Adversarial attacks on network intrusion detection systems using machine learning
Kulvinder Singh, Ankit Panigrahi, Deepanshu Jindal · 2025
In the past few years, machine learning (ML) based NIDS solutions have shown great promise to detect and prevent cyber attacks in real-time. Nonetheless, these systems are becoming much more vulnerable to adversarial attacks, where attackers create similar inputs which can skirt around the detection. This paper illustrates a variety of techniques designed to fool NIDSs through the use of adversarial attack against machine learning models. We present evasion and poisoning attacks to target vulnerabilities of state-of-the-art NIDS algorithms like decision tree, support vector machine, and deep learning models. This attack works by perturbing network traffic data to generate adversarial samples destined for some location that the automotive is not meant to reach, hence we lower detection rates. Furthermore, we evaluate the resilience of these systems with adversarial training and defensive strategies towards greater robustness. We highlight the potential of robust defenses to significantly enhance detection performance and show that, even though new attacks can evade existing defenses, they are still blocked by another defense mechanism.