Abstract
Crash injury determinants vary by setting, yet the mechanisms influencing severity in Iran remain insufficiently understood. This study analyzes 2,217,740 police-reported crashes from Iran over four years using an integrated Classification and Regression Tree