An Efficient Explainable Artificial Intelligence (XAI)-Based Framework for a Robust and Explainable IDS
This work proposes an efficient Explainable Artificial Intelligence (XAI)-based framework designed to enhance both the robustness and explainability of IDS, and integrates traditional statistical methods with XAI techniques, providing transparent and interpretable decision-making while maintaining high detection performance.
Abstract
Artificial Intelligence (AI)-based Intrusion Detection Systems (IDS) significantly advance network security by leveraging Machine Learning (ML) and Deep Learning (DL) for highly accurate and dynamic cyber threat detection. However, a critical limitation of current AI-based IDS is their inherent “black box” nature, which disrupts the decision-making processes, thereby compromising trust and accountability. In response to these challenges, we propose an efficient Explainable Artificial Intelligence (XAI)-based framework designed to enhance both the robustness and explainability of IDS. Our two-stage process integrates traditional statistical methods with XAI techniques, specifically SHAP and LIME, for feature selection and explanation analysis, providing transparent and interpretable decision-making while maintaining high detection performance. Our experimental evaluation, conducted using the CIC-DDoS2019 and CICIoT2023 datasets, demonstrates that the framework can sustain high detection accuracy while significantly enhancing the interpretability of IDS decisions. Moreover, the proposed feature reduction process lowers the computation effort of the XAI techniques, resulting in up to 87% faster.