Analysis of Machine Learning Algorithms for Cyber Attack Detection in SCADA Power Systems

Abstract

Cybersecurity is a rapidly growing concern in many technological areas worldwide. Supervisory Control and Data Acquisition (SCADA) systems are especially vulnerable to cyber attacks due to increased inter-connectivity. SCADA systems need to be equipped with the proper tools and techniques to detect cyber attacks, distinguish them accurately from normal traffic, overcome cyber attacks when present, and prevent future cyber attacks from disrupting these systems. In this paper, we first analyze 10 well-known traditional machine learning algorithms in terms of how effective they are when detecting cyber attacks. Then, we construct a stacking ensemble learner using these methods via different meta learners. Our experiments show that ensemble methods perform better than individual methods, demonstrating the need for a more comprehensive solution when defending against cyber attacks in SCADA systems.