Consensus machine learning for interpretable road-user fatality risk analysis: Evidence from the Canadian National Collision Database

OBJECTIVES: Fatal road traffic collisions account for a small proportion of reported crashes but impose a disproportionate societal and public health burden. This study evaluates interpretable machine-learning models and proposes a consensus machine learning (CML) framework to identify robust predictors associated with fatal collision risk using Canadian collision data. METHODS: Person-level records from the Canadian National Collision Database (NCDB) covering 2014-2021 were used to classify fatal and nonfatal collision outcomes. Balanced random forest, extreme gradient boosting (XGBoost), and neural network models were developed using a common preprocessing framework. Model performance was evaluated using receiver operating characteristic area under the curve (ROC-AUC) and precision-recall area under the curve (PR-AUC). Permutation-based feature-importance rankings were subsequently aggregated using the proposed CML framework to identify predictors that remained consistently important across models and reporting years. RESULTS: XGBoost achieved the strongest predictive performance (ROC-AUC = 0.932; PR-AUC = 0.714), followed by balanced random forest (ROC-AUC = 0.898; PR-AUC = 0.585) and the neural network (ROC-AUC = 0.886; PR-AUC = 0.587). Across the three machine-learning models, collision configuration, traffic control, collision hour, and the number of vehicles involved consistently ranked among the most influential predictors associated with fatal collision risk. Roadway configuration, roadway alignment, road-surface condition, weather conditions, vehicle characteristics, and safety-device use also contributed to prediction, although their relative importance varied among models. The CML framework further identified the leading predictors of fatal collisions: collision configuration, road-surface conditions, time of collision, number of vehicles involved, roadway configuration, and weather conditions. CONCLUSIONS: The proposed CML framework consistently identified important predictors across multiple machine-learning algorithms and reporting years, providing a robust and interpretable approach to understanding fatal collision risk. Two of the defined categories, crash characteristics and traffic and environmental conditions, were identified as key feature sets for predicting collision fatalities. These findings provide evidence to support engineering, policy, enforcement, and education initiatives aimed at reducing fatal collisions while highlighting the value of consensus-based machine learning as a complementary decision-support tool for transportation safety planning.