Leveraging Tree-Based Machine Learning and Explainable AI for UAV GPS Spoofing Detection

The increasing reliance on GPS systems in Unmanned Aerial Vehicles (UAVs) has made the need for an efficient and robust system to detect, mitigate, and prevent potential disruption during operation like jamming and spoofing. Machine learning (ML) approaches have played a vital role in GPS spoofing detection which can be enhanced further by gaining the insights of the underlying factors that leads to prediction. For this, a popular Explainable AI(XAI) framework named Shapley Additive Explanations (SHAP) is applied providing feature explanation of benign and malicious classes in ML model predictions. This study compares the four tree-based ML models namely Random Forest, eXtreme Gradient Boosting (XGBoost), Categorical Boost (CatBoost), and Light Gradient Boosting Machine(LightGBM) on a real-world GPS spoofing dataset for UAVs. The features of best performing model, XGBoost with an accuracy of 99.89% is further interpreted using SHAP technique to provide feature-specific explanations for the results. These explanations of the framework aim to improve understanding, build trust, and support effective defenses against GPS spoofing.