PREDICTING K-COVERAGE IN WIRELESS MULTIHOP

NETWORKS WITH BOUNDARY EFFECTS USING SUPPORT

VECTOR REGRESSION: A FEATURE SENSITIVITY
ANALYSIS FOR SMART CITY APPLICATIONS

A Dissertation Part-2 report submission for the partial fulfillment of the requirement for the
award of the degree

FIVE YEAR INTEGRATED [Link]. (CSE) - [Link]. (DATA

SCIENCE)

Submitted By: -
Name: Sachin Singh
Roll No: 20/ICS/028
UNDER THE SUPERVISION OF
Dr. RAJESH MISHRA
(Assistant Professor)

DEPARTMENT OF COMPUTER SCIENCE AND ENGINEERING

UNIVERSITY SCHOOL OF INFORMATION AND COMMUNICATION
TECHNOLOGY
GAUTAM BUDDHA UNIVERSITY
GREATER NOIDA – 201312, GAUTAM BUDDHA NAGAR
UTTAR PRADESH, INDIA

May 2025
 UNIVERSITY SCHOOL OF INFORMATION AND COMMUNICATION
TECHNOLOGY
GAUTAM BUDDHA UNIVERSITY, GREATER NOIDA, 201312, U. P., (INDIA)

Candidate’s Declaration
I, Sachin Singh, hereby certify that the work embodied in this report entitled
“Predicting k-Coverage in Wireless Multihop Networks with Boundary
Effects Using Support Vector Regression: A Feature Sensitivity Analysis for
Smart City Applications” in partial fulfillment of the requirements for the award
of the Five Year Integrated [Link]. (CSE) - [Link]. (Data Science) submitted to
the School of Information and Communication Technology, Gautam Buddha
University, Greater Noida, is an authentic record of my own work carried out
under the guidance of Dr. Rajesh Mishra, Assistant Professor, Department of
Computer Science and Engineering, School of Information and Communication
Technology.

The matter presented in this report has not been submitted to any other
University/Institute for the award of any other degree or diploma. Responsibility
for any plagiarism-related issue stands solely with me.

Date: 11th March 2025

Sachin Singh
(20/ICS/028)
II
 UNIVERSITY SCHOOL OF INFORMATION AND COMMUNICATION
TECHNOLOGY
GAUTAM BUDDHA UNIVERSITY, GREATER NOIDA, 201312, U. P., (INDIA)

Certificate
This is to certify that the work titled “Predicting k-Coverage in Wireless
Multihop Networks with Boundary Effects Using Support Vector Regression:
A Feature Sensitivity Analysis for Smart City Applications” submitted for the
award of the Five-Year Integrated [Link]. (CSE) -[Link]. (Data Science) degree
from the Department of Computer Science and Engineering, University School of
Information and Communication Technology, Gautam Buddha University,
Greater Noida (UP), has been carried out under my supervision.

The contents of this report are not submitted as part of any other degree award to
the candidate or any other individual. This report represents original research and
study conducted solely by the student. Any responsibility for issues related to
plagiarism rests entirely with the student.

Dr. Rajesh Mishra

Assistant Professor
Department of Computer Science and Engineering
School of Information and Communication Technology
Gautam Buddha University
Greater Noida 201312
III
 Acknowledgement

I have made efforts in this report. However, it would not have been possible
without the kind support and help of several individuals and organizations, and I
would like to extend my sincere thanks to all of them.

I am deeply indebted to Dr. Rajesh Mishra for his invaluable guidance and
constant supervision, as well as for providing essential information and support
throughout the preparation of this report.

I would also like to thank Dr. Arpit Bhardwaj, Dean of School of Information
and Communication Technology and Dr. Arun Solanki, Head of the
Department of Computer Science and Engineering, for their guidance and
support in the completion and evaluation of my report on the topic, “Predicting
k-Coverage in Wireless Multihop Networks with Boundary Effects Using
Support Vector Regression: A Feature Sensitivity Analysis for Smart City
Applications”. Their encouragement and insights greatly contributed to the
successful completion of this work.

Lastly, I wish to express my gratitude to my friends and family members for their
cooperation and encouragement, which has been instrumental in the completion
of this report.

IV
 Abstract

Achieving the optimal network coverage is very essential for wireless multihop
networks (WMNs), mainly in finite rectangular shape areas where the boundary
effects (BEs) can distort the coverage metrics. Traditionally the simulations
approach many times struggle very much to scale effectively as the network
parameters grow increasingly complex.

In this study, we used a Support Vector Regressor (SVR) model for prediction of
the k-coverage probability in those scenarios where we see both with and without
boundary effects. The model we used is trained using key features such as region
dimensions, sensing ranges, and the density of node. Compared to other models
like GRNN, Random Forest and Gradient Boosting, the SVR is demonstrating
superior performance. It is acquiring a root mean square error (RMSE) of 0.1287
and R² value of 0.7752 in the scenarios with boundary effects, and an RMSE of
0.1371 with an R² value of 0.7610 in scenarios without them. Additionally, our
feature sensitivity analysis gives valuable insights for the optimization of network.
This makes the proposed framework not only scalable but also highly accurate for
multiple applications, including urban sensor networks, disaster management,
proving effective for applications and other critical infrastructure monitoring.

V
 List of Abbreviations

Abbreviation Full Form

ML Machine Learning
MSE Mean Squared Error
R2 R-Squared Score
k-coverage k-coverage Probability
SVM Support Vector Machine
SVR Support Vector Regressor
RF Random Forest
GBM Gradient Boosting Model
NN Neural Network
IoT Internet of Things
CSE Computer Science and
Engineering
WSN Wireless Sensor Network

VI
 List of Figures
Figure No. Title Page
No.
Figure 1.1.1 A diagram of a multi-hop 2
network layout, showing sensor
nodes with overlapping
coverage areas and illustrating k-
coverage.
Figure 1.7.1 Bar Graph Comparison of Model 9
Performance (MSE and R2)
Figure 1.7.2 Line Graph Comparison of 9
Model Performance (MSE and
R2)
Figure 1.8.1 Gives the overall Research 10
Structure for the proposed work
Figure [Link] Supervised vs Unsupervised 25
Learning
Figure [Link] Working of Gradient Boosting 28
Figure 5.2.1 Heatmap of Model Performance 39
Summary
Figure 5.3.1 Feature Importance Analysis 41
Figure [Link] Model Performance Comparison 42
using Stacked Bar Graph
Figure 6.1.1 Model Performance Comparison 45
using Grouped Bar Graph

VII
 List of Tables
Table No. Title Page
No.
Table 1.5.1 Dataset Overview 6
Table 1.6.1 Model Characteristics Table 8
Table 2.2.1 Literature Review Table 18
Table [Link] Comparison of Supervised Learning 24
Algorithms
Table 4.2.1 Dataset Overview 32
Table 5.2.1 Model Performance Comparison 38
Table 5.3.1 Performance Comparison of Machine 40
Learning Models vs. GRNN
Table 6.2.1 Future Research Directions and 47
Benefits

VIII
Candidate’s Declaration………………………………………………………….....…II
Certificate……………………………………………………………………………..III
Acknowledgement…………………………………………………………….…..…..IV
Abstract…………………………………………………………………………….….V
List of Abbreviation……………………………………………………………….….VI
List of Figures…………………………………………………………..…………….VII
List of Tables………………………………………………………………………...VIII

TABLE OF CONTENTS

Chapter 1: INTRODUCTION………………………………………….…..01-10
1.1 Introduction.............................................................................................................. 02
1.2 Problem Statement................................................................................................... 03
1.3 Objectives of the Study............................................................................................ 04
1.4 Significance of the Study......................................................................................... 05
1.5 Methodology Overview........................................................................................... 05
1.6 Machine Learning Models Overview....................................................................... 07
1.7 Evaluation Metrics................................................................................................... 08
1.8 Thesis Structure....................................................................................................... 10
1.9 Summary................................................................................................................... 11

Chapter 2: LITRATURE REVIEW………………………………………..13-20

2.1 Main Base Paper…………………………………………………………..……….14
2.2 Related Literature…………………………………………………….………..…..14
2.3 Summary ……………………………………………………………………..…... 20

Chapter 3: METHODS AND METHODOLOGIES………………………22-29

3.1 Introduction…..……………………………………………………….…..……….23
3.2 Types of Machine Learning……………………………………………………..…23
3.3 Models …………………………………………………...………...………….......25
3.4 Tools and Programming Languages ………………………………….……………28
3.5 Summary……………………………………………………...….…………..…….29
Chapter 4: K-COVERAGE PROBABILITY PREDICTION MODEL .30-35
4.1 Introduction……………………………………………………………….…….....31
4.2 Dataset……………………………………………………..……………...…….....31
4.3 Process Flow……………………………………………………………………….33
4.4 Model Training……………………………………………………….……………34
4.5 Summary…………………………………………………………..……………….35

