jacob, and R Dhanalakshmi

Sentimental Analysis on Twitter Data of Political Domain 205

Seenaiah Pedipina, S Sankar, and R Dhanalakshmi

Cloud-Based Smart Environment Using Internet of Things (IoT) 217

E Laxmi Lydia, Jose Moses Gummadi, Sharmili Nukapeyi,

Sumalatha Lingamgunta, A Krishna Mohan, and Ravuri Daniel

A Review of Healthcare Applications on Internet of Things 227

S Chitra and V Jayalakshmi

Big Social Media Analytics: Applications and Challenges 239

Sonam Srivastava and Yogendra Narain Singh

A Cost and Power Analysis of Farmer Using Smart Farming IoT

System 251

P Darshini, S Mohana Kumar, Krishna Prasad, and S N Jagadeesha

Intelligent Computing Application for Cloud Enhancing

Healthcare Services 261

Anandakumar Haldorai and Arulmurugan Ramu

Contents xv

Coronavirus Detection and Classification Using X-Rays and CT

Scans with Machine Learning Techniques 277

Moulana Mohammed, P V V S Srinivas, Veldi Pream Sai Gowtham,

Adapa V Krishna Raghavendra, and Garapati Khyathi Lahari

Johnson’s Sequencing for Load Balancing in Multi-Access Edge

Computing 287

P Herbert Raj

A Study on MPLS Vs SD-WAN 297

S Rajagopalan

Security Issues and Solutions in E-Health and Telemedicine 305

Deemah AlOsail, Noora Amino, and Nazeeruddin Mohammad

Accident Alert System with False Alarm Switch 319

S Alen, U Advaith, Joveal K Johnson, Kesia Mary Joies,

Rahul Sunil, Aswathy Ravikumar, and Jisha John

Metaheuristics Algorithms for Virtual Machine Placement

in Cloud Computing Environments—A Review 329

Jyotsna P Gabhane, Sunil Pathak, and Nita M Thakare

Prostate Image Segmentation Using Ant Colony

Optimization-Boundary Complete Recurrent Neural

Network (ACO-BCRNN) 351

J Ramesh and R Manavalan

A Deep Learning Approach to Detect Lumpy Skin Disease in Cows 369

Gaurav Rai, Naveen, Aquib Hussain, Amit Kumar, Akbar Ansari,

and Namit Khanduja

Prediction of Influenza-like Illness from Twitter Data and Its

Comparison with Integrated Disease Surveillance Program Data 379

Monica Malik and Sameena Naaz

Review of Denoising Framework for Efficient Removal of Noise

from 3D Images 395

Anand B Deshmukh and Sanjay V Dudul

Algorithmic Trading Using Machine Learning and Neural Network 407

Devansh Agarwal, Richa Sheth, and Narendra Shekokar

Analysis on Intrusion Detection System Using Machine Learning

Techniques 423

B Ida Seraphim and E Poovammal

Content Related Feature Analysis for Fake Online Consumer

Review Detection 443

Dushyanthi Udeshika Vidanagama, Thushari Silva, and Asoka Karunananda

xvi Contents

Big Data Link Stability-Based Path Observation for Network

Security 459

Nedumaran Arappali, Melaku Tamene Mekonnen,

Wondatir Teka Tefera, B Barani Sundaram, and P Karthika

Challenging Data Models and Data Confidentiality Through

“Pay-As-You-Go” Approach Entity Resolution 469

E Laxmi Lydia, T V Madhusudhana Rao, K Vijaya Kumar,

A Krishna Mohan, and Sumalatha Lingamgunta

Preserving and Scrambling of Health Records with Multiple

Owner Access Using Enhanced Break-Glass Algorithm 483

Kshitij U Pimple and Nilima M Dongre

Malignant Web Sites Recognition Utilizing Distinctive Machine

Learning Techniques 497

Laki Sahu, Sanjukta Mohanty, Sunil K Mohapatra, and Arup A Acharya

Speech Parameter and Deep Learning Based Approach

for the Detection of Parkinson’s Disease 507

Akhila Krishna, Satya prakash Sahu, Rekh Ram Janghel,

and Bikesh Kumar Singh

Study on Data Transmission Using Li-Fi in Vehicle to Vehicle

Anti-Collision System 519

Rosebell Paul, Neenu Sebastian, P S Yadukrishnan, and Parvathy Vinod

Approaches in Assistive Technology: A Survey on Existing

Assistive Wearable Technology for the Visually Impaired 541

Lavanya Gupta, Neha Varma, Srishti Agrawal, Vipasha Verma,

Nidhi Kalra, and Seemu Sharma

Stateless Key Management Scheme for Proxy-Based Encrypted

Databases 557

Kurra Mallaiah, Rishi Kumar Gandhi, and S Ramachandram

Exploration of Blockchain Architecture, Applications,

and Integrating Challenges 585

Jigar Mehta, Nikunj Ladvaiya, and Vidhi Pandya

Filter Bank Multicarrier Systems Using Gaussian Pulse-Based

Filter Design for 5G Technologies 601

Deepak Singh and Mukesh Yadav

LIMES: Logic Locking on Interleaved Memory for Enhanced

Security 613

A Sai Prasanna, J Tejeswini, and N Mohankumar

Contents xvii

A Novel IoT Device for Optimizing “Content Personalization

