Development of Machine Learning Subsystem in Intrusion Detection System for Cyber Physical System
Fabian Savero Diaz Pranoto, Yudistira Dwi Wardhana Asnar · 2023
The growth of cyber-physical systems (CPS) has caused a rise of cyber-attacks. These attacks such as Stuxnet have great social and economic impact. Research and development of intrusion detection systems have emerged to protect CPS. Many research results have found that anomaly-based IDS using deep learning models have high detection rates towards attacks on various datasets. However, these models are not usually integrated with a running CPS. In this study, we propose an IDS for Process Instrumentation Trainer (PIT), a CPS in Institut Teknologi Bandung. This paper focuses on the detection and model training functionalities and their integration towards the whole system. A comparison is done between multiple models. Trained models were tested on the Secure Water Treatment (SWaT) dataset and PIT dataset. Test results show that 1D-CNN has the best detection performance. The model was able to detect 29 out of 36 attacks on SWaT and 5 out of 5 attacks on PIT. However, robustness of the proposed IDS has yet to be tested towards adversarial attacks.