A Robust and Trustworthy Intrusion Detection System Using Adversarial Machine Learning and XAI

Network attacks are increasingly sophisticated. Advances in Artificial Intelligence (AI), particularly deep learning, have improved intrusion detection systems (IDS). However, deep learning (DL) in cybersecurity faces challenges, such as imbalanced training data, vulnerability to adversarial attacks and a lack of transparency regarding AI systems. To address these problems, we developed the RobustAdvTrain (Robustness Adversarial Training) framework to train IDS models for high accuracy and resilience against adversarial attacks. This framework provides explainable AI for transparent IDS predictions. We propose the sAoEGAN (self-Attention on Explanation Generative Adversarial Network) model, which combines explainable AI and self-attention mechanisms to generate high-quality adversarial samples. Our approach improves intrusion detection, resilience to adversarial samples, and transparency in deep learning-based IDS systems.