An ML Approach for Detection of GNSS Spoofing

Global Navigation Satellite Systems (GNSS) are pivotal for autonomous vehicle navigation, but are vulnerable to spoofing attacks, where malicious actors alter signals to transmit misleading data. This paper proposes a spoofing detection app focusing on anomalies such as wrong path, speed modification, signal interrupt, erratic behavior, U-turn, sudden jumps, and sudden lane changes. The detection model combines a multi-layered LSTM, Kalman filter and Haversine formula, trained and tested on a custom dataset developed using the apps LocaEdit and GNSSLogger. A weighted sum approach combines the results of these components, achieving an accuracy of 94.64%. In conclusion, the proposed model displays a robust and balanced performance across different spoofing scenarios, highlighting its reliability and potential for real-time applications in navigation systems.