Machine learning-based Intrusion Detection Systems (IDS) play a critical role in identifying network threats, but they are increasingly vulnerable to adversarial attacks that can manipulate inputs and evade detection. This challenge is particularly important in time-sensitive environments such as 6G and Open RAN (O-RAN). This paper proposes a robust IDS framework enhanced with Explainable Artificial Intelligence (XAI) to detect and mitigate adversarial behavior in real time. Instead of relying on computationally intensive methods like adversarial training, the approach integrates XAI to analyze how input data influences model decisions and to identify abnormal patterns.
A key contribution is the use of SHAP-based feature importance distributions to characterize normal system behavior during training. At run-time, incoming data are evaluated against this baseline, and deviations in SHAP values are used to detect potential adversarial manipulation. The framework is validated in an O-RAN environment, focusing on detecting RRC signaling storm attacks. Results show that the proposed method improves detection performance and enables faster, zero-touch responses to emerging threats, enhancing the overall resilience of IDS systems. Overall, this work demonstrates how combining XAI and IDS can provide more transparent, adaptive, and effective security mechanisms for next-generation wireless networks.
Robust_Intrusion_Detection_System_with_Explainable_Artificial_Intelligence