A bio-inspired hybrid deep learning model for network intrusion detection

Advances in network communications have resulted in an explosion of connected device usages in several business domains with phenomenal surge in network traffic. Though the capabilities of the business models are empowered with Internet driven service providers, they are vulnerable to severe security threats. Intrusion detection systems capable of identifying malicious attacks from traffic data are potential tools in securing organizational resources from unauthorized access. This paper proposes an intrusion detection system modeled as a two stage framework with feature selection performed by a Generalized Mean Grey Wolf Algorithm and an ElasticNet Contractive Auto Encoder. It is tested with the NSL-KDD and BoT-IoT datasets resulting in an overall classification accuracy of 0.999 for binary and multi-class classification of attacks. This system performs classification with an optimal subset of traffic features and surpasses the performances of state-of-the-art intrusion detection systems. Further, the proposed model also demonstrates its effectiveness to learn from unlabeled data without additional training, signal drifts from normal data and generalization with arbitrary test data.