Detecting Cyber Threats with Limited Dataset Using Generative Adversarial Network on SCADA System

Cyber-attack and threat pose a significant risk to critical infrastructure especially in Supervisory Control and Data Acquisition (SCADA). These are essential to manage energy, water and industrial manufacturing processes. Detecting cyber-attack on SCADA systems is very important to keep the integrity and security of SCADA systems. However, currently the biggest challenge in SCADA cybersecurity is limitation of dataset. There is very limited public dataset, and generating realistic simulation data is challenging. Therefore, finding or create a dataset to training robust detection model is not easy. We propose to use Generative Adversarial Network (GAN) to address this challenge. GAN is one of the well-known deep learning algorithms that generate synthetic data that resembles real data; therefore, GAN is useful when training data is limited. We research to find the best model and layer structure to detect cyber-attack. Our test result shows higher than 97% accuracy. This paper details what GAN is, browsing dataset, experimental and result.