An explainable multi-modal fusion framework for robust detection and classification of GNSS jamming attacks in critical infrastructures

Abstract Global Navigation Satellite Systems (GNSS) serve as critical components for positioning and timing in modern infrastructures, yet their inherent signal vulnerability makes them prime targets for intentional jamming attacks. While existing detection methods often rely on single-modality data, they frequently fail to capture the complex temporal and spectral signatures of diverse interference types, leading to a "black-box" decision process that lacks engineering trust. This study proposes an explainable multi-modal intrusion detection framework that fuses statistical features, raw 1D temporal dynamics via 1D-CNN, and 2D time-frequency representations using EfficientNetB0. The integrated features are classified using an XGBoost ensemble, achieving a high classification accuracy of 99.33% on a diverse dataset of six jamming classes. To ensure the reliability of the system for security operators, a comprehensive Explainable AI (XAI) strategy using SHAP and Grad-CAM is implemented. This research provides a transparent and robust security solution for protecting GNSS dependent systems against evolving electronic threats. Contribute 74.22% to the detection mechanism, highlighting the importance of raw signal dynamics over traditional image-centric approaches.