Chapter 5: RESULTS AND ANALYSIS………………………..……….…36-43

5.1 Introduction………………………………………………………….………….....37
5.2 K-Coverage Probability Prediction Model…………….……….……………….....37
5.3 Comparison with the Existing Model………………………………..….………….40
5.4 Results…………………………………………………………………..…………41
5.5 Summary……………………………………………………….………………….43

Chapter 6: CONCLUSION AND FUTURE SCOPES……...………44-47

6.1 Introduction…………………………………………………………….……….....45
6.2 Future Scopes………………………………………….……….……………..……46
6.3 Application of this Study…………...………………………………..…………….47

REFRENCES………………………………………………………………..49-55
APPENDICES……………………………………………………………….56-65
Appendix A – Similarity Score
Appendix B – Paper Communication Mail and Presentation Certificate
Appendix C – Communicated Paper
 CHAPTER – 1
INTRODUCTION

1
1.1 Introduction
With the rapid development of Wireless Sensor Networks (WSNs), many sectors such as
environmental monitoring, healthcare, industrial automation, and smart cities now rely on
interconnected sensor nodes to collect and transmit data [1]. These nodes operate via a
multihop network, where data is transmitted across multiple intermediary nodes to reach a
central hub or base station [2]. To ensure robust and reliable system performance, WSNs are
designed to provide k-coverage—a concept ensuring every point in the target area is covered
by at least k sensor nodes [3].

The k-coverage probability (KCP) represents the likelihood that any arbitrary point in a
monitored area is within the sensing range of k nodes [4]. A higher value of KCP implies
better fault tolerance, redundancy, and data reliability, as multiple sensors validate and relay
the same information [5]. This is especially critical in sensitive environments such as military
operations, healthcare monitoring, and disaster management, where uninterrupted, accurate
data collection is vital for operational safety [6].

Figure 1.1.1 A diagram of a multi-hop network layout, showing sensor nodes with
overlapping coverage areas and illustrating k-coverage.

2
1.2 Problem Statement
WSNs play a pivotal role in real-time sensing, monitoring, and automation tasks. Ensuring
k-coverage is essential for maintaining reliable sensing and communication [7]. However,
predicting KCP with accuracy is challenging due to various factors:

• Node mobility and density [8]

• Connectivity fluctuations

• Environmental interference (e.g., physical obstructions, weather)

• Node deployment randomness [9]

These factors contribute to uncertainty in network performance and coverage reliability.

Conventional analytical/statistical techniques often assume ideal conditions and static
configurations, which do not hold in real-world deployments [10]. Moreover, these
techniques typically fail to account for:

• Boundary Effects (BEs): Reduced sensing/coverage near the network's physical edges
[11]

• Shadowing Effects (SEs): Signal attenuation due to obstacles or terrain variations [12]

Such effects lead to overestimation of coverage metrics and inefficient sensor deployment.
Additionally, traditional techniques are often computationally expensive and unsuitable for
large-scale networks or real-time predictions [13].

In contrast, machine learning (ML) models are now being explored to address these
challenges by:

• Learning from historical or simulated data

• Capturing non-linear relationships

• Supporting real-time decision-making [14]

3
However, there is a lack of consensus on which ML models are best suited for KCP prediction
under varying conditions. With each model having distinct advantages and limitations, a
systematic comparative evaluation is essential.

This study aims to bridge this gap by evaluating four prominent ML models: Random Forest
(RF), Gradient Boosting (GB), Support Vector Regressor (SVR), and Neural Network
(NN).

Key Research Questions:

• How accurately can each model predict KCP under varying environmental and network
conditions?

• What are the computational demands of each model, and are they feasible for real-time
application?

• How well does each model generalize across different node deployment patterns and
network geometries?

1.3 Objective of the Study

This research aims to advance the understanding and practical deployment of ML models for
KCP prediction in Wireless Multihop Networks (WMNs). It does so by comparing model
performance and proposing improvements where necessary.

Specific Objectives:

1. Evaluate the Performance of Existing ML Models

o Assess prediction accuracy and computational efficiency of models like RF, GB,
SVR, and NN on KCP estimation.

2. Conduct Comparative Analysis

4
 o Use performance metrics (MSE, R²) to compare existing models against a
proposed AI-enhanced approach.

3. Design a Novel AI-Driven Algorithm

o Integrate advanced techniques (e.g., feature importance, regularization) to

improve prediction accuracy.

4. Feature Engineering and Dataset Enhancement

o Include critical parameters such as node density, sensing range, boundary effect
indicators, and network size to boost model reliability.

1.4 Significance of the Study

The outcomes of this study are expected to:

• Provide a scalable, accurate framework for predicting KCP in dynamic environments

• Enable better energy-efficient and cost-effective sensor deployments

• Inform IoT infrastructure, smart city, and cyber-physical system designers on best
practices for robust coverage estimation

Further, by highlighting model strengths and weaknesses under various scenarios

(with/without BEs), this study supports real-world optimization of WSN deployments.

1.5 Methodology Overview

The research methodology involves a structured approach to comparing ML models,
utilizing a dataset that represents various WSN configurations and network conditions.

1. Data Collection and Preparation: The dataset contains variables influencing KCP,
including sensor node count (SN Count), network dimensions (Length, Breadth), and
Boundary Effects (BEs). Data preprocessing steps include feature scaling and splitting
data into training and testing sets to ensure reliable model evaluation.

5
 2. Model Training and Evaluation: Four ML models—Random Forest (RF), Gradient
Boosting (GB), Support Vector Regressor (SVR), and Neural Network (NN)—are
trained and evaluated on the dataset. Model performance is measured using the following
metrics:

o Mean Squared Error (MSE): Reflects prediction error, with lower values indicating greater
model accuracy.

o R² Score: Indicates the variance explained by each model, with higher scores reflecting
better predictive accuracy.

3. Feature Importance and Sensitivity Analysis: To understand each model's sensitivity

to different network features, Partial Dependence Plots (PDPs) and Individual
Conditional Expectation (ICE) curves are used. These analyses provide insights into
the most influential factors for KCP prediction.

Table 1.5.1: Dataset Overview

Feature Name Description Range Importance

Score
Sensor Node Total number of sensor nodes in Low to High High
Count the network
Network Length of the WSN deployment Low to High Medium
Length area
Network Breadth of the WSN Low to High Medium
Breadth deployment area
Boundary Effects of the network boundary 0 (no) / 1 (yes) High
Effects on coverage
Shadowing Signal Blockage by obstacles 0 (no) / 1 (yes) High
Effects

6
1.6 Machine Learning Models Overview
This study utilizes four distinct machine learning (ML) models known for their strengths
in handling complex regression tasks and capturing non-linear patterns within data. Each
model brings specific advantages and limitations, making them suitable for evaluating k-
coverage probability (KCP) in wireless sensor networks (WSNs):

• Random Forest (RF): An ensemble learning model that combines multiple decision
trees to reduce overfitting and increase accuracy. RF captures complex patterns in data
effectively and performs well with high-dimensional datasets. However, it can be less
interpretable than simpler models due to its ensemble nature.

• Gradient Boosting (GB): This sequential ensemble method builds models iteratively,
where each new model corrects the errors of the previous one. GB is powerful for datasets
with complex interactions, but it requires careful tuning to prevent overfitting and is
computationally intensive, which can extend training time.

• Support Vector Regressor (SVR): An extension of Support Vector Machines (SVM)

tailored for regression tasks, SVR can capture non-linear relationships in data and
performs well with smaller datasets. However, it may struggle with large datasets due to
higher computational demands, making it less scalable than other models.

• Neural Network (NN): A deep learning model with multiple layers capable of
identifying intricate relationships between input features. NNs are highly flexible and
adaptive but require larger datasets and higher computational resources to perform
optimally. They are often less interpretable than traditional models.

7
Table 1.6.1: Model Characteristics Table

Model type Strengths Limitations Computational

Requirements
Random Forest Resilient to Limited Moderate
(RF) overfitting, interpretability
captures complex
patterns
Gradient Accurate with High training High
Boosting (GB) complex time, needs
interactions, tuning
flexible
Support Vector Robust to Not scalable for Moderate to High
Regressor (SVR) overfitting, large datasets
effective with
smaller datasets
Neural Network Capable of Requires large Very High
(NN) capturing datasets, high
intricate computation
relationships

1.7 Evaluation Metrics

• Mean Squared Error (MSE): This metric measures the average of squared differences
between predicted and actual values, providing an estimate of prediction error. Lower
MSE values indicate higher accuracy, as they reflect smaller deviations between
predictions and actual observations.

