Analysis of GPS spoofing attacks in unmanned aerial vehicles and detection using artificial intelligence

Unmanned Aerial Vehicles are utilized in a wide range of fields, including military operations, civil infrastructure applications, agricultural monitoring, and disaster management. The increasing level of autonomy in these systems has led to a growing dependence on various technologies, primarily Global Navigation Satellite Systems. However, positioning and navigation processes that rely on GPS infrastructure are vulnerable to cyberattacks due to the weak, unauthenticated, and openly accessible nature of GPS signals. This study experimentally analyzes the effects of GPS jamming and spoofing attacks on UAV systems and proposes an artificial intelligence-based detection system capable of identifying such threats in real time. In the experimental setup, fake GPS signals were generated using HackRF One, and both jamming and spoofing scenarios were implemented on a UAV platform equipped with a Pixhawk 2.4.8 flight control unit. During these scenarios, data were collected via telemetry and analyzed using seven different machine learning algorithms. The algorithms; AdaBoost, Artificial Neural Networks, K-Nearest Neighbors, XGBoost, Random Forest, Support Vector Machines, and CatBoost; were evaluated based on accuracy, precision, recall, and processing time. Among these, XGBoost was identified as the most suitable model, achieving a high accuracy of 96.40% and a low test time of 0.0101 seconds. The selected model was integrated into the UAV system via a Raspberry Pi and successfully tested in real-time conditions. The developed system, which operates on low-cost hardware and open-source software, is capable of detecting both spoofed GPS signals and signal disruptions with high accuracy, thereby enhancing the mission safety of UAVs. The findings of this study indicate that the proposed approach has the potential to improve GNSS security in both military and civilian UAV applications. Furthermore, future developments may enable the system to become more robust, practical, and scalable through the integration of multi-sensor data and the implementation of adaptive learning mechanisms.