A weight optimized deep learning model for cluster based intrusion detection system

In wireless sensor networks (WSNs), the implemented conventional intrusion detection frame works need more energy and computation time, which impact the overall WSNs lifespan. Additionally, few of these models create considerable IDS traffic volume, which creates congestion band width constrained of WSN. This paper presents a new hierarchical type intrusion detection system for determining the malicious sensor nodes. This recommended intrusion detection framework is mainly works in two levels (1) Rule-based specification-based detection system (Level 1) and (2) Anomaly-based IDS via clustering (Level 2). In the initial phase, the IDS agent monitors the other sensor node for maliciousness using the specific set of rules. Owing to the second phase, the cluster head monitors the other cluster head for maliciousness, which is performed via Deep Belief Network (DBN) model, which is already trained with the node parameters. In order to make precise prediction process, this paper involves the optimization concept, which optimally tunes the weight of DBN by introducing a new hybrid optimization algorithm which named as Cuckoo Insisted Lion Algorithm, is the combination of Cuckoo Search and Lion Algorithm. In this recommended model, the overall performance is verified over the other modernization models relating to certain prediction parameters.

Automated summary

This paper proposes a new hierarchical intrusion detection system for accurately determining malicious sensor nodes. The system operates in two levels and incorporates optimization techniques to improve prediction accuracy, performing better compared to other modernization models.