8
• R² Score: The R² Score represents the proportion of variance in the target variable
explained by the model, with values closer to 1.0 indicating a better fit. It offers insights
into how well the model can capture underlying patterns in the data, with higher scores
reflecting greater predictive accuracy.

Figure 1.7.1 Bar Graph Comparison of Model Performance (MSE and R2)

Figure 1.7.2 Line Graph Comparison of Model Performance (MSE and R2)

9
1.8 Thesis Structure
To ensure a coherent flow, the thesis is organized as follows:

Literature Review

Problem Identification

Process Planning

Data Collection and Preprocessing

Model Implementation

Results and Discussion

Implications and Future Work

Conclusion

Figure 1.8.1 Gives the overall Research Structure for the proposed work

10
• Chapter 2: Literature Review – Reviews existing research on k-coverage, KCP
prediction, machine learning applications in WSNs, and relevant methodologies. It
includes a comparison of previous studies, highlighting gaps addressed in this research.
• Chapter 3: Methodology and Data Preprocessing – Describes the dataset, feature
selection, preprocessing techniques, and an overview of the machine learning models
used (RF, GB, SVR, GRNN). It includes a detailed discussion on data preparation, feature
scaling, and exploratory data analysis.
• Chapter 4: Model Implementation – Explains the implementation of machine learning
models for KCP prediction, including training procedures, hyperparameter tuning,
evaluation criteria, and computational setup. It also includes a flowchart illustrating the
process.
• Chapter 5: Results and Analysis – Presents experimental results, including model
performance evaluation using MSE, RMSE, and R² scores. Comparative analysis of
models, feature importance analysis, and visual representations (heatmaps, bar charts)
are provided.
• Chapter 6: Conclusion and Future Scope – Summarizes key findings, WSN
optimization, and practical applications of KCP prediction.
• References – Lists all the cited research papers and sources in accordance with academic
standards

1.9 Summary
This chapter introduced the fundamental concepts and motivation behind the study on k-
coverage probability (KCP) prediction in Wireless Sensor Networks (WSNs). It outlined the
significance of KCP in ensuring network reliability, data accuracy, and fault tolerance in real-
world applications such as healthcare, military surveillance, and smart city infrastructure.

The problem statement highlighted the challenges of accurately predicting KCP due to
factors such as boundary effects, shadowing effects, and dynamic network conditions.

11
Traditional analytical methods were found to be computationally expensive and limited in
adaptability, leading to the adoption of machine learning (ML) models as a more effective
alternative.

The research objectives were defined, emphasizing a comparative evaluation of different ML

models (Random Forest, Gradient Boosting, Support Vector Regressor, and General
Regression Neural Network) to determine the most suitable approach for KCP prediction.
The significance of the study was also discussed, highlighting its potential impact on WSN
optimization and broader applications in IoT and cyber-physical systems.

Furthermore, an overview of the research methodology was presented, detailing data

collection, preprocessing, model training, and performance evaluation metrics. The chapter
concluded with an outline of the thesis structure, ensuring a clear and logical flow for the
subsequent chapters.

This study aims to bridge the gap between theoretical modeling and practical applications by
leveraging data-driven approaches to enhance WSN efficiency and scalability. The next
chapter provides an in-depth literature review, examining previous research and
methodologies relevant to KCP prediction in wireless networks.

12
CHAPTER – 2
LITRATURE
REVIEW

13
2.1 Main Base Paper
1. A Machine Learning Approach to Predict the 𝑘-Coverage Probability of Wireless
Multihop Networks Considering Boundary and Shadowing Effects
Authors: Jaiprakash Nagar, Sanjay Kumar Chaturvedi, Sieteng Soh, Abhilash Singh

This foundational paper emphasizes the importance of k-coverage probability (KCP) in

Wireless Multihop Networks (WMNs) as a critical measure of network reliability and
Quality of Service (QoS). Traditional approaches to coverage estimation often ignore
boundary effects (BEs) and shadowing effects (SEs), leading to optimistic and
inaccurate predictions. The authors propose a machine learning-based solution using a
Generalized Regression Neural Network (GRNN) to address this challenge [10].

Their model is trained using six significant input parameters: network length, breadth,
sensing range, sensor node count, shadowing standard deviation, and desired k-value.
Using ensemble regression trees for sensitivity analysis, the model achieves a
correlation coefficient (R) of 0.78 and RMSE of 0.14 with BEs, demonstrating
improved prediction accuracy and reduced computational complexity compared to
traditional analytical methods [10]. This paper forms the cornerstone for the present
research.

2.2 Related Literature

2. 𝜅-Coverage Reliability for Wireless Multihop Network incorporating Boundary

Effect
Authors: Bhawna Kankane, Sandeep Sharma, Rajesh Mishra

This study presents an analytical framework for assessing Network Coverage Reliability
(NCR) in WMNs deployed in finite circular regions. It incorporates boundary effects and
node failure, examining three sensing models: Boolean, Shadowing & Multipath Fading

14
 (SMF), and Elfes. The analysis reveals that increasing node density and sensing range
enhances NCR, while higher fading values reduce it. Mean Time to Failure (MTTF) is also
derived as a reliability metric [11].

3. A Gaussian Process Regression Approach to Predict the k-Barrier Coverage

Probability for Intrusion Detection in Wireless Sensor Networks
Authors: Abhilash Singh, Jaiprakash Nagar, Sandeep Sharma, Vaibhav Kotiyal

This paper utilizes Gaussian Process Regression (GPR) to estimate k-barrier coverage
probability in WSN-based intrusion detection. Key features include sensor count, sensing
range, Sensor-to-Intruder Velocity Ratio (SIVR), and Mobile to Static Node Ratio
(MSNR). The model achieves R = 0.85 and RMSE = 0.095, outperforming traditional
detection methods [12].

4. A Deep Learning Approach to Predict the Number of 𝑘-Barriers for Intrusion

Detection Over a Circular Region Using Wireless Sensor Networks
Authors: Abhilash Singh, J. Amutha, Jaiprakash Nagar, Sandeep Sharma

This work proposes a fully connected feedforward neural network for predicting k-
barriers in circular deployment regions. The study incorporates Gaussian and uniform
sensor distributions, achieving correlation coefficients of R = 0.78 and 0.79,
respectively [13].

5. A Deep Learning Approach to Accurately Predict the κ-Coverage Probability

in Wireless Sensor Networks
Authors: Mini Arora, Ashok Pal

The authors develop an ANN-based prediction model that incorporates sensor count,
sensing range, shadowing standard deviation, and κ-value. The model outperforms
ANFIS by achieving R = 0.98 and RMSE = 0.07, with reduced time complexity [14].

15
6. Estimating the Coverage Performance of a Wireless Sensor Network
Considering Boundary Effects in the Presence of Sensor Failure
Authors: Mini, Ashok Pal

This analytical model accounts for sensor failures and boundary effects, illustrating
that higher k-values reduce KCP. The model is validated with simulations, achieving
an RMSE of 0.03, thus emphasizing the need to consider environmental and
topological constraints [15].

7. Analysis of Multi-hop Wireless Sensor Networks Using Probability Propagation

Models
Authors: Komgrit Jaksukam, Teerawat Tongloy, Santad Chuwongin, Siridech
Boonsang

A probability propagation model is developed for multi-hop communication

reliability using cluster-tree topology. The study evaluates data transmission
probabilities at cluster heads and sink nodes using the IEEE 802.15.4 MAC protocol,
offering insights into energy-efficient multi-hop communication [16].

8. A Universal Approach to Coverage Probability and Throughput Analysis for

Cellular Networks
Authors: Hui Zhang, Sheng Chen, Liang Feng, Yifeng Xie, Lajos Hanzo

This research proposes a general framework called Equivalent Uniform-Density

Plane-Entity (EUDPE) to analyze coverage and throughput in cellular systems under
different topologies and spatial point processes, such as Poisson. The approach offers
accurate lower-bound estimates for system planning [17].

16
9. SINR-based k-Coverage Probability in Cellular Networks with Arbitrary
Shadowing
Authors: H.P. Keeler, Błaszczyszyn, M.K. Karray

The authors present explicit expressions for Signal-to-Interference-and-Noise Ratio

(SINR) distributions in Poisson-distributed base station layouts, considering arbitrary
shadowing and fading, enabling robust SINR-based k-coverage evaluations [18].

10. Expected k-Coverage in Wireless Sensor Networks

Authors: Li-Hsing Yen, Chang-Wu Yu, Yang-Min Cheng

This paper offers an exact formula for expected k-coverage in finite sensor networks.
The findings assist in energy-efficient node scheduling and deployment planning,
particularly for minimizing over-provisioning [19].

