Improving the Performance of IIoT Intrusion Detection System Using Hybrid Synthetic Data

The convergence of Industrial Control Systems (ICS) and Information Technology networks has enabled manufacturers to embrace modern automation technologies with the Industrial Internet of Things (IIoTs). However, the existing ICS network protocols lack robust security mechanisms, as they were originally designed for isolated environments. This vulnerability has exposed ICS networks to emerging cyber threats, necessitating the development of robust intrusion detection systems to promptly alert security personnel. The effectiveness of an IDS heavily depends on the quality and quantity of its training data. Insufficient and imbalanced data present challenges that impact detection rates. Generative Adversarial Networks (GANs) offer a solution by generating realistic data based on existing information. This study utilizes Conditional Tabular GAN, known as CTGAN, tailored for generating tabular data, along with traditional fuzzing for synthetic data generation. Specific rules are defined to ensure data accuracy. Evaluation results demonstrate that IDS trained with a combination of CTGAN, fuzzing data, and existing data yields the best performance.