Cyber-attack detection in cyber-physical systems using supervised machine learning

Abstract

Cyber-Physical Systems (CPS) are where the physical processes are controlled by computation and other technology components. Although, the collaboration of the computer with traditional physical infrastructure can improve the efficiency of such facility-based systems. However, it increases the scope of attack from physical security to a cybersecurity perspective. Thus, it becomes critical for authorities of such systems to be able to identify the cyber-attacks on such systems and can impact the functioning of such geographically distributed systems. Machine learning techniques have been used to accurately parse such data of Cyber-physical systems to detect attacks. In this paper, we use four different supervised machine learning algorithms to build models to detect cyber-attack activities on a CPS water treatment plant. The result of the four classification models K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Decision Tree (DT), and Random Forest (RF) are compared on the basis of evaluation matrices to perform the comparative analysis. The comparative analysis results show that the DT model performs better than the other models with an overall accuracy of 99.9% and other improved evaluation metrics.