11. Randomized k-Coverage Algorithms for Dense Sensor Networks

Authors: Mohamed Hefeeda, Majid Bagheri

Two approximation algorithms—RKC (centralized) and DRKC (distributed)—are

introduced for activating a minimal number of sensors to maintain k-coverage while
significantly reducing energy consumption [20].

12. A Probabilistic Model of k-Coverage in Minimum Cost Wireless Sensor

Networks
Authors: S.A.R. Zaidi, M. Hafeez, D.C. McLernon, M. Ghogho

This study formulates a cost-efficient probabilistic model that estimates the minimum
number of sensors required for ensuring reliable k-coverage under budget constraints,
optimizing coverage-cost trade-offs [21].

17
Table 2.2.1: Literature Review Table
Year Title Techniques Used Performance Metrics
2024 K-Coverage Reliability Analytical Reliability metrics:
for Wireless Multihop Modeling, NCR, MTTF. Failure
Network Incorporating MATLAB rate: 0.01–0.05/hr.
Boundary Effect Simulations NCR decreases with σ
and Ω (SMF).
Coverage degraded by
4% for 2 dB σ increase.
2023 A Machine Learning Generalized R=0.78, RMSE=0.14
Approach to Predict the Regression Neural (with BE),
k-Coverage Probability Network (GRNN) RMSE=0.15 (without
of Wireless Multihop BE).
Networks Considering
Boundary and
Shadowing Effects
2022 A Deep Learning Feed-Forward R=0.98, RMSE=0.07.
Approach to Accurately ANN, ANFIS
Predict the κ-Coverage
Probability in Wireless
Sensor Networks
2021 A Distributed Border Log-normal Performance
Surveillance (DBS) Shadowing improvement: 75%
System for Rectangular Model, NS-2.35 over binary sensing
and Circular Region of Simulator model
Interest with Wireless
Sensor Networks in

18
 Shadowed
Environments
2021 ECS-NL: An Enhanced Enhanced Cuckoo Average Localization
Cuckoo Search Search (ECS) with Error (ALE): 0.5–0.8
Algorithm for Node Early Stopping m; Time Reduction:
Localization in (ES) mechanism 80%
Wireless Sensor
Networks
2020 WSN Strategies Based Review of State-of-the-art
on Sensors, strategies and classification and
Deployment, Sensing classification tables summarizing
Models, Coverage and framework open issues
Energy Efficiency:
Review, Approaches
and Open Issues
2019 Network Coverage in Analytical Decoding probability,
Interference Limited modeling, critical density
Wireless Sensor simulations
Networks
2018 Kalman Filtering GRNN, Kalman RMSE, Average
Framework based Real Filter (KF), Localization Error
Time Target Tracking in Unscented KF
Wireless Sensor
Networks using
Generalized Regression
Neural Networks
2017 Coverage Contribution Coverage Network Lifetime,
Area Based k-Coverage Contribution Area Sensor Spatial Density

19
 for Wireless Sensor (CCA),
Networks Centralized and
Distributed
Protocols
2014 Border Effect Analysis Border Connectivity
for Reliability Dominance Probability, MTTD,
Assurance and Solution, Border MNSR
Continuous Effect Avoidance,
Connectivity of Connectivity
Wireless Sensor Analysis, Sensor
Networks in the Redeployment
Presence of Sensor
Failures
2013 Distance Distributions CDF & PDF Distance Distribution,
in Regular Polygons derivation, Euclidean Distance
Rotational
Symmetry,
Algorithm Design
2002 Statistical Mechanics of Empirical Topology, Robustness
Complex Networks Analysis, Random
Graphs, Network
Models

2.3 Summary
This literature review explored a wide range of research efforts addressing k-coverage
prediction, network reliability, and efficient deployment strategies in WSNs and WMNs.

20
Analytical models, regression-based approaches, and deep learning models were compared,
with a particular focus on their performance in handling boundary effects, shadowing, and
sensor failures.

The review establishes that machine learning—especially models like GRNN, GPR, and
ANNs—demonstrates significant promise in achieving accurate and scalable KCP
predictions. These insights directly shape the research direction in the subsequent chapters,
focusing on model implementation and evaluation.

21
 CHAPTER – 3
METHODS AND
METHODOLOGIES

22
3.1 Introduction
Machine learning (ML) has emerged as a transformative approach for addressing complex,
data-driven problems, particularly in wireless multihop networks (WMNs). The ability of
ML models to learn from data and adapt to changing environments makes them ideally suited
for predicting k-coverage probability (KCP)—a key performance metric in WMNs [22].

Conventional statistical and analytical methods, though valuable, often struggle to adapt to
the dynamic, high-dimensional nature of real-world wireless environments. In contrast, ML
approaches can leverage historical and simulated datasets to uncover intricate patterns in
sensor placement, environmental interference, and node density. This chapter outlines the
ML paradigms, algorithms, and tools utilized in this study. It begins with a detailed
explanation of ML types, followed by an overview of the implemented models and a
description of the software stack used to build, train, and evaluate the models.

3.2 Types of Machine Learning

Machine learning is broadly categorized into three types: supervised learning, unsupervised
learning, and reinforcement learning, each with its own algorithms, goals, and applications
[23].

3.2.1 Supervised Learning

Supervised learning involves mapping inputs to known outputs using labeled datasets. It is
especially effective for regression and classification tasks. The models learn from the data
by minimizing prediction error, and they generalize well to unseen samples. In this study,
supervised learning is used for regression-based KCP prediction.

Popular supervised learning algorithms include:

• Support Vector Regressor (SVR): Effective in high-dimensional spaces.

• Random Forest (RF): Robust ensemble method that handles large datasets well.

23
• Gradient Boosting (GB): Known for high prediction accuracy in tabular datasets.

Supervised learning is widely applied in healthcare diagnostics, financial modeling, and

network performance analysis [24].

Table [Link]: Comparison of Supervised Learning Algorithms

Algorithm Strengths Limitations

Random Forest (RF) Handles large datasets Prone to overfitting with
well. noisy data.
Gradient Boosting High predictive accuracy. Requires careful tuning.
(GB)
Support Vector Effective in high- Computationally
Regressor (SVR) dimensional spaces intensive

3.2.2 Unsupervised Learning

Unsupervised learning algorithms operate on unlabeled datasets, aiming to identify hidden

patterns and structures. Unlike supervised methods, they do not rely on predefined outputs,
making them suitable for exploratory data analysis, clustering, and dimensionality reduction.

Key algorithms include:

• K-Means Clustering

• Hierarchical Clustering

• Principal Component Analysis (PCA)

Unsupervised learning is valuable for tasks such as anomaly detection, customer

segmentation, and sensor fault analysis [25].

24
 Figure [Link]: Supervised vs Unsupervised Learning

3.2.3 Reinforcement Learning

Reinforcement Learning (RL) is a reward-driven learning framework where agents learn to

take actions that maximize cumulative rewards over time. Unlike supervised learning, RL
does not require labeled datasets—it learns through interaction with the environment.

In wireless communication, RL is used for:

• Dynamic network configuration

• Resource allocation

• Coverage optimization

RL models such as Q-learning and Deep Q-Networks (DQNs) have been effectively applied
to autonomous systems and adaptive wireless networks [26].

3.3 Models
This study evaluates and compares four widely used machine learning models for predicting
k-coverage probability:

25
1. Support Vector Machine (SVM)

2. General Regression Neural Network (GRNN)

3. Random Forest (RF)

4. Gradient Boosting (GB)

Each model was chosen based on its ability to handle non-linear data, multivariate
features, and regression tasks.

3.3.1 Support Vector Machine (SVM)

SVM is a supervised learning model that seeks the optimal hyperplane for classification or
regression tasks. For regression, SVM is adapted into Support Vector Regressor (SVR),
which fits a function that deviates from actual targets by a value not more than ε and tries to
minimize the margin.

Key benefits:

• Robust to outliers.

• Effective in high-dimensional spaces.

• Suitable for non-linear regression using kernel tricks.

Applications include biomedical signal processing, text classification, and network

performance estimation [27].

3.3.2 General Regression Neural Network (GRNN)

GRNN is a feedforward neural network that uses a radial basis function (RBF)
architecture. It is primarily designed for regression and function approximation tasks.

26
Key features:

• Rapid training and convergence.

• Excellent performance on small datasets.

• No iterative training required (non-backpropagation-based).

GRNN is effective in domains such as real-time prediction, financial modeling, and signal
interpolation [28].

3.3.3 Random Forest

Random Forest is an ensemble learning method that constructs multiple decision trees and
averages their outputs for regression.

Advantages:

• Handles large datasets and numerous features well.

• Provides feature importance scores, enabling explainability.

• Resistant to overfitting and noise.

