Modeling Attacks on Machine Learning Components of Intrusion Detection Systems

Intrusion detection systems (IDS) perform a critical security function and can detect both known and unknown and more complex attacks. However, intrusion detection systems based on machine learning mechanisms (primarily deep learning) are susceptible to so-called adversarial attacks. This study examines the impact of attacks on the machine learning components of intrusion detection systems used in complex heterogeneous infrastructures by simulating such attacks. Various machine learning models are presented to detect anomalies in most cases used in intrusion detection systems. For the first time, a simulation of an adversarial attack on the machine learning components of IDSs was carried out. The dependence of the influence of the FGSM attack on the efficiency metrics: Precision, Recall, f1 was built.