Detecting GPS spoofing in smart AVS: an accuracy-based machine learning approach

GPS spoofing attacks represent a significant threat to autonomous vehicular systems (AVS), where malicious actors transmit counterfeit GPS signals to manipulate the position or navigation of vehicles. Addressing GPS spoofing is crucial for ensuring reliability, safety, and operational integrity in AVS. If undetected, such attacks could lead to dangerous mis-navigation, system failures, or compromise passenger safety. This paper uses statistical analysis and machine learning to present a comprehensive method for detecting GPS spoofing. The proposed methodology involves two phases: model construction and model evaluation. In the construction phase, PyCaret is initialized with a GPS dataset containing spoofed and non-spoofed signals, with key features such as PDOP (Position Dilution of Precision), HDOP (Horizontal Dilution of Precision), and VDOP (Vertical Dilution of Precision) specified. Preprocessing steps handle missing data and categorical encoding, while multiple classification algorithms are tested, and hyperparameters are optimized. In the evaluation phase, models are assessed using MAE, MSE, RMSE, and R2 metrics. Feature importance is analyzed, focusing on PDOP, HDOP, and VDOP’s role in identifying spoofed signals. Visualizations, including scatter plots, histograms, and heat maps provide insights into feature distributions and relationships with spoofing. The results indicate that PDOP plays a key role in distinguishing spoofed signals, making this approach effective in enhancing AVS reliability by mitigating spoofing risks and improving passenger safety. This methodology offers a robust framework for enhancing GPS spoofing detection which emphasizes the ability to detect GPS spoofing directly improving the reliability and safety of autonomous vehicles.