Strategy” 627

Vijay A Kanade

IoT Based Self-Navigation Assistance for Visually Impaired 635

Nilesh Dubey, Gaurang Patel, Amit Nayak, and Amit Ganatra

An Overview of Cyber-Security Issues in Smart Grid 643

Integration of IoT and SDN to Mitigate DDoS with RYU Controller 673

Mimi Cherian and Satishkumar Verma

Low Rate Multi-vector DDoS Attack Detection Using Information

Gain Based Feature Selection 685

R R Rejimol Robinson and Ciza Thomas

A Framework for Monitoring Patient’s Vital Signs with Internet

of Things and Blockchain Technology 697

A Christy, MD Anto Praveena, L Suji Helen, and S Vaithyasubramanian

IoT Based Smart Transport Management and Vehicle-to-Vehicle

Communication System 709

Vartika Agarwal, Sachin Sharma, and Piyush Agarwal

An Analytical and Comparative Study of Hospital Re-admissions

in Digital Health Care 717

Aksa Urooj, Md Tabrez Nafis, and Mobin Ahmad

An Edge DNS Global Server Load Balancing for Load Balancing

in Edge Computing 735

P Herbert Raj

Network Intrusion Detection Using Cross-Bagging-Based Stacking

Model 743

S Sathiya Devi and R Rajakumar

Enterprise Network: Security Enhancement and Policy

Management Using Next-Generation Firewall (NGFW) 753

Md Taslim Arefin, Md Raihan Uddin, Nawshad Ahmad Evan,

and Md Raiyan Alam

xviii Contents

Comparative Study of Fault-Diagnosis Models Based on QoS

Metrics in SDN 771

Anil Singh Parihar and Nandana Tiwari

A Brief Study on Analyzing Student’s Emotions with the Help

of Educational Data Mining 785

S Aruna, J Sasanka, and D A Vinay

IoT-PSKTS: Public and Secret Key with Token Sharing Algorithm

to Prevent Keys Leakages in IoT 797

K Pradeepa and M Parveen

Investigation and Analysis of Path Evaluation for Sustainable

Communication Using VANET 813

D Rajalakshmi, K Meena, N Vijayaraj, and G Uganya

Performance Study of Free Space Optical System Under Varied

Atmospheric Conditions 827

Hassan I Abdow and Anup K Mandpura

Malicious URL Detection Using Machine Learning and Ensemble

Modeling 839

Piyusha Sanjay Pakhare, Shoba Krishnan, and Nadir N Charniya

Review on Energy-Efficient Routing Protocols in WSN 851

G Mohan Ram and E Ilavarsan

Intelligent Machine Learning Approach for CIDS—Cloud

Intrusion Detection System 873

T Sowmya and G Muneeswari

In-network Data Aggregation Techniques for Wireless Sensor

Networks: A Survey 887

T Kiruthiga and N Shanmugasundaram

Comparative Analysis of Traffic and Congestion

in Software-Defined Networks 907

Anil Singh Parihar, Kunal Sinha, Paramvir Singh, and Sameer Cherwoo

A Comparative Analysis on Sensor-Based Human Activity

Recognition Using Various Deep Learning Techniques 919

V Indumathi and S Prabakeran

FETE: Feedback-Enabled Throughput Evaluation for MIMO

Emulated Over 5G Networks 939

B Praveenkumar, S Naik, S Suganya, I Balaji, A Amrutha,

Jayanth Khot, and Sumit Maheshwari

Automatic Vehicle Service Monitoring and Tracking System Using

IoT and Machine Learning 953

M S Srikanth, T G Keerthan Kumar, and Vivek Sharma

Contents xix

Machine Learning-Based Application to Detect Pepper Leaf

Diseases Using HistGradientBoosting Classifier with Fused HOG

and LBP Features 969

Matta Bharathi Devi and K Amarendra

Efficacy of Indian Government Welfare Schemes Using

Aspect-Based Sentimental Analysis 981

Maninder Kaur, Akshay Girdhar, and Inderjeet Singh

Author Index 989

About the Editors

A Pasumpon Pandian received his Ph.D degree in the Faculty of Information and

Communication Engineering under Anna University, Chennai, TN, India, in 2013

He received his graduation and postgraduation degree in Computer Science andEngineering from PSG College of Technology, Coimbatore, TN, India, in the year

1993 and 2006, respectively He is currently working as Professor in the ComputerScience and Engineering department of KGiSL Institute of Technology, Coimbatore,

TN, India He has twenty-six years of experience in teaching, research and ITindustry He has published more than 20 research articles in refereed journals Heacted as Conference Chair in IEEE and Springer conferences and Guest Editor inComputers and Electrical Engineering (Elsevier), Soft Computing (Springer) andInternational Journal of Intelligent Enterprise (Inderscience) Journals His researchinterest includes image processing and coding, image fusion, soft computing andswarm intelligence

