OPTIMIZATION-DRIVEN DEEP RECURRENT NEURAL NETWORK FOR INTRUSION DETECTION AND HEALTH RISK ASSESSMENT IN WIRELESS BODY SENSOR NETWORK

Abstract

Wireless body sensor network (WBSN) has gained great attention in the environmental and military applications, but security is the major issue, nowadays. In addition, the data exchanged through the wireless sensor network (WSN) is vulnerable to several malicious attacks because of the physical defense equipment needs. Hence, various intrusion detection methods are required for defending against such attacks. Accordingly, an effective method, named deep recurrent neural network (Deep RNN), is proposed in this research for detecting the intrusion in WBSN. At first, the WBSN nodes are utilized to sense the data from the health records of patient for acquiring certain parameters to make risk assessment. Then, WBSN nodes transmit the data to the target nodes using the obtained parameters. After the determination of parameters, the WBSN nodes are responsible to collect the information of the patient and transfer the obtained information to cluster heads (CHs) based on the hybrid harmony search algorithm–particle swarm optimization (HSA–PSO). HSA–PSO is utilized for identifying the optimal CH node iteratively. From the selected CHs, secure communication is done to exchange the data packets. After that, the KDD features are extracted and intrusion detection is done using the proposed Deep RNN. After the genuine users are detected, the classification is done using fractional cat-based salp swarm algorithm (FCSSA) for the risk assessment. The performance of the intrusion detection and health risk assessment in WBSN based on the proposed model is evaluated based on accuracy, sensitivity, and the specificity. The developed model achieves the maximal accuracy of 95.79%, maximal sensitivity of 95.97%, and the maximal specificity of 95.61% using Cleveland dataset.