Evaluating performance of scalable fair clustering machine learning techniques in detecting cyber attacks in industrial control systems

Abstract

The widespread use of machine learning to build intelligent systems which can make real-life decisions has led to the introduction of fairness notion in the machine learning techniques. Over the years, many fair machine learning algorithms have been established to reduce the discrimination factor in machine learning. The fair variants of machine learning techniques such as fair clustering models provide a solution to the biased data analysis problem. However, these models can produce fair, but less accurate results. Therefore, it is critical to analyze the performance of such fair clustering algorithms. In this paper, we use a scalable fair clustering algorithm to build Fairlet Decomposition (FD) model. Using this FD model, we validate the stated time linearity result of the algorithm using three different datasets: Internet of things (IoT), Industrial control system (ICS) and Secure Water Treatment plant (SWAT) and further quantify the fairness factor of FD model by computing the accuracy and other evaluation metrics. Our experimental findings show that the best performance of FD model was observed with IoT dataset with an overall accuracy of 83% and a low FPR of 10%.