It is frequently used for classification, forecasting, and network modeling [29].

3.3.4 Gradient Boosting

Gradient Boosting is another ensemble method where models are built sequentially. Each
new model corrects the errors of its predecessor by focusing on residuals.

Core benefits:

• Highly accurate for structured/tabular data.

• Customizable via learning rate, max depth, and subsampling.

• Often achieves better results than Random Forest when properly tuned.

Common implementations include XGBoost, LightGBM, and CatBoost [30].

27
 Figure [Link]: Working of Gradient Boosting

3.4 Tools and Programming Languages

A robust stack of tools and libraries was used to implement the above models efficiently. All
components were selected for their interoperability, performance, and community support.

3.4.1 Jupyter Notebook

Jupyter Notebook is an interactive development environment supporting literate

programming—merging code, visualizations, and markdown into a unified workflow. It is
widely used in education, prototyping, and research due to its versatility [31].

3.4.2 Python 3

Python 3 is the primary programming language for this study. With its concise syntax and
massive library ecosystem, it simplifies data preprocessing, model training, evaluation, and
visualization [32].

3.4.3 NumPy

28
NumPy enables fast numerical computations on large multi-dimensional arrays. Functions
for linear algebra, statistical operations, and matrix transformations make it essential in
machine learning pipelines [33].

3.4.4 Pandas

Pandas provides high-level data structures (Series and DataFrames) for handling tabular
data. It streamlines:

• Data cleaning

• Missing value imputation

• Feature extraction

It is a core part of any data science or ML project [34].

3.4.5 Matplotlib

Matplotlib is a comprehensive visualization library that enables the creation of static,

animated, and interactive plots. It is extensively used for generating graphs and charts to
analyze model performance and interpret dataset characteristics.

3.5 Summary
This chapter detailed the learning paradigms, algorithms, and tools utilized in this
research. Machine learning’s potential to model complex, high-dimensional relationships
makes it well-suited for predicting k-coverage in wireless multihop networks. The selected
models—SVR, GRNN, Random Forest, and Gradient Boosting—cover a spectrum of
learning strategies, from kernel-based regression to ensemble learning and neural networks.

The use of Python, Jupyter, and essential libraries like NumPy and Pandas ensures
replicability and efficiency. These foundations pave the way for the next chapter, which will
present the experimental setup, dataset details, and model evaluation.

29
 CHAPTER – 4
K-COVERAGE
PROBABILITY
PREDICTION MODEL

30
4.1 Introduction
The prediction of k-coverage probability plays a pivotal role in optimizing the architecture
and performance of Wireless Multi-hop Networks (WMNs). K-coverage refers to the
scenario where each point in the target area is covered by at least 'k' sensor nodes, ensuring
fault tolerance, data redundancy, and improved communication reliability. Accurate
estimation of this probability enables network designers to deploy sensor nodes more
efficiently, minimize energy usage, and enhance overall robustness [36]. In WMNs, the
network’s performance is highly sensitive to environmental variables, node density, and
spatial deployment, which makes predictive modeling both complex and crucial [37].

This chapter elaborates on the systematic methodology adopted to predict k-coverage

probability using machine learning (ML) models. The process encompasses data acquisition,
preprocessing, feature engineering, model selection, training, and performance evaluation.
Machine learning techniques are ideal for this task due to their adaptability to nonlinear data
patterns and capability to generalize from training data to unseen conditions [38].

The approach discussed in this chapter ensures the ML models can address real-world
uncertainties, such as boundary effects and signal shadowing, which are typically
encountered in WMNs. Additionally, data preprocessing ensures that the dataset used is free
from inconsistencies and redundancies. Feature engineering and careful model selection
contribute significantly to the predictive accuracy and reliability of the proposed solutions.

4.2 Dataset
The dataset utilized in this research comprises vital network parameters including
geographical dimensions (area, length, and breadth), number of deployed nodes, sensing
range, environmental shadowing effects (Sigma), and the target variable—k-coverage
probability. These parameters collectively influence how effectively a wireless sensor
network achieves the desired level of coverage. Such datasets are essential for modeling the
behavior of wireless networks under varied deployment scenarios [39].

31
Table 4.2.1: Dataset Overview

Feature Name Description

Area (m²) Total coverage area of the network
Length (m) Length of the deployment area
Breadth (m) Breadth of the deployment area
Nodes Number of sensor nodes deployed
Sensing Range (m) Range of each sensor
Sigma (dB) Standard deviation of shadowing effects
Minimum number of sensors required
Required k
for coverage
Target variable indicating coverage
K-Coverage Probability
probability

Prior to training, the dataset underwent comprehensive preprocessing. Missing values were
imputed using the mean strategy to ensure the dataset remained intact without loss of critical
information. Skewed features were transformed using logarithmic functions to normalize
their distributions, a vital step for enhancing the learning capability of many ML models
[40]. Furthermore, standardization techniques such as Z-score normalization were applied,
especially beneficial for distance-based algorithms like Support Vector Regressor (SVR) that
are sensitive to scale.

Feature independence and multicollinearity were also evaluated to retain only those
parameters contributing uniquely to the model. An initial exploratory data analysis (EDA)
was conducted to visualize distributions, relationships, and dependencies among the features.
Tools such as correlation heatmaps and pair plots helped identify redundant or irrelevant
features, thereby improving the efficiency and interpretability of subsequent models.

32
4.3 Process Flow
The process flow of the k-coverage probability prediction model consists of the following
key steps:

1. Load Dataset - Import and inspect the dataset for missing values and anomalies. Data
loading ensures that the dataset is correctly structured before further processing.

2. Data Cleaning - Handle missing values and remove redundant columns. Cleaning
ensures that inconsistencies in the dataset do not affect model training.

3. Feature Engineering - Transform skewed features and standardize variables to improve

predictive performance.

4. Exploratory Data Analysis (EDA) - Generate histograms, correlation heatmaps, and

pair plots to understand feature relationships. Identifying patterns and dependencies
allows for better model training.

5. Feature/Target Definition - Define independent features and target variables to

streamline model training.

6. Data Splitting - Divide data into training and testing sets (80:20 ratio) to evaluate model
generalizability.

7. Scaling - Apply feature scaling for models that require normalized inputs (e.g., SVM),
ensuring fair model comparisons.

8. Model Training - Train different machine learning models on the dataset by fitting them
to the training data.

This methodological approach helps to reduce model bias and variance, thereby improving
both prediction accuracy and robustness in deployment scenarios. A structured process flow
also ensures model reproducibility, which is critical in scientific studies [41].

33
4.4 Model Training
The models selected for this study include Random Forest, Gradient Boosting, Support
Vector Regressor (SVR), and General Regression Neural Network (GRNN). These
algorithms are known for their efficacy in handling structured datasets and delivering robust
regression performance in varied real-world contexts [42].

4.4.1 Random Forest

Random Forest is a powerful ensemble learning method that operates by constructing

multiple decision trees during training and outputting the average prediction. It excels in
modeling complex, non-linear relationships without requiring significant parameter tuning
[43]. An important feature of Random Forest is its ability to rank feature importance, thus
aiding in identifying critical variables that influence k-coverage probability.

4.4.2 Gradient Boosting

Gradient Boosting is a boosting technique that iteratively builds an ensemble of weak

learners, typically decision trees, and combines them to minimize prediction error. It is
known for achieving state-of-the-art results in many regression and classification problems
[44]. While computationally intensive, its ability to reduce both bias and variance makes it
particularly effective in coverage prediction tasks.

4.4.3 Support Vector Regressor (SVR)

SVR applies the principles of support vector machines to regression problems. It is

particularly well-suited for problems involving high-dimensional feature spaces and non-
linear relationships. However, SVR’s performance is sensitive to kernel selection and
hyperparameter tuning, which requires careful cross-validation [45]. Its utility in modeling
complex sensor network environments makes it a strong candidate in this research.

34
 4.4.4 General Regression Neural Network (GRNN)

GRNN is a type of probabilistic neural network that performs well with noisy or limited
datasets. Unlike deep neural networks, GRNN does not require long training times and
converges quickly. It utilizes a radial basis function kernel to estimate continuous functions,
making it suitable for real-time applications where fast adaptation is essential [46].

All models were trained on two variations of the dataset:

• With Boundary Effects – Includes constraints imposed by the physical boundaries of the
deployment area, which significantly influence sensor placement and coverage [47].

• Without Boundary Effects – Ignores edge constraints, thereby simplifying the environment
and offering a baseline for comparative analysis.

These dual scenarios allow a holistic evaluation of model behavior under practical and
idealized conditions.

4.5 Summary
This chapter has detailed the methodology employed to predict k-coverage probability in
WMNs using machine learning. The process began with a detailed dataset comprising
environment-specific and network-related parameters, followed by data cleaning,
transformation, and exploration. Feature selection and scaling were carried out meticulously
to improve model performance.

