Leveraging Temporal and Statistical Features for Improved Falsification Attack Detection in Vehicular Ad Hoc Networks

Falsification attacks in Vehicular Ad Hoc Networks (VANETs) pose a significant threat to vehicle-to-vehicle communication integrity by manipulating Basic Safety Messages (BSM) that convey critical information such as position and speed. Traditional detection methods relying solely on instantaneous position or velocity variations are less effective in identifying gradual or fluctuating attack patterns. To address this limitation, this study introduces a rolling-window-based feature engineering approach to capture vehicle movement dynamics more comprehensively. The engineered features include position variance, total speed, average speed, and speed standard deviation. The importance of each feature in detecting falsification attacks was evaluated using the Mean Absolute SHAP Value (MASV). Experimental results demonstrate that feature engineering yields a 16.50% improvement in CatBoost model recall compared to the baseline version without feature engineering. The proposed method achieves 98.12% recall in multiclass classification, representing a 12.86 percentage point improvement over the previous work (85.26%), with substantial gains in detecting challenging attack types such as Type 8 and Type 16.