Xavier Fernando is Professor at the Department of Electrical and Computer

Engi-neering, Ryerson University, Toronto, Canada He has (co-)authored over 200research articles and two books (one translated to Mandarin) and holds few patentsand non-disclosure agreements He was IEEE Communications Society Distin-guished Lecturer and delivered close over 50 invited talks and keynote presentationsall over the world He was Member in the IEEE Communications Society (COMSOC)Education Board Working Group on Wireless Communications He was Chair IEEECanada Humanitarian Initiatives Committee 2017–2018 He was also Chair of theIEEE Toronto Section and IEEE Canada Central Area He is a program evaluator forABET (USA) He was a visiting scholar at the Institute of Advanced Telecommu-nications (IAT), UK, in 2008, and MAPNET Fellow visiting Aston University, UK,

in 2014 Ryerson University nominated him for the Top 25 Canadian Immigrantsaward in 2012 in which was a finalist His research interests are in signal processingfor optical/wireless communication systems He mainly focuses on physical andMAC layer issues He has special interest in underground communications systems,

of cognitive radio systems, visible light communications and wireless positioningsystems

xxi

xxii About the Editors

Dr Syed Mohammed Shamsul Islam completed his Ph.D with Distinction in

Computer Engineering from the University of Western Australia (UWA) in 2011 Hereceived his M.Sc in Computer Engineering from King Fahd University of Petroleumand Minerals in 2005 and B.Sc in Electrical and Electronic Engineering from IslamicInstitute of Technology in 2000 Before joining ECU as a Lecturer in ComputerScience, he worked in different teaching and research positions at UWA and CurtinUniversity (2011–2016) He was promoted to Senior Lecturer in November 2020 Hehas published over 60 research articles and got 17 public media releases, including

a TV news story and four live radio interviews He has received the NHMRC Ideasgrant 2019 (AUD 467,980) and nine other external research grants He is serving the

scientific community as an Associate Editor of IEEE Access, a guest editor of

Health-care, a Technical Committee Member of 25 conferences and a regular reviewer of

26 journals He is a Senior Member of IEEE and Australian Computer Society His

research interest includes Artificial Intelligence, Computer Vision, Pattern nition, Big-Data Analysis, Biometrics, Medical Imaging, Internet of Things (IoT),Image Processing and Biomedical Engineering

Recog-Maximizing Network Lifetime in WSN

Using Ant Colony Algorıthm

M D Saranya, G Pradeepkumar, J L Mazher Iqbal, B Maruthi Shankar, and K S Tamilselvan

Abstract A wireless network is a cluster of specific transducers with statement

transportation intend to study it frequently operate in an unpredictable wireless ground with vigour constriction Several types of research are mainly interest invigour consciousness and statement dependability of a wireless sensor network tomaximize network lifetime In this article, a greedy algorithm and ACO algorithmaims at obtaining the best clarification that satisfies the given set of Greedy algo-rithm The aim of the Greedy algorithm obtains a most favorable explanation thatsatisfies the given set of constraints and also maximizes the given objective func-tion Ant colony algorithm has been practical to the traveling salesman problem tofind the optimal solution in a short time However, the performance of the ACOalgorithm is considered for both high energy efficiency and good power balancing,and maximal energy utilization throughout the network ACO algorithm it is under-standable that the network time increases and extends the life cycle of the wirelesssensor network since it manages the energy and power management The GreedyAlgorithm creates the problem of selecting a communication path using the trav-eling salesman and cracks the logic by using this algorithm The algorithm uses thesingle source to all destination technique to find through path for optimum networkconnectivity The simulation results were demonstrated with the help of the algorithmand it outperforms the shortest path length concerning the network lifetime

back-Keywords Wireless network·Maximizing network lifetime·Greedy algorithm·ACO·Communication reliability·Connectivity·Coverage·Traveling salesmanproblem

M D Saranya (B) · G Pradeepkumar · K S Tamilselvan

Department of ECE, KPR Institute of Engineering and Technology, Coimbatore, Tamilnadu, India

J L Mazher Iqbal

Department of ECE, Veltech Rangarajan Dr Sagunthala R&D Institute of Science and

Technology, Chennai, Tamilnadu, India

B Maruthi Shankar

Department of ECE, Sri Krishna College of Engineering and Technology, Coimbatore,

Tamilnadu, India

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd 2021

A Pasumpon Pandian et al (eds.), Computer Networks, Big Data and IoT, Lecture Notes

on Data Engineering and Communications Technologies 66,

https://doi.org/10.1007/978-981-16-0965-7_1

1

2 M D Saranya et al.

The wireless group is the way of contact along with sensors and observers A Sensornetwork is a collection of dedicated transducers among a transportation infrastructureintended to monitor as well as record conditions at the varied location WSN naturallyencompasses numerous integer of spatially detached array-operated, an embeddedoption that is a network to accumulate the data to the users, and it has the restrictedcomputing and processing capabilities Full coverage with connectivity suggests thatevery location in the field is enclosed by at least one node and the information on thisposition can be running scared to the fusion center The complexity of data routingand processing also depends on topology In a WSN the sensors are work in a longerlifetime Each communication task has an implementation time, relative deadline,and a time The duration of a WSN is increased by scheduling the vigorous interval