Four machine learning models—Random Forest, Gradient Boosting, SVR, and GRNN—
were employed, each offering unique advantages in handling non-linear, high-dimensional
data. By using both boundary-constrained and unconstrained datasets, the study ensures a
thorough evaluation of each model’s robustness. The forthcoming chapter will delve into the
comparative analysis and results derived from these models.

35
CHAPTER – 5
RESULTS AND
ANALYSIS

36
5.1 Introduction
This chapter presents the experimental findings and comparative performance evaluation of
the proposed machine learning models for predicting k-coverage probability in wireless
multi-hop networks (WMNs). It evaluates and contrasts the behavior of Random Forest,
Gradient Boosting, Support Vector Regressor (SVR), and General Regression Neural
Network (GRNN) models using multiple evaluation metrics to assess accuracy,
generalization ability, and robustness. By analyzing results under both ideal and realistic
conditions (i.e., with and without boundary effects), this chapter provides a comprehensive
insight into each model’s suitability for practical network deployments.

The key regression metrics used include Mean Squared Error (MSE), Root Mean Squared
Error (RMSE), and R² score. These measures help quantify model accuracy, explainability,
and error characteristics. Additionally, results are visualized using comparative heatmaps,
bar charts, and feature importance plots to enhance interpretability. This evaluation serves as
a basis for selecting the most appropriate predictive model for optimizing sensor node
deployment in WMNs under varied environmental constraints [48], [49].

5.2 K-Coverage Probability Prediction Model

The k-coverage probability prediction model is designed to estimate the probability that
every point in a wireless multi-hop network is covered by at least k sensor nodes. It relies on
supervised learning techniques, trained on structured datasets consisting of features such as
deployment area, node density, sensing radius, shadowing factor (σ), and required k. These
features encapsulate both spatial and environmental aspects of the network, making them
crucial to understanding coverage behavior [50].

Two experimental setups are considered to simulate real-world and ideal scenarios:

37
1. With Boundary Effects: This setup accounts for environmental constraints such as physical
obstructions, terrain limitations, and edge-based node sparsity, which affect sensing
performance and coverage density [51].

2. Without Boundary Effects: This baseline assumes an ideal environment, disregarding real-
world constraints to assess theoretical performance.

Each model is evaluated using the following metrics:

• Mean Squared Error (MSE): Measures the average of squared prediction errors. Lower
values signify better accuracy.

• Root Mean Squared Error (RMSE): Interpretable as the average prediction error in the
original units.

• R² Score: Represents the proportion of variance in the dependent variable that is predictable
from the independent variables.

Table 5.2.1: Model Performance Comparison

MSE (With R² (With MSE (No R² (No

Model
BEs) BEs) BEs) BEs)
Random
0.0404 0.4523 0.0442 0.4382
Forest
Gradient
0.0455 0.3828 0.0592 0.2475
Boosting
Support
Vector 0.0166 0.7752 0.0188 0.7610
Regressor

As seen in Table 5.2.1, the SVR model achieves the lowest MSE and the highest R² values
in both scenarios, indicating that it captures the relationship between network parameters

38
and k-coverage probability more effectively than the others. Gradient Boosting, while robust,
exhibits slightly higher error margins, potentially due to sensitivity to hyperparameter tuning
[52]. Random Forest delivers reliable results, especially in scenarios with noise or boundary
variation.

Figure 5.2.1: Heatmap of Model Performance Summary

39
This heatmap visually emphasizes the variation in MSE and R² values across models and
scenarios. Brighter regions correspond to stronger model accuracy and fit. It highlights
SVR’s superior consistency and the decline in performance when boundary effects are
introduced.

5.3 Comparison with the Existing Models

To contextualize the effectiveness of the proposed machine learning models, they are
compared against GRNN—a widely used baseline in previous k-coverage probability
studies. Traditional techniques, such as heuristic optimization and statistical regression, often
underperform when modeling complex, non-linear interactions typical in sensor networks.
ML algorithms, by contrast, are data-driven and capable of adapting to high-dimensional
input spaces [53].

The comparative evaluation is based on three key aspects:

• Accuracy Improvements: ML models outperform GRNN in MSE and R² metrics,

confirming their superiority in learning non-linear patterns.

• Computational Efficiency: While GRNN is computationally lightweight, tree-based

models offer a better trade-off between runtime and accuracy, making them ideal for near-
real-time systems.

• Robustness to Variability: Models like SVR and Random Forest demonstrate greater
adaptability across boundary-aware and ideal conditions.

Table 5.3.1: Performance Comparison of Machine Learning Models vs. GRNN

MSE (With R² (With MSE (No R² (No

Model
BEs) BEs) BEs) BEs)
SVM 0.0166 0.7752 0.0188 0.7610
GRNN 0.0225 0.6084 0.0196 0.6084

40
The data shows that while GRNN performs acceptably in the no-boundary condition, it fails
to generalize well when boundary effects are introduced. In contrast, SVR maintains high
accuracy across both settings, reinforcing its robustness [54].

Figure 5.3.1: Feature Importance Analysis

This bar chart ranks input features based on their contribution to model prediction accuracy.
Key predictors include sensing range, node count, and required k, consistent with findings
from prior sensor network optimization studies [55]. The analysis assists network planners
in prioritizing deployment parameters that most affect network reliability.

5.4 Results
This section consolidates the individual performance outcomes of the trained models,
compares predicted vs. actual values, and visualizes error distributions.

41
 5.4.1 Model Performance Analysis

• Random Forest: Offers good generalization and low variance, ideal for large datasets with
mixed data types. Slightly underperforms compared to SVR but provides interpretability and
resilience to overfitting [56].

• Gradient Boosting: Delivers accurate predictions through iterative refinement but is more
sensitive to noise and requires careful tuning.

• Support Vector Regressor (SVR): Excels in modeling non-linear relationships and yields the
highest prediction accuracy. However, it is computationally intensive, especially with large
datasets [57].

• General Regression Neural Network (GRNN): Provides fast inference with minimal training
but lower accuracy and poor generalization in complex settings.

Figure [Link]: Model Performance Comparison using Stacked Bar Graph

The bar chart illustrates the MSE and R² values side by side for all models, offering an
intuitive comparison of prediction quality. SVR leads in both metrics, with GRNN trailing
in accuracy.

42
5.4.3 Visualizing Model Predictions

In addition to heatmaps and bar charts, residual plots and prediction scatter plots were used
to analyze error distribution. Closer alignment of predicted and actual values across samples
indicated stronger model calibration. SVR’s residuals were the most tightly clustered,
indicating minimal bias.

This visualization validates that SVR and Random Forest not only achieve low average error
but also maintain consistency across different test samples, a critical property for deployment
in unpredictable environments [58].

5.5 Summary
This chapter provided a detailed performance evaluation of four machine learning models
applied to the problem of predicting k-coverage probability in WMNs. The SVR model
emerged as the most accurate across both ideal and realistic deployment scenarios. While
GRNN was faster, it underperformed in generalization and accuracy. Random Forest and
Gradient Boosting provided competitive performance with good scalability and
interpretability.

Feature importance analysis emphasized the role of sensing range, node density, and required
coverage level in influencing k-coverage outcomes. These findings support the use of data-
driven models for intelligent network design. By leveraging advanced ML techniques,
practitioners can make informed deployment decisions that optimize resource use and ensure
robust, scalable network coverage.

The next chapter will conclude this thesis and outline potential avenues for future work.

43
 CHAPTER – 6
CONCLUSION AND
FUTURE SCOPES

44
6.1 Introduction
This chapter concludes the research conducted on k-coverage probability prediction in
wireless multi-hop networks (WMNs) using machine learning (ML) models. The study
explored and implemented four predictive models—Random Forest, Gradient Boosting,
Support Vector Regressor (SVR), and General Regression Neural Network (GRNN)—to
estimate k-coverage probability under varying network configurations and environmental
conditions.

The research emphasizes the importance of data-driven techniques in enhancing the

performance of wireless sensor networks (WSNs). These models not only improved
prediction accuracy but also allowed adaptive deployment strategies that can reduce
redundant node placement, minimize energy consumption, and increase system reliability
[59].

The comparative analysis demonstrated that SVR exhibited the highest predictive accuracy,
particularly under complex deployment scenarios. Random Forest and Gradient Boosting
models offered a practical balance between computational efficiency and accuracy, while
GRNN, although fast, was outperformed in generalization ability.

