Deep Learning-Enhanced Cluster Head Optimization for Intrusion Detection in Wireless Sensor Networks
Alaa Sabree Awad, Mohammed Khalaf, Mahmood Abdulrazzaq Alsaadi · Ingénierie des systèmes d information · 2024
Securing wireless sensor networks (WSNs) is imperative, particularly for an intrusion detection system (IDS) deployed in inaccessible terrains, which are susceptible to a multitude of security threats.This study introduces a novel IDS framework employing deep learning to curtail a spectrum of cyber assaults, including but not restricted to DoS, tampering, and sinkhole attacks.In addition to that, The crux of the proposed model depends on the optimization of the cluster head (CH) selection among sensor nodes, where nodes with superior energy levels are preferentially considered for CH roles.This research advances beyond energy-centric CH selection criteria by incorporating delay and distance considerations, culminating in the development of the Particle Distance Updated Bottlenose Dolphin Optimization (PDU-BDO) algorithm for the CH election process.Subsequently, an intrusion detection analysis is conducted via an optimized deep hierarchical voting neural network (DHVNN), with the PDU-BDO algorithm facilitating the neural network's (NN) weight tuning during training.The efficacy of the PDU-BDO algorithm, benchmarked against three extant methodologies using the NSL-KDD dataset, reflects significant performance enhancements, yielding an accuracy of 91.6%, precision of 88.2%, recall of 86%, F1-score of 82%, and a kappa score of 71.4%.Moreover, deep learning-based IDS against adversarial attacks is corroborated through real-world application scenarios, signalling a stalwart defense mechanism within the WSN paradigm.