Seasonal driving style assessment for truck drivers: Integrating alarm response behaviors

Truck drivers exhibit significant individual variability in their responses to in-vehicle alarms, which critically impacts road freight safety. This study aims to dynamically assess truck driving styles by integrating alarm-response behaviors and exploring their seasonal variations. We established a comprehensive framework utilizing 20 indicators that encompass vehicle operations, driving behaviors, and alarm responses based on real-world GPS and alarm data. A K-means++ clustering algorithm coupled with an XGBoost (Extreme Gradient Boosting) model was developed to classify and identify driving styles across four seasons, while SHAP (SHapley Additive exPlanations) was employed to interpret feature impacts globally. The proposed approach demonstrated robust predictive performance across all seasons (Accuracy > 0.94). Results identified six distinct driving styles: Responsive-Agile, Risk-Negligent, Conservative-Delayed, Nocturnal-Adapted, Overreactive, and Trajectory-Unstable. Notably, individual drivers exhibit significant seasonal transitions in their styles; for instance, alarm responses tend to be slower in spring and summer but faster in autumn and winter. Furthermore, SHAP analysis revealed that initial response time, speed decay ratio, and nighttime driving intensity are critical determinants in style classification. These findings provide freight companies with actionable insights for targeted safety training and dynamic risk management.