Figure 6.1.1: Model Performance Comparison using Grouped Bar Graph

45
As illustrated in Figure 6.1.1, the grouped bar graph provides a comparative view of MSE
and R² scores across all trained models. The visualization confirms that SVR had the most
consistent and accurate performance. The heatmap and residual plots discussed in Chapter 5
further validated that the machine learning models are capable of capturing non-linear
patterns and environmental variabilities that influence coverage [60].

Additionally, key influential parameters such as sensing range, node count, and
environmental noise (σ) were identified as critical drivers in determining network coverage
levels. Feature importance analysis further supported these findings and provides a valuable
reference for future deployment optimization strategies [61].

6.2 Future Scope

While this study offers substantial contributions to the prediction of k-coverage in WMNs,
several avenues remain open for enhancement and further exploration. Future work can
address both methodological improvements and real-world applicability, including:

1. Deep Learning Integration: Implementing deep learning models such as Convolutional

Neural Networks (CNNs) for spatial data and Long Short-Term Memory (LSTM)
networks for temporal data could improve the prediction of dynamic network behaviors
[62].

2. Real-World Deployment and Validation: Field experiments in actual WSN

deployments would help validate the generalizability of trained models under
uncontrolled, real-world conditions. This includes consideration of terrain, hardware
faults, and mobility.

3. Adaptive Network Optimization: Incorporating reinforcement learning or adaptive

control strategies that enable real-time adjustments to network configurations based on
environmental feedback.

46
4. Hybrid Machine Learning Models: Using ensemble or stacking techniques that
combine predictions from multiple models can further reduce error and increase
robustness to unseen scenarios [63].

5. Scalability and Efficiency Improvements: Optimizing algorithms for parallel

processing or lightweight deployment in embedded systems would allow real-time
predictions even in resource-constrained environments such as edge computing devices.

6. Incorporating Additional Parameters: Introducing additional environmental or

contextual features—such as wind speed, temperature, or infrastructure interference—
could improve model granularity and context-aware performance.

Table 6.2.1: Future Research Directions and Benefits

Future Research Direction Future Research Direction

Deep Learning Integration Deep Learning Integration
Ensuring model accuracy in
Real-World Deployment and Validation
practical scenarios
Adaptive Network Optimization Adaptive Network Optimization
Hybrid Machine Learning Models Enhancing accuracy and robustness
Scalability and Efficiency Improvements Scalability and Efficiency
Improvements
Incorporating Additional Parameters Incorporating Additional Parameters

These directions represent promising paths toward building intelligent, autonomous, and
energy-efficient sensor networks that adapt seamlessly to dynamic conditions and user
requirements [64].

6.3 Application of this Study

47
The The research findings on k-coverage probability prediction are applicable across a broad
spectrum of domains that rely on wireless sensor networks for real-time data acquisition,
monitoring, and control. The integration of machine learning in WSNs enhances network
intelligence, allowing smarter deployment, monitoring, and decision-making. Some key
application areas include:

1. Smart Cities: Optimizing the deployment of sensors for traffic control, environmental
monitoring, waste management, and public safety systems [65].

2. Disaster Management: Real-time deployment of sensor networks in disaster-prone

zones enables early warning systems, damage assessment, and search and rescue
coordination.

3. Agriculture and Precision Farming: Monitoring soil moisture, temperature, and crop
health through distributed sensors helps farmers optimize irrigation and increase yield
efficiency.

4. Healthcare and Remote Monitoring: Wireless medical sensors can support continuous
health monitoring for patients, especially in remote or resource-scarce regions [66].

5. Military and Defense: Tactical deployment of sensor nodes enhances surveillance,

perimeter monitoring, and real-time threat detection in hostile environments.

6. Industrial IoT (IIoT): WSNs improve predictive maintenance, environmental

compliance, and asset tracking in manufacturing and heavy industries [67].

The growing fusion of AI and IoT technologies allows these applications to benefit from
adaptive intelligence, where predictive models like the one proposed can be deployed in real
time for better autonomy, efficiency, and cost-effectiveness.

48
REFERENCES

49
[1] M. A. Alsheikh, S. Lin, D. Niyato, and H.-P. Tan, “Machine learning in wireless sensor
networks: Algorithms, strategies, and applications,” IEEE Communications Surveys &
Tutorials, vol. 16, no. 4, pp. 1996–2018, 2014.
[2] C. Li, Z. Sun, H. Wang, and H. Song, “A novel energy-efficient k-coverage algorithm
based on probability-driven mechanism of wireless sensor networks,” Int. J. Distrib.
Sens. Netw., vol. 12, no. 9, pp. 1–11, 2016.
[3] J. Yu, S. Ren, S. Wan, D. Yu, and G. Wang, “A stochastic k-coverage scheduling algorithm
in wireless sensor networks,” Int. J. Distrib. Sens. Netw., vol. 2012, Article ID 615989,
2012.
[4] K. Tarnaris, I. Preka, D. Kandris, and A. Alexandridis, “Coverage and k-coverage
optimization in wireless sensor networks using computational intelligence methods: A
comparative study,” Electronics, vol. 9, no. 4, pp. 1–18, Apr. 2020.
[5] M. Chen, U. Challita, W. Saad, C. Yin, and M. Debbah, “Artificial neural networks-based
machine learning for wireless networks: A tutorial,” IEEE Communications Surveys &
Tutorials, vol. 21, no. 4, pp. 3039–3071, 2019.
[6] J. Amutha, S. Sharma, and J. Nagar, “WSN strategies based on sensors, deployment,
sensing models, coverage and energy efficiency: Review, approaches and open issues,”
Wireless Pers. Commun., vol. 111, no. 2, pp. 1089–1115, 2020.
[7] M. Kulin, T. Kazaz, I. Moerman, and E. De Poorter, “A survey on machine learning-
based performance improvement of wireless networks: PHY, MAC and network layer,”
IEEE Communications Surveys & Tutorials, vol. 22, no. 3, pp. 1646–1681, 2020.
[8] J. Nagar, S. K. Chaturvedi, and S. Soh, “Connectivity analysis of finite wireless multihop
networks incorporating boundary effects in shadowing environments,” IET Commun.,
vol. 14, no. 11, pp. 1801–1810, 2020.
[9] S. Mohammadjafari, S. Roginsky, E. Kavurmacioglu, M. Cevik, J. Ethier, and A. B.
Bener, “Machine learning-based radio coverage prediction in urban environments,” IEEE
Trans. Netw. Serv. Manag., vol. 17, no. 4, pp. 2117–2130, 2020.

50
[10] J. Nagar, S. K. Chaturvedi, S. Soh, and A. Singh, “A machine learning approach to predict
the k-coverage probability of wireless multihop networks considering boundary and
shadowing effects,” in Proc. IC3ECSBHI, 2023.
[11] B. Kankane, S. Sharma, and R. Mishra, “κ-coverage reliability for wireless multihop
network incorporating boundary effect,” in Proc. ICCCS, 2024.
[12] A. Singh, J. Nagar, S. Sharma, and V. Kotiyal, “A Gaussian process regression approach
to predict the k-barrier coverage probability,” Comput. Commun., vol. 187, pp. 49–59,
2023.
[13] A. Singh, J. Amutha, J. Nagar, and S. Sharma, “A deep learning approach to predict the
number of k-barriers for intrusion detection,” J. Ambient Intell. Humaniz. Comput., vol.
13, no. 4, pp. 1791–1805, 2022.
[14] M. Arora and A. Pal, “A deep learning approach to accurately predict the κ-coverage
probability,” Sensors, vol. 22, no. 2, pp. 501–512, 2022.
[15] Mini and A. Pal, “Estimating the coverage performance of a WSN considering boundary
effects and sensor failure,” Ad Hoc Netw., vol. 107, pp. 102267, 2021.
[16] K. Jaksukam, T. Tongloy, S. Chuwongin, and S. Boonsang, “Analysis of multi-hop
WSNs using probability propagation models,” in Proc. IEEE WCNC, 2020.
[17] H. Zhang, S. Chen, L. Feng, Y. Xie, and L. Hanzo, “A universal approach to coverage
probability and throughput analysis for cellular networks,” IEEE Trans. Veh. Technol.,
vol. 69, no. 1, pp. 681–695, Jan. 2020.
[18] H. P. Keeler, B. Błaszczyszyn, and M. K. Karray, “SINR-based k-coverage probability in
cellular networks,” IEEE Trans. Commun., vol. 67, no. 1, pp. 550–565, 2019.
[19] L.-H. Yen, C.-W. Yu, and Y.-M. Cheng, “Expected k-coverage in WSNs,” IEEE Commun.
Lett., vol. 22, no. 10, pp. 2080–2083, Oct. 2018.
[20] M. Hefeeda and M. Bagheri, “Randomized k-coverage algorithms for dense sensor
networks,” IEEE Trans. Mob. Comput., vol. 10, no. 11, pp. 1540–1552, Nov. 2011.

