Study on Anomaly Detection Technique in an Industrial Control System Based on Machine Learning

Abstract

This study proposed an anomaly detection technique in an industrial control system using supervised and unsupervised machine learning algorithms. For the dataset for learning, the HIL-based Augmented ICS (HAI) dataset provided for the study on security in industrial control systems was used. For the learning model, Light Gradient Boosted Machine -- a supervised learning algorithm and One-Class Support Vector Machine and Isolation Forest as unsupervised learning algorithms were employed. The proposed technique is presented in this paper, which is organized as follows: Feature selection, Data preprocessing, Hyperparameter optimization and verification, and Experiment and analysis of results. The performance difference according to the algorithm and model configuration was exhibited through the experimental results. In addition, issues to be considered in model configuration and future study directions for anomaly detection techniques in industrial control systems were presented based on the experimental results.