An Interpretable Ensemble Learning Method for GPS Spoofing Detection with Feature Selection

Global Positioning System (GPS) spoofing poses severe threats to navigation safety, necessitating robust detection mechanisms with enhanced interpretability. While existing methods have been explored for GPS spoofing detection, algorithmic interpretability is rarely addressed and detection performance remains suboptimal. This study proposes Stack-TabNet, a stacked ensemble learning framework integrating various decision trees and the attentive transformer-based TabNet network. To address model opacity, an interpretable feature attribution mechanism is employed to quantify feature contributions and guide optimization. Experiments are conducted on a complex dataset comprising authentic and spoofed GPS signals across four classes, characterized by high-dimensional signal metrics and severe class imbalances. The initial model utilizing all available features demonstrates robust detection capability. Subsequently, an optimized variant employs a subset of top-ranked features identified by the interpretation mechanism, yielding further improved accuracy. Comparative analysis confirms that the proposed framework outperforms traditional baselines, achieving a maximum accuracy of 95.91% after SHAP-guided feature selection, which exceeds all baseline models by at least 3.54%. The analysis identifies Pseudorange and Time of Code Delay as the most important features, consistently ranking at the top across all base learners within the ensemble architecture.