of devices It could affect the performance of the network for connectivity Rc/Rsratio has to be considered At each node in the system should form a linked envelop

to attain sensing coverage and system connectivity

General optimization carried at improved force effectiveness and improved networklifetime in wireless sensor network; two critical factors are information packet sizeand broadcast control level Conversely, the effect of slighter package size is disinte-gration into additional information packets and thereby indulgence of increased force[1,2] Wireless Sensor Network Lifetime (NL) is a critical metric since the antennanodes frequently rely on the limited power supply Cross layer network lifetime maxi-mization consider the joint best possible proposal of the substantial, Medium AccessControl (MAC), and system layers to exploit the NL of the power-constrained WSN[4] The problem of NL maximization can be formulated as a nonlinear optimizationproblem encompassing the routing stream, link scheduling, contact rate, and controlallocation operations for all energetic Time Slots (TSs) [5] The complexity of NLMaximization (NLM) can be formulated as a mixed integer-convex optimizationdifficulty with the implementation of the Time Division Multiple Access (TDMA)technique [3,6] The wireless sensor system topology, the convenience of resources,and the energy consumption of nodes in different paths of the data collection treemay vary largely, thus affecting the general network lifetime [8, 7] Cross-layerprotocol, which incorporates a multipath routing protocol and an information inter-leaving practice based on Reed–Solomon code Formulate the trouble of selectingsensor announcement paths as knapsack trouble and resolve it by a greedy algorithm[10,9] The multipath routing protocol then enables all sensors to select multiplestatement paths using the proposed optimization algorithm [11,12] On the basis

of numerous communication paths, the method of data interleaving is working byusing a Reed–Solomon code to provide reliable data transmission Simulation results

Maximizing Network Lifetime in WSN Using … 3

can display the available multipath routing protocols to the system lifetime since itbalances energy utilization and promotes communication reliability [13,14]

Routing is the method of selecting a path for traffic in several networks Directionfinding is performed for many types of networks, such as the path switched networks,public switched telephone network such as the Internet, as well as in networks used

in society and covert transportation, such as the organization of streets, roads, andhighways in universal communications In packet switching networks, routing is thehigher-level judgment construction that directs network packets from their sourcetoward their destination through middle network nodes by exact packet forwardingmechanisms

Traveling Salesman Problem belongs to a set of troubles in computational complexityanalysis called NP-complete problem If could find a way to declare an NP-completeproblem and can employ the algorithm to explain all NP problems rapidly TSP has

a number of applications silent in its purest formulation preparation, logistic, andcreates of microchips The challenge of the complexity is that the traveling salesmanneeds to reduce the total length of the trip The goal of the Traveling SalesmanProblem (TSP) is to find the majority competent way to tour a choose number of

“cities” Traveling Salesman Problem conserve be modeled as an undirected weightedgraph, such that cities are the graph’s vertices Figure1shows the TSP crisis to findthe minimum length of the path using the ACO algorithm The shortest path= a–d–e,Net weight= 22

Fig 1 Example of TSP

problem

an idle state due to or another local host Greedy algorithm preserve is characterized

as human life from tiny sight and in addition to non-recoverable A greedy session is atime-limited packet flow or data flow at the maximum possible rate A greedy sourcetransfer creation simulation model, or a greedy transfer generator, is useful whensimulating and analyzing or measuring the lifetime and throughput of a network.These locally optimal solutions will lastly include a globally best resolution Greedyalgorithm is to resolve the problem, it must be that the best explanation to the bigdifficulty contains the best possible solution to sub-problems

1 ACO is an inhabitants-based Meta-heuristic to facilitate can exist used to locateestimated solutions to the complicated optimization problem

2 In ACO, a location of a software agent called synthetic ants investigates excellentsolutions to a known optimization difficulty To apply ACO, the optimizationtrouble is transformed and addicted to the problem of finding the finest path on

a weighted diagram

3 ACO is a probabilistic technique searching for the finest pathway in the graphbased on the performance of ants seeking a path connecting their colony andbased on food Ants steer from shell to food source; ants are blind

Ant Colony Optimization (ACO) studies reproduction systems that obtain ulation from the performance of authentic ant colonies and which are used to estab-lish discrete optimization problems The method that ants discover their grub in thestraight path is interesting; ants are secreting pheromones to memorize their path.These pheromones disappear with time; whenever an ant finds food, it marks itsentrance journey with pheromones

stim-Pheromones disappear faster on longer paths Shorter paths make obtainable asthe way to food for most of the other ants The shorter path will be unbreakable bythe pheromones further Finally, the ants revolve up at the shortest path Ants leavepheromone trail when they make an adaptation trails that are used in prioritizingevolution The communication along with individuals, or stuck between individualsand the atmosphere, is based on the employ of chemicals produced by the ants Thesechemicals are called pheromones

Maximizing Network Lifetime in WSN Using … 5

Fig 2 Flowchart of traveling salesman predicament

ACO is proficient in solving the Traveling Salesman Problem (TSP) TSP is anNP-hard problem Figure2shows the Flowchart traveling salesman problem given

a set of n cities, the Traveling Salesman Problem requires a salesman to discoverthe shortest way between the given cities and go back to the starting city, whilekeeping in mind that each metropolitan preserve to be visited only once The mostprimitive ACO algorithms used the Traveling Salesman Problem (TSP) as an instance

application The TSP is characteristically represented by a graph G = (V, E), V is the set of integers of vertices, representing the cities, and E being the position of edges that completely connects the vertices To every edge (i, j) a distance dij is

associated The ACO algorithm, called Ant System (AS), has been applied to theTraveling Salesman Problem (TSP) TSP is also called the Hamiltonian circuit

Step 1: In the TSP, each ant finds paths in a network and every work is a node.