51
[21] S. A. R. Zaidi, M. Hafeez, D. C. McLernon, and M. Ghogho, “A probabilistic model of
k-coverage in minimum cost WSNs,” IEEE Trans. Commun., vol. 62, no. 8, pp. 2937–
2949, Aug. 2014.
[22] M. Z. Win, Y. Shen, and W. Dai, “A framework for signal coverage analysis in wireless
networks,” IEEE Trans. Signal Process., vol. 67, no. 3, pp. 619–634, Feb. 2019.
[23] S. Raschka and V. Mirjalili, Python Machine Learning, 3rd ed. Packt Publishing, 2019.
[24] C. M. Bishop, Pattern Recognition and Machine Learning. Springer, 2006.
[25] J. Han, M. Kamber, and J. Pei, Data Mining: Concepts and Techniques, 3rd ed. Morgan
Kaufmann, 2011.
[26] R. S. Sutton and A. G. Barto, Reinforcement Learning: An Introduction, 2nd ed. MIT
Press, 2018.
[27] N. Cristianini and J. Shawe-Taylor, An Introduction to Support Vector Machines and
Other Kernel-based Learning Methods. Cambridge Univ. Press, 2000.
[28] D. F. Specht, “A general regression neural network,” IEEE Trans. Neural Netw., vol. 2,
no. 6, pp. 568–576, Nov. 1991.
[29] L. Breiman, “Random forests,” Mach. Learn., vol. 45, no. 1, pp. 5–32, 2001.
[30] T. Chen and C. Guestrin, “XGBoost: A scalable tree boosting system,” in Proc. KDD,
2016, pp. 785–794.
[31] T. Kluyver et al., “Jupyter Notebooks – a publishing format for reproducible
computational workflows,” in Positioning and Power in Academic Publishing: Players,
Agents and Agendas, IOS Press, 2016.
[32] G. van Rossum and F. L. Drake, The Python Language Reference Manual, Network
Theory Ltd., 2011.
[33] S. Van Der Walt, S. C. Colbert, and G. Varoquaux, “The NumPy array: A structure for
efficient numerical computation,” Comput. Sci. Eng., vol. 13, no. 2, pp. 22–30, 2011.
[34] W. McKinney, “Data structures for statistical computing in Python,” in Proc. 9th Python
Sci. Conf., 2010, pp. 51–56.

52
[35] J. D. Hunter, “Matplotlib: A 2D graphics environment,” Comput. Sci. Eng., vol. 9, no. 3,
pp. 90–95, 2007.
[36] A. A. Abbasi and M. Younis, “A survey on clustering algorithms for wireless sensor
networks,” Computer Communications, vol. 30, no. 14–15, pp. 2826–2841, 2007.
[37] J. Yick, B. Mukherjee, and D. Ghosal, “Wireless sensor network survey,” Computer
Networks, vol. 52, no. 12, pp. 2292–2330, 2008.
[38] X. Liu, “A survey on clustering routing protocols in wireless sensor networks,” Sensors,
vol. 12, no. 8, pp. 11113–11153, 2012.
[39] M. Cardei and J. Wu, “Energy-efficient coverage problems in wireless ad-hoc sensor
networks,” Computer Communications, vol. 29, no. 4, pp. 413–420, 2006.
[40] I. Guyon and A. Elisseeff, “An introduction to variable and feature selection,” Journal of
Machine Learning Research, vol. 3, pp. 1157–1182, 2003.
[41] G. E. Hinton, “Learning multiple layers of representation,” Trends in Cognitive Sciences,
vol. 11, no. 10, pp. 428–434, 2007.
[42] L. Breiman, “Random forests,” Machine Learning, vol. 45, no. 1, pp. 5–32, 2001.
[43] J. H. Friedman, “Greedy function approximation: A gradient boosting machine,” Annals
of Statistics, vol. 29, no. 5, pp. 1189–1232, 2001.
[44] C. Cortes and V. Vapnik, “Support-vector networks,” Machine Learning, vol. 20, no. 3,
pp. 273–297, 1995.
[45] D. F. Specht, “A general regression neural network,” IEEE Transactions on Neural
Networks, vol. 2, no. 6, pp. 568–576, 1991.
[46] T. Clouqueur, V. Phipatanasuphorn, P. Ramanathan, and K. K. Saluja, “Sensor
deployment strategy for target detection,” Proceedings of the First ACM International
Workshop on Wireless Sensor Networks and Applications, pp. 42–48, 2002.
[47] J. A. Stankovic et al., “Wireless sensor networks for in-home healthcare: Potential and
challenges,” Proceedings of the High Confidence Medical Device Software and Systems
Workshop, pp. 2–3, 2005.

53
[48] C. Intanagonwiwat, R. Govindan, and D. Estrin, “Directed diffusion: A scalable and
robust communication paradigm for sensor networks,” in Proceedings of the 6th Annual
International Conference on Mobile Computing and Networking, pp. 56–67, 2000.
[49] R. Min, M. Bhardwaj, S. Cho, E. Shih, A. Sinha, A. Wang, and A. Chandrakasan, “Low-
power wireless sensor networks,” in Proceedings of the 14th International Conference on
VLSI Design, pp. 205–210, 2001.
[50] T. S. Rappaport, Wireless Communications: Principles and Practice, 2nd ed., Prentice
Hall, 2002.
[51] W. Wang, V. Srinivasan, and K. Chua, “Coverage in hybrid mobile sensor networks,”
IEEE Transactions on Mobile Computing, vol. 7, no. 5, pp. 640–653, 2008.
[52] J. H. Friedman, “Stochastic gradient boosting,” Computational Statistics & Data
Analysis, vol. 38, no. 4, pp. 367–378, 2002.
[53] S. Haykin, Neural Networks: A Comprehensive Foundation, 2nd ed., Prentice Hall,
1998.
[54] S. Smola and B. Schölkopf, “A tutorial on support vector regression,” Statistics and
Computing, vol. 14, no. 3, pp. 199–222, 2004.
[55] H. Zhang and J. C. Hou, “Maintaining sensing coverage and connectivity in large sensor
networks,” Ad Hoc & Sensor Wireless Networks, vol. 1, no. 1–2, pp. 89–124, 2005.
[56] L. Breiman, “Random forests,” Machine Learning, vol. 45, no. 1, pp. 5–32, 2001.
[57] C. Chang and C. Lin, “LIBSVM: A library for support vector machines,” ACM
Transactions on Intelligent Systems and Technology, vol. 2, no. 3, pp. 1–27, 2011.
[58] Y. Bengio, “Practical recommendations for gradient-based training of deep
architectures,” in Neural Networks: Tricks of the Trade, Springer, pp. 437–478, 2012.
[59] M. A. Matin and M. M. Islam, “Overview of wireless sensor network,” in Wireless Sensor
Networks - Technology and Protocols, IntechOpen, 2012.
[60] K. Akkaya and M. Younis, “A survey on routing protocols for wireless sensor networks,”
Ad Hoc Networks, vol. 3, no. 3, pp. 325–349, 2005.

54
[61] C. Perkins, E. Belding-Royer, and S. Das, “Ad hoc on-demand distance vector (AODV)
routing,” RFC 3561, 2003.
[62] Y. LeCun, Y. Bengio, and G. Hinton, “Deep learning,” Nature, vol. 521, no. 7553, pp.
436–444, 2015.
[63] Z.-H. Zhou, Ensemble Methods: Foundations and Algorithms, CRC Press, 2012.
[64] T. Hossain, M. U. Siddique, and H. T. Mouftah, “Dynamic coverage and connectivity
maintenance for WSNs,” Computer Communications, vol. 34, no. 10, pp. 1231–1241,
2011.
[65] D. Niyato, X. Lu, and P. Wang, “Machine-to-machine communications for smart cities,”
IEEE Communications Magazine, vol. 51, no. 6, pp. 84–90, 2013.
[66] A. Pantelopoulos and N. G. Bourbakis, “A survey on wearable sensor-based systems for
health monitoring and prognosis,” IEEE Transactions on Systems, Man, and Cybernetics,
Part C, vol. 40, no. 1, pp. 1–12, 2010.
[67] L. Da Xu, W. He, and S. Li, “Internet of things in industries: A survey,” IEEE
Transactions on Industrial Informatics, vol. 10, no. 4, pp. 2233–2243, 2014.

55
APPENDIX

56
Appendix A

Similarity Score

11% SIMILARITY SCORE

9% INTERNET SOURCES

6% PUBLICATIONS

<1% STUDENT PAPERS

Report pic

57
Appendix B

Paper Communication Mail

58
Certificate of Paper Presentation

59
Appendix C

Communicated Paper

60
61
62
63
64
65