Adversarial Machine Learning Attacks Against Network Intrusion Detection Systems: Classification Analysis
Dimitriya Mihaylova · 2025
In recent years much research has been focused on the improvement of Network-based Intrusion Detection Systems (NIDS) through the implementation of Machine Learning (ML) approaches. However, together with the numerous assets and promising results from the use of this emerging technology in traditional security systems, novel vulnerabilities arise from the inherent nature of the incorporated ML models. In this paper one such security violation, namely the Adversarial Machine Learning (AML) attack, is studied. While on the one hand, ML can be used to improve the security of the system, conversely it can be also exploited by an adversary, who can benefit from existing ML models to develop more refined attacks in a reduced time and with greater impact. An AML attack is an ML-driven intervention that contaminates the original data for training or testing the ML model of the NIDS, and results in degradation of the performance of the security system. This paper explores the nature of AML attacks and presents a classification analysis of their main types.