Moreover, no one has to return to the start node and the path is completed when eachnode is visited

Step 4: Construct Solutions: Each ant starts at a meticulous state, and then traverse

the states one by one

Step 5: Apply Local Search: Before updating the ant’s trail, a restricted search can

be applied to each solution constructed

Step 6: Update Trails: after the solutions are constructed and calculated, pheromone

levels increase and reduce on paths according to favorability

Step 7: Use d ij to indicate the detachment between any two cities in the difficulty

As such d ij = [(x i − x j)2+ (y i − y j)2]1/2

Step 8: Letτ ij (t) indicate the strength of the track on edge (i, j) at time t, at which

time every ant will have completed an expedition

Figure3 shows the shortest path identification to find the minimum length of thepath using the ACO algorithm

The ACO algorithm has been applied to the shortest path problem, and each antfinds paths in a network where each node is referred to as a node Upon completion

of every node, there is no intention to reach the start node once again Therefore, atotal number of available paths in the network and the corresponding path weight getcalculated

(a) Available Path

1 Path1—A–B–E–F–I–J

2 Path2—A–B–E–F–G–H–I

Fig 3 Shortest path

problem

Maximizing Network Lifetime in WSN Using … 7

No of Connected covers-7

The number of Connected Covers in Shortest Path Identification by GreedyAlgorithm is scheduled in Table1

Proteus mechanism to test, debug, and regulate a program is problematic In tion, greedy simulation allows non-intrusive monitoring and debugging, and alsomakes it simple to repeat executions Proteus provides users with unparallelededibility in choosing or customizing the period of accuracy in the wireless sensornetwork and memory simulations Proteus was initially designed for evaluatinglanguage, compiler, and runtime system mechanisms to carry portability SeveralCAD users release schematics detain as essential immorality in the procedure ofcreating printed circuit board (PCB) arrangement encloses the disputed in the peak

distinc-8 M D Saranya et al.

of investigation With PCB layout now offering automation of equally section ment and path routing, getting invent hooked on the computer can frequently be themain occasion overriding constituent of expending even more time working on theschematic Proteus provides repeatability, non-instructive monitoring and debugging,and incorporated graphical output The power of its structural design has allowed us

assign-to combine first conventional graph-based imitation and now with Proteus VSM—interactive circuit reproduction keen on the invent WSN environment For the firsttime eternally, it is sufficient to explain a total circuit for a TSP based Ant ColonyOptimization and then test it interactively It provides in general control of drawingemergence in turns of line widths, fills styles, colours and fonts

4.1.1 Features of Proteus

1 Regular wire steering and mark post/removal

2 Powerful equipment for selecting substances and passing on their property

3 Entire maintenance for buses including element pins, inter-sheet terminal,module ports, and wires

4 Schedule of equipment and Electrical Rules ensure report

5 Netlist output is to furnish all fashionable PCB layout tools

The power of its construction has allowed us to assimilate primary conventionalchart based imitation and now with Proteus VSM—interactive circuit replicationkeen on the plan surroundings Figure 4shows the schematic diagram of shortestpath identification using greedy in Proteus simulation

Figure5the simulation yield of shortest path identification using a greedy algorithmfor Proteus simulation It provides the whole administer of drawing appearance inturns of row thickness, pack style, and standard in addition to fonts

Maximizing Network Lifetime in WSN Using … 9

Fig 4 Schematic figure of shortest path using greedy

Fig 5 Simulation output of shortest path classification using greedy algorithm

10 M D Saranya et al.

Table 2 Power consumption

Path Weight Power loss transmit

5.3 Power Analysis of Greedy Algorithm

The amount of power utilization for communicating through dissimilar nodes insidethe range is listed in Table2

The relationships connecting the numbers of nodes and initial energy capacity using

a greedy algorithm, and the results of the greedy algorithm is shown in Fig.6In thisfigure, it is obvious that the network life span increases along with the increasinginitial energy capacity

Fig 6 Numbers of nodes and initial energy capacity using a greedy algorithm

Maximizing Network Lifetime in WSN Using … 11

Table 3 Power analysis of

loss/transmit mW

Power loss/receive mW

The sum of power utilization for communicating with dissimilar nodes surrounded

by the variety is listed in Table3

Figure7shows the relations connecting the system lifetime initial power capacitiesalong with different algorithms; it is clear that the net lifetime increases and extendsthe life series of the wireless sensor network since it manages the energy and controlmanagement

Fig 7 Relationships between the network lifetime and Initial energy capacity along with different

