Intelligent Approach for Anomaly Detection using Machine learning Techniques in Industrial Control system

Anomaly detection is crucial for cybersecurity in industrial control systems due to the continuous injection of malicious payloads by attackers. This study examines the effectiveness of various machine learning algorithms in identifying these payloads throughout the system. The experiments utilize three datasets: the Hardware in Loop HAI dataset, the Gas Pipeline dataset, and the ICS cyber-attack dataset. Python programming and a high-performance GPU facilitate the experimental procedures. This comprehensive analysis, based on these experiments, highlights that the CNN-Dense net and Random Forest algorithms deliver the highest performance in terms of accuracy, precision, recall, and Receiver Operating Characteristic (ROC) curve across all datasets. It is important to note that other algorithms also perform comparably to CNN-Dense net and Random Forest. Consequently, the choice of machine learning algorithm should be guided by the specific data generated by the application system.