The Role of Rule Mining in Generating Synthetic Cyber-Physical System Attack Samples

In recent years, Cyber-Physical Systems (CPSs) have increasingly been exposed to potential exploitation by the sophisticated adversary due to their vulnerabilities. The ever-evolving threat landscape for CPSs can impact their control logic, leading to system and process disruptions. Several Machine Learning (ML) based Intrusion Detection Systems (IDS) have been proposed to detect cyber threats in CPS. However, the issue of class imbalance in CPS datasets must be addressed to develop robust and effective security controls to mitigate cyber threats to CPS. We propose a novel method to generate synthetic ICS data by customising data generation methods specifically tailored for transactional datasets. The proposed scheme merges the process of mining frequent itemsets with a generative modeling method. A collection of items that frequently appear together is referred to as an itemset. We verify the validity of generated synthetic samples by comparing them with the original data samples. Furthermore, we apply three machine learning classifiers to evaluate the quality of the generated synthetic datasets with the aim to address the issue of class imbalance. The generated synthetic datasets to address the issue of class imbalance. The synthetic datasets generated contribute to the development of robust security controls which can detect evolving threats faced by CPSs.