In this project also, using the competent ant colony algorithm has been functional

to the traveling salesman problem to locate the optimal solution in a short time.However, the performance of the ACO algorithm is designed for both high energyefficiency and good power balancing, and maximal energy utilization throughout thenetwork The relationship between the statistics of nodes and original power capacityusing ACO algorithm is evident that the network lifetime increases and extends thelife cycle of the wireless sensor network since it manages the energy and powermanagement

In the future, another proposed ACO algorithm will be regularly considered to recoverthe network lifetime and communication reliability, for example, the active adaptation

of multiple routing problems that are essentially multiple paths multiple destinations

of the original TSP; this is similar to vehicle routing problems For these problems,numerous routes are considered, which makes them closer to real-world applications

References

1 Kakhandki AL, Hublikar S, Priyatamkumar (2018) Energy efficient discriminatory hop tion optimization toward maximize lifetime of wireless sensor networks Alexandria Eng J 57(2):711–718

selec-2 Akbas A, Yildiz HU, Tavli B, Uludag S (2016) Joint optimization of transmission power level and packet size for WSN lifetime maximization IEEE Sens J 16(12):5084–5094

3 Deng W, Xu J, Zhao H (2019) An improved ant colony optimization algorithm based on hybrid strategies for scheduling problem IEEE Access 5(7):20281–20292

4 Lin Y, Zhang J (2012) An ant colony optimization approach used for maximizing the lifetime

of heterogeneous wireless sensor networks IEEE Trans Syst 42(3):408–420

5 Sun Y, Dong W, Chen Y (2016) An improved routing algorithm based on ant colony optimization inside wireless sensor networks IEEE Commun Lett

Maximizing Network Lifetime in WSN Using … 13

6 Bagula A, Mazandu K (2008) Energy constrained multipath routing in wireless sensor networks In: Proceeding of the 5th international conference on ubiquitous intelligence and computing, Oslo, Norway, pp 453–467

7 Chang CT, Chang CY, Zhao S, Chen JC, Wang TL (2016) SRA: a sensing radius adaptation mechanism for maximizing network lifetime in WSNs IEEE Trans Veh Technol 65(12):9817– 9833

8 Fonseca R, Gnawali O, Levis K (2007) Four-bit wireless link estimation In: Proceedings of the 6th workshop taking place hot topics in networks (Hot Nets VI), Atlanta, GA, USA

9 Yetgin H, Cheung KTK, El-Hajjar M, Hanzo L (2014) Cross-layer network lifetime tion in interference-limited WSNs IEEE Trans Veh Technol 64(8):3795–3803

maximiza-10 Wang H, Agoulmine N, Ma M, Jin Y (2010) Network lifetime optimization in wireless sensor networks IEEE J Sel Areas Commun 28(7):1127–1137

11 Cohen K, Leshem A (2010) A time-varying opportunistic approach to lifetime maximization

of wireless sensor networks IEEE Trans Sig Process 58(10):5307–5319

12 Lin K-Y, Wang P-C (2010) A greedy algorithm in WSNs intended for maximum network lifetime and communication reliability In: IEEE 12th international conference on networking, sensing and control Howard Civil Service International House, Taipei, Taiwan

13 Younis M, Senturk I (2012) Topology management techniques for tolerating node failures in wireless sensor network A review Comput Netw pp 254–283

14 Imon SK, Khan A, Di Francesco M, Das SK (2014) Energy-efficient randomized switching for maximizing lifetime in tree-based wireless sensor networks IEEE/ACM Trans Netw 23(5):1401–1415

Deep Ensemble Approach for Question

Answer System

K P Moholkar and S H Patil

Abstract Researches on question answering systems has been attracting significant

research attention in recent years with the explosive data growth and breakthroughs

in machine learning paradigm Answer selection in question answering segment isalways considered as a challenging task in natural language processing domain Themajor difficulty detected here is that it not only needs the consideration of semanticmatching between question answer pairs but also requires a serious modeling ofcontextual factors The system aims to use deep learning technique to generate theexpected answer Sequential ensemble approach is deployed in the proposed model,where it categorically boosts the prediction of LSTM and memory network to increasethe system accuracy The proposed model shows a 50% increase in accuracy whencompared to individual systems with a few number of epochs The proposed systemreduces the training time and boosts the system-level accuracy

Keywords Question answer system·Deep neural network·LSTM·Memory·Network ensemble·CatBoost

In general, people tend to seek information through conversation, Internet, books,etc Due to information overload, the process of extracting the required answers usingmachine learning techniques has become a challenging task The task become morecomplicated when the users have started to post different types of questions like

