<title>Abstract</title> Intrusion Detection Systems (IDS) are essential for securing computer networks against malicious activities. However, the rise of adversarial attacks seriously threatens the robustness and efficacy of IDS models. With the increasing prevalence of adversarial attacks on intrusion detection systems (IDS), it has become crucial to develop robust defence mechanisms to make sure the integrity and reliability of these systems. This paper presents a novel approach that combines Particle Swarm Optimization (PSO), Gradient Boosting Machines (GBM), genetic operators, and deep neural networks (DNN) with defence mechanisms to improve the resilience of IDS in order to stop adversarial attacks. The proposed approach starts with a feature engineering stage, where PSO and GBM are utilised to select and optimise the most informative features from the input dataset. Genetic operators are then employed to refine the feature selection process further, ensuring the creation of robust and discriminative feature subsets. In the subsequent stage, a deep neural network model is constructed with defence mechanisms, including adversarial training, input perturbation, and ensemble learning. These defence mechanisms work synergistically to monitor and improve the IDS's capacity to find and classify normal and adversarial network traffic accurately. The well-known NSL-KDD dataset is utilised to assess how successful the suggested method is. Experimental findings show that the integrated framework outperforms current techniques. Additionally, the system shows increased resistance to various adversarial techniques, such as evasion, poisoning, and adversarial samples. Overall, this study bridges the gap between adversarial attacks and intrusion detection, offering a powerful defence framework that can be integrated into existing IDS architectures to extenuate the consequence of adversarial threats and ensure the integrity and reliability of network security systems.
Authors: Kanak Giri, Pankaj Dadheech, Mukesh Kumar Gupta
DOI: https://doi.org/10.21203/rs.3.rs-5350806/v1
Publish Year: 2024
