Predicting risky driving behaviour in real-time is crucial for road safety, allowing for early intervention to prevent crashes. Using the Safety Tolerance Zone (STZ) to classify driving performance into three levels (i.e. normal, dangerous and avoidable accident), The aim of this study was to identify dangerous behaviour and explore the key factors influencing each level. To achieve this objective, data from a driving simulator experiment were exploited and four classification models: Ridge Classifier (RC), Support Vector Machines (SVM), Random Forests (RF) and eXtreme Gradient Boosting (XGBoost) were developed. To better understand the contribution of individual features, SHAP analysis was conducted. Through the systematic feature selection process, the most relevant variables were identified and organized into two distinct Groups, A and B. Results revealed that RF and XGBoost models consistently achieved the best performance, outperforming the other techniques across all three safety levels, reaching 95% in prediction accuracy. It was also demonstrated that time to collision was the most influential factor in Group A, whereas speed emerged as the dominant predictor in Group B, with higher values strongly associated with risky behaviour. This research highlights the risk indicators, offering valuable insights for shaping safety interventions and ultimately enhancing road safety.
| ID | pc696 |
| Presentation | |
| Full Text | |
| Tags |
