EVOLUTIONARY ALGORITHMS FOR IMPROVING ENERGY EFFICIENCY AND SECURITY IN WIRELESS SENSOR NETWORKS

Abstract

Wireless Sensor Network (WSN) finds vast real-world applications in the field of energy control, security, health care, defense, and environment monitoring. WSNs are subdued by limited power with a specific battery backup. Due to the large distance between Sensor Nodes (SNs) and the sink, more power consumption occurs in the sensors. The limited energy of SNs is a major drawback to empower a large network coverage area. Therefore, the battery life and location of Cluster Heads (CHs) play an important role in increasing the efficiency and lifetime of SNs for long-term operation in WSNs. Researchers face significant challenges in developing more energy-efficient and secure clustering and routing protocols for WSNs. The bulk of existing routing protocols focuses on CH election while disregarding other important aspects of routing including cluster formation, data aggregation, and security, among others. Nature-inspired algorithms like Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Ant Colony Optimization (ACO), Gray Wolf Optimization (GWO), and Butterfly Optimization Algorithm (BOA) have been used for addressing important challenges in WSNs, such as sensor lifespan, transmission distance, and energy consumption. The introduction of hybrid techniques has proven to be more effective. Hybrid techniques like GA PSO, PSO-ACO, PSO-GWO, etc. have gained traction as viable solutions for bio inspired algorithms to improve the energy efficiency of WSNs. Such techniques have been found more optimum in comparison to the conventional techniques. The present thesis focuses on comparative analysis of hybridization techniques with conventional techniques for improving the energy efficiency of WSNs. In addition, an Improved Butterfly Optimization Algorithm (IBOA) has been proposed for global optimization problems in WSNs. Lastly, the present thesis focuses on addressing threats and security issues in WSNs through a feature selection and Machine Learning (ML) based intrusion detection pipeline. The Fast Correlation based Feature Selection (FCBFS) has been utilized as the feature selection method. The ML classifiers include Decision Tree (DT), Random Forest (RF), ix Naïve Bays (NB), Extra Tree (ET), and Extreme Gradient Boosting (XG-Boost). The contributions presented in this thesis are outlined below: • We propose a comprehensive review of Bio-inspired Hybrid Optimization Algorithms for Energy-Efficient WSN. We have aimed to discuss and compare various newly implemented, conventional, and hybrid methodologies for establishing a robust energy-efficient WSN wherein parameters like packet loss, energy, throughput, delay, and overhead have been utilized. Various open issues and challenges in WSN development using bio-inspired optimization techniques such as network stability, network dynamic character, secure transmission lines, methods to improve QoS, etc., have been addressed. • We propose a variable sensor modality IBOA for global optimization problems. The modified optimization approach focuses on unconstrained issues, performs on restricted problems, and remains the future scope of this work. • We propose an FCBFS method with XG-Boost for the National Security Laboratory-Knowledge Discovery Dataset (NSL-KDD) intrusion detection benchmark dataset to address the threats and security issues in a complicated WSN for IoT applications. Evaluation metrics such as accuracy, precision, recall, and F1-Score have been calculated to gauge the performance and robustness of the proposed research work. A classic accuracy score of 99.84% is achieved in the case of the XG-Boost classifier, wherein the best ten obtained features were selected after applying the proposed FCBFS. The proposed technique, which has ten features, outperforms the existing techniques in the literature for the NSL-KDD dataset. Experimental analysis has been done extensively to prove the efficacy of the developed solutions.