Answers are grounded in this paper and cite their sources.
Cyber-physical systems are integral to various technological ecosystems, enabling interactions between computing systems, communication networks, and physical processes. Machine learning algorithms are employed to train anomaly-based Intrusion Detection Systems (IDS) to secure these ecosystems. However, they remain vulnerable to adversarial attacks, where training and testing data can be manipulated to degrade performance. This paper evaluates machine learning algorithms in anomaly-based IDSs under adversarial attacks using two datasets from the power system context. The results indicate that ensemble algorithms excelled against JSMA attacks, while single classifiers performed better against FGSM attacks, with Decision Tree and Random Forest being the top performers in their respective groups.
No takes yet. Share an insight, caveat, or question.
Silva et al. (2024) studied this question.