Identification and Interpretation of Aggressive Driving Behavior under Time Pressure

Abstract

In the context of accelerated modern lifestyles, time pressure has become a significant factor influencing driving behavior. It often leads drivers to adopt aggressive driving practices to cope with traffic congestion or pursue efficiency, thereby increasing accident risks. This study investigates the impact of time pressure on driving behavior through driving simulator experiments. Initially, multisource data were collected and driving behavior samples were extracted. Subsequently, the K-means algorithm was employed to cluster driving styles, identifying three types: aggressive, normal, and cautious. Following this, statistical analysis was performed to reveal the association between time pressure and driving behavior. An ensemble-learning approach was developed to identify aggressive driving behavior under differing time-pressure conditions. The XGBoost ensemble achieved 94.5% accuracy, 94.53% precision, 94.5% recall and a 94.49% F1-score. For comparison, we trained a second ensemble model (Random Forest) and three single classifiers (Decision Tree, Logistic Regression, and Support Vector Machine). Both ensemble models outperformed all single classifiers on every major metric, confirming the advantage of ensemble learning for this identification task. The SHapley Additive exPlanations (SHAP) method was then applied to explore the influential factors for different driving behavior types. SHAP analysis interprets that aggressive driving behavior is notably dependent on features such as the driver’s identity category, driving experience, mileage driven, educational background, and time pressure. The findings have potential applications in the implementation of sensor-based real-time intervention strategies, which could significantly enhance road safety in high-pressure driving environments.