Reducing overfitting in deep learning intrusion detection for power systems with CTGAN

Deep learning based Intrusion Detection Systems (IDS) show potential, in identifying zero day attacks in power systems by learning common attack patterns. However the challenge lies in the scarcity of training data and the intricate dynamics of power systems leading to overfitting and reduced detection accuracy. This study introduces an approach to combat overfitting in learning based IDS for power systems by leveraging Conditional Tabular Generative Adversarial Networks (CTGAN) for data enhancement. We delve into the drawbacks of existing data augmentation techniques in power systems. Emphasizes the benefits of using CTGAN to generate diverse synthetic data. The research then outlines the construction and assessment of a CTGAN augmented dataset integrated into a learning based IDS for zero day attack detection. Experimental findings demonstrate enhancements, in model adaptability and detection precision compared to models trained on genuine data. Our method presents a resolution to tackle overfitting issues and bolster the efficiency of learning based IDS in safeguarding power systems against emerging cyber threats.