Information System Security Reinforcement with WGAN-GP for Detection of Zero-Day Attacks
Ziyu Mu, Xiyu Shi, Safak Dogan · 2024
Growing sophistication among cyber threats has posed increasing challenges to the security and reliability of information systems, especially in the face of zero-day attacks that exploit unknown vulnerabilities. This paper introduces an innovative application of Artificial Intelligence (AI), specifically the adoption of Wasserstein Generative Adversarial Networks with Gradient Penalty (WGAN-GP), to support Intrusion Detection Systems (IDS) to strengthen defences against such attacks. This research focuses on using the WGAN-GP to generate network traffic data in simulating the unpredictable patterns of zero-day attacks. It utilises the widely used network traffic dataset NSL-KDD to conduct data expansion. This approach leverages data generated by the WGAN-GP to train detection systems, enabling them to learn and identify subtle signatures of zero-day attacks. Experimental evaluation demonstrates that the WGAN-GP model can improve the accuracy of zero-day attack detection. In comparison to other methods, such as Convolutional Neural Networks (CNN), the detection accuracy is increased by 2.3% and 2% for binary and multi-classification, respectively. This work shows that combining IDS with advanced generative AI models, such as WGAN-GP, can significantly enhance the security of information systems in identifying and mitigating risks posed by zero-day attacks.