Wh, short question, factoid, reasoning questions, counting, etc A system is alwaysexpected to learn facts and apply reasoning to discover new facts that can assist

in answering the given question Extraction and identification of suitable answer

© The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd 2021

A Pasumpon Pandian et al (eds.), Computer Networks, Big Data and IoT, Lecture Notes

on Data Engineering and Communications Technologies 66,

https://doi.org/10.1007/978-981-16-0965-7_2

15

16 K P Moholkar and S H Patil

depends on understanding the main intent of a question This makes the task ofbuilding an appropriate model as a challenging task The conventional method ofquestion system will have data retrieval and handcrafted rules The performance oftraditional systems is restricted as they heavily depend on manually extracted rule.The task becomes difficult because of imbalanced dataset The QA process can bebroken into two parts: information retrieval and reading comprehension The process

of finding the document containing the answer for the raised question is known asinformation retrieval Reading comprehension deals with identifying answer from

a given the passage However, the recent schemes on deep learning have showntheir potential in improving the performance of QA system Deep learning modelsrequire a substantial amount of training GRU, LSTM, and bidirectional LSTM help

to handle longer text sequences unrolled during time Enhancements like attentionmechanism and memory networks help the model to identify important features.Dialog systems [8] like chat bots [9], Alexa, and IBM Watson [10] which simulatehuman conversation are derived from question answer systems With the increase indigital awareness and reach in rural area, the need of multilingual question answersystem has started arising Availability of multilingual QA corpus is a challengingtask Until recently, multilingual QA system was dependent on machine translation.Identifying appropriate answers and framing syntactically and semantically correctsentence require consideration of language-specific grammar, which is also difficulttask This would further affect the accuracy of system Therefore, advancements in

QA system are the need of the hour Ensemble models help to combine the strengths

of individual models to enhance the accuracy of system The major contribution ofthe paper is an ensemble model which categorically boosts the combined predictions

of LSTM model and memory model The ensemble model is designed to boost theaccuracy of question answering system capable of handling different question cate-gories The proposed research work has been arranged as follows: Section2provides

a brief review of literature Section3explains the overview of the proposed questionanswering system Section4presents experimental setup Section5 illustrates theresults and discussion, and Sect.6presents the conclusion

Abishek et al [1] proposed a CNN—LSTM model for identifying duplicate tions using Quora dataset The author proved that identifying duplicate questionsbefore the classification enhances the performance of model tremendously Bae et

ques-al [2] employed word weighting method for question classification The author posed mechanism to identifying positive and negative words, and positive wordshaving negative sentiments, and vice versa for query processing Razzaghnoori et

pro-al [3] experimented with recurrent neural network (RNN) for extracting featurevector mechanism and employed classical support vector machine (SVM) for queryclassification Hong et al [4] have proposed a model that constructs a binary treestructure to take care of composite reasoning implied in the language The recursive

Deep Ensemble Approach for Question Answer System 17

grounding tree (RVG-TREE) parses the language and then performs a visual ing along the tree in a bottom-up fashion Liu [5] proposed a Siamese architecturethat compromises of multilayer long short-term memory (LSTM) network and con-volutional neural networks (CNN) The author uses concept interaction graph formatching long text The graph vertex acts as a concepts, and interaction level act

reason-as edges The model aggregates the matching signals through a graph convolutionalnetwork which checks similarity relationship between two articles Ullah et al [6]proposed an attention model that emphasizes on the importance of context and timethat helps to provide contextual information regarding the question while generatingthe answer representations In 2017, Yue et al [7] developed a dynamic memory net-works to carry out “textual QA.” The model takes the inputs that are processed to takeout hierarchical and global salient features simultaneously Consequently, they weredeployed to formulate numerous feature sets at every interpretation phases Exper-imentation was performed on a public textual Q&A dataset using without and withsupervision approach from labels of constructive details Finally, when distinguishedwith preceding works, the developed technique demonstrates improved stability andaccuracy

LSTM helps to defend the error that is backpropogated through time LSTM is thememory element in the model A LSTM cell consists of neural network with threedifferent activation functions LSTM cell has three basic gates which are neural

networks with sigmoid activation function known as input gate ‘i’, forget gate ‘f ’, and output gate ‘o’ Neural net with Tanh activation is known as candidate layer ‘c’ Output is a current hidden state id denoted by vector ‘h’, and memory state denoted

by vector ‘c’ Forget layer ‘F’ decides which information to discard Inputs to the LSTM cell at each step are X the current input and H the previous hidden state, and C is the previous memory state Outputs from the LSTM cell are H the current hidden state and C the current memory state The following equations demonstrate

the working of LSTM layers

f t = σ (W f [h(t − 1), x t ] + b f ) (1)

i t = σ (W f [h(t − 1), x t ] + b i ) (2)

˜C = tanh(W C [h(t − 1), x t ] + b C ) (3)

C = f ∗ C(t − 1) + i ∗ ˜C) (4)

18 K P Moholkar and S H Patil

Fig 1 LSTM cell

o t = σ(W o [h(t − 1), x t ] + b o ) (5)

h t = o t ∗ tanh(C t ) (6)Figure1demonstrates the working of LSTM cell The cell compares its memory withprevious output of LSTM cell to identify the importance and reads a current wordconsidering previous

For a QA system to be efficient, inference mechanism and memory elements arerequired to memorize long sequences with context These long sequences serve asknowledge for effective prediction of answers This model decodes the context andquestion in fixed length vector and produces the expected answer This model is also

known as encoder–decoder model Consider a network with M = μ1, μ2, , μ n

memory elements, I input feature of vector of size e, u update parameter for ing old memory with present input, O the output feature map, and R the expected

updat-response The encoder processes the input feature and produces a two-dimensionaloutput matrix The model takes input as sentence and stores in available memoryμ.

The input length depends on the number of memory cells in the layer A new

mem-ory is stored in u module without changing the old memmem-ory μ LSTM layer acting

as a decoder takes input vector, the time steps, and features to produce a decodedsequence, i.e., answer (Fig.2)

Deep Ensemble Approach for Question Answer System 19

Fig 2 Memory network

Algorithm 1: Memory Cell

Data: input text character/word/sentence

Result: Memory Network Algorithm

CatBoost is an implementation of gradient boosting, which uses binary decision trees

as base predictors [12] Existing gradient boosting algorithms like random forest,Adaboost, and Xgboost suffer from prediction shift problem Boosting algorithmrelies on target of all training examples which causes prediction shift Traditionalapproach [14] converts categorical features to statistical model leading to targetleakage CatBoost algorithm provides solution to both the said issues by employingprinciple of ordering CatBoost is based on gradient boosted decision trees Decision

20 K P Moholkar and S H Patil

trees are built repetitively during training phase Each successive tree is built withreduced loss compared to the previous trees When the algorithm identifies overfitting, it stops construction of new tress

Algorithm 2: Categorical Boosting

Data: input text character/word/sentence

Result: tree

Preliminary calculation of splits.;

Choosing the tree structure.;

while not over-fitting do

Divide objects into disjoint ranges;

Split data into bucket;

Select tree structure;

Calculating values in leaves.;

while not over-fitting do

Calculate penalty function;

to train and optimize the results CatBoost takes more time to train a system withnumerical features It does not support sparse matrices

A deep neural network (DNN) learns complex nonlinear relationship in data Bias,variance, and noise are prediction errors While noise cannot be handled by algorith-mic approach, bias and variance can be reduced by proper choice of model Everytime a DNN is trained, different version of mapping function is learnt This stochas-tic nature affects the performance of model The DNN suffers from low bias andhigh variance The proposed model uses ensemble approach to reduce the problem

of high bias Sequential ensemble approach is used in the proposed system to reducebias error This approach combines predictions from base learners and applies cat-egorical boosting to enhance the accuracy of system Several instances of the samebase model are trained sequentially to train the LSTM learner with previous weaklearners Figure3shows the working of the model Long short-term recurrent neuralnetwork and memory networks are used as base learners Categorical boosting isdone by CatBoost algorithm which is used as meta learners to enhance the accu-racy of ensemble model The proposed model works in two phases The learning

Deep Ensemble Approach for Question Answer System 21

Fig 3 Proposed model

algorithms for LSTM are based on minimizing the error rate The first phase usesdata subsets to produce results The results generated by first phase are combinedtogether, and performance boost is achieved by categorical boosting (Fig.3)

LSTM and memory networks are trained in full supervision by providing context,question, and expected answers to the system Each model is tested in isolation foraccuracy and loss parameters Evaluation of the proposed system is done on babiquestion answering dataset The dataset consists of 20 different question answeringtasks The models are trained with 40 epochs and batch size of 32 The model is trained

on 1000 samples A dropout of 0.3 is applied to avoid over fitting The efficiency of

a QA system is judged by the correctness of answer predicted Precision and recallare traditional factors for judging a classification problem Apart from these, meanreciprocal rank (MRR) is the ability of a QA system to select appropriate answer for

22 K P Moholkar and S H Patil

a given question from the ranked list of answers If suitable answer is not found, thenrank becomes zero The proposed model was evaluated on precision, recall, and f1measure

Precision(Micro) = No of correctanswers for Question q

answers retrieved for Question q (7)

Recall(Micro) = Correct answers retrieved for Question q

gold standard answers retrieved for q (8)

Precision= Average of Precision for all questions (9)Recall= Average of Recall for all questions (10)F1 Score(Micro) = Harmonic mean of Precision (Micro) and Recall (Micro)

(11)F1 Score= Harmonic Mean of Precision and Recall (12)

(13)

From experiments, it was observed that each classifier requires different settings ofhyperparameters for producing best model With fixed hyperparameter setting, indi-vidual performance of classifier varies a lot The proposed ensemble boosting helps

to overcome this bottleneck Considerable amount of time needs to be invested toidentify optimal hyperparameters In cases where single classifier is not sufficient

to correctly predict the result, the proposed approach provides a better alternative.Metric classification report indicates that proposed ensemble model performs bet-ter than individual models The accuracy of LSTM model is obtained around 60%and memory network model around 60% for 40 epochs The accuracy of proposedensemble model obtained is 95.45% The f1 score of proposed model is 93% Aconsiderable boost in accuracy is observed in the result (Table1)

Table 1 Comparing different models on precision, recall, and f1 score

Deep Ensemble Approach for Question Answer System 23

The ROC curve is plotted with true positive rate on y-axis and false positive rate

on x-axis The ROC curve demonstrates the probability distribution of system

Clas-sifiers that produce curves closer to the top-left corner indicate a better performance.The ROC curve and FPR FNR curve indicate that the performance of proposed model

is stable FNR indicates the miss rate for given data (Figs.4and5)

Fig 4 ROC curve

Fig 5 FPR/FNR curve