A Robust Network Intrusion Detection System Using Random Forest Based Random Subspace Ensemble to Defend Against Adversarial Attacks

In recent years, machine learning (ML) has had a significant influence on the discipline of computer security. In network security, intrusion detection systems increasingly employ machine learning techniques. Approaches based on machine learning have substantially improved the efficacy of intrusion detection. Adaptive adversaries who comprehend the underlying principles of ML techniques can initiate attacks against the classification engine of an intrusion detection system. Malicious actors exploit machine learning model vulnerabilities. Network security, specifically intrusion detection systems, requires the development of defensive strategies to combat this threat. The RF-RSE (Random Forest based Random Subspace Ensemble) and RF-RSE-AT (RFRSE-Adversarial Training) methods are proposed as network intrusion detection systems to defend against adversarial attacks. The methodologies proposed are evaluated using the NSL-KDD dataset. The RF-RSE method demonstrates remarkable resistance to adversary attacks. The RF-RSE-AT method performs exceptionally well in correctly identifying network traffic classes when presented with adversarial attacks, and it maintains its accuracy even when no attack is present.