A machine learning model to predict the severity of road traffic injury based on aberrant driving behaviors and driver characteristics.

OBJECTIVES: Road traffic injuries remain a leading cause of mortality and disability worldwide, especially in low- and middle-income countries. This study aimed to develop and validate a machine-learning model to predict road traffic injury severity, using aberrant driving behaviors as measured by the Manchester Driver Behavior Questionnaire, along with demographic and driving exposure variables. METHODS: We conducted a secondary analysis of a dataset including 800 drivers. Participants were categorized into three road traffic injury severity classes: no injury (n = 400), mild injury (n = 200), and severe or fatal injury (n = 200). The Boruta feature selection algorithm was applied to confirm predictor relevance. An XGBoost classification model was trained and tuned through 10-fold cross-validation. Model performance was evaluated using multiclass AUC, accuracy, F1 score, and class-specific diagnostic indices. RESULTS: The optimized XGBoost model achieved a multiclass AUC of 0.886 and an overall accuracy of 74.4%. Sensitivity was highest for predicting no injury (0.883), while specificity peaked for the severe/fatal injury class (0.910). The most influential predictors included crash history, education level, and daily driving hours. Among behavioral variables, slips and deliberate violations emerged as meaningful contributors to injury severity predictions. CONCLUSIONS: Integrating psychological constructs from the Driver Behavior Questionnaire with demographic and exposure data provides predictive accuracy in injury severity models. Our behavior-informed framework offers a practical and theoretically grounded tool for identifying high-risk drivers and informing targeted road safety interventions.