Enhanced Dynamic Temporal Features Based Lightweight Xgboost Framework for in-Vehicle Can Bus Attack Detection

The Controller Area Network (CAN) is widely utilized in automotive systems. However, in-vehicle CAN networks face multiple security threats due to the lack of encryption and authentication mechanisms, while conventional rule-based or shallow machine learning detection methods exhibit limited generalization capabilities against unknown attack patterns. This paper proposes an XGBoost-based attack detection framework integrating dynamic temporal statistical features, which constructs a unified detection model for multiple attack types through multidimensional feature engineering. Experimental validation on public in-vehicle CAN bus datasets demonstrates that the proposed method achieves 98.5 % F1-Scorein detecting DoS, replay, and other attack scenarios, outperforming comparable models by 6 % mostly in accuracy while reducing model training time by over <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{9 0 \%}$</tex>.