From Nature to Networks: Harnessing Funnel Dynamics for Effective Intrusion Detection in IoT Environments

Abstract

Abstract The rapid evolution of technology and the proliferation of interconnected systems have given rise to an alarming increase in cyber threats. In this context, Intrusion Detection Systems (IDS) have emerged as crucial tools for detecting and mitigating unauthorized access and malicious activities within computer networks. This paper introduces a novel deep learning architecture inspired by the working principle of a funnel for detecting intrusions in IoT networks. The proposed architecture incorporates a feature selection model that leverages the hunting behavior of the yellow saddle goatfish and the swarm behavior of birds. This nature-inspired optimization algorithm enhances the deep learning model's ability to learn high-quality features, thereby improving the intrusion detection rate. Additionally, the proposed approach addresses the challenge of working with diverse environmental datasets by analyzing the identification capabilities for modern attacks separately. The simulation of the proposed framework is conducted using MATLAB software, and performance evaluation is carried out using various performance metrics. The proposed architecture demonstrates improvements of 1.51% for the KDD-CUP99 dataset, 2.87% for the NSL-KDD dataset, and 22.29% for the UNSW-NB15 dataset. These enhancements highlight the efficacy of the proposed architecture in advancing intrusion detection capabilities in IoT networks. The promising results obtained from this study open up several exciting avenues for future research.