Evaluation of Machine Learning Algorithms for Anomaly Detection in Industrial Networks

Abstract

The cyber-physical security of Industrial Control Systems (ICSs) represents an actual and worthwhile research topic. In this paper, we compare and evaluate different Machine Learning (ML) algorithms for anomaly detection in industrial control networks. We analyze supervised and unsupervised ML-based anomaly detection approaches using datasets extracted from the Secure Water Treatment (SWaT), a testbed developed to emulate a scaled-down real industrial plant. Our experiments show strengths and limitations of the two ML-based anomaly detection approaches for industrial networks.