A Technique for Protecting Machine Learning Components of Intrusion Detection Systems from Evasion Attacks

Today, intrusion detection systems (IDS) are important means to protect network infrastructure from cyber attacks. Digitalisation requires the use of tools that can cope not only with known types of attacks, but also with previously unspecified ones. Machine learning (ML) can be used to defend against such threats. This paper presents a methodology for defending against evasion attacks on machine learning (ML) components of intrusion detection systems. For the first time, Long short-term memory (LSTM) defence modelling against Fast Gradient Sign Method (FGSM) was performed for IDS ML components. Precision, Recall, and F-measure metrics were used for evaluation. The study shows the possibility of protecting IDS ML components from evasion attacks and provides an effective technique for such protection.