Crash severity analysis on developing countries using random forest: a case study of Vietnam’s expressways

Road traffic accidents (RTAs) remain a pressing global issue, particularly in Southeast Asia, including Vietnam, where the severity of crash outcomes poses a major challenge. This study analyzes and models the severity of RTAs on Vietnam’s expressway network over a seven-year period, compiling a dataset of 491 crashes that include key accident characteristics, roadway features, and vehicle involvement patterns. Critical trends were identified, such as the higher occurrence of accidents on expressways with elevated speed limits, the significant role of truck- and coach-related RTAs, and the direct influence of infrastructure elements like median barriers and emergency lanes on crash severity. Using the Random Forest (RF) model, the study identified key predictors of severity—including vehicle type, collision type, speed limits, and lane count—and compared its performance with XGBoost and SVM. RF achieved the best overall predictive accuracy, while partial dependence plots (PDPs) were applied to interpret how major factors affect severity outcomes. The findings provide evidence for traffic safety policy, supporting stricter regulation of high-risk vehicles, targeted enforcement, infrastructure upgrades, and context-specific interventions based on time and environmental conditions are essential to mitigate the consequences of crashes and reduce the severity of injuries and fatalities. Future work should establish a standardized national crash database with detailed behavioral and environmental factors, explore advanced machine learning (ML) methods when larger datasets are available, and validate the model across diverse regions and road types.