Intrusion detection in Wireless Sensor Networks using T-MAXOUTNET: Taylor series-based Deep Maxout Network

Abstract

Wireless sensor networks (WSNs) are susceptible to various types of attacks, which are degraded the performance of entire network. Hence, the primary purpose of this research is to design and develop intrusion detection system in WSN for detecting the attackers. The series of steps carried out for detecting the intrusions in WSN are network initialization phase, setup phase, routing phase, and the intrusion detection phase. The network initialization phase involves three models, such as energy model, mobility model and trust model, which are employed for detecting the shortest path in network. The cluster head selection is carried out in setup phase through O-SEED method, which is optimized through Rider Cat Swarm Optimization (RCSO). RCSO model is intended by the integration of Rider optimization algorithm (ROA) and Cat Swarm Optimization algorithm (CSO). In the routing phase, the multipath transmission is done using Rider Atom Search Optimization (RASO). RASO algorithm is designed by incorporating the ROA and Atom Search Optimization (ASO). In the intrusion detection phase, the packet log data is recorded, and the important attributes, like BOT-IOT data is taken for further processing. Then, the feature selection is performed through Kendall tau distance (KTD) to select the appropriate features for detecting the intrusions. After that, the selected attributes are given to the Deep Maxout Network for intrusion detection. The Deep Maxout Network is trained using the developed optimization algorithm, termed as Taylor Rider Atom Search Optimization (Taylor RASO). Here, the proposed Taylor RASO is designed through integrating Taylor series, ASO, and ROA. The performance of the proposed scheme is evaluated through three metrics, namely accuracy, energy, and throughput, and is compared with that of the existing approaches in order to reveal the efficiency of developed technique. The developed Taylor-RASO outperformed other methods with maximum accuracy of 0.935, maximal average residual energy 0.087J, maximal throughput of 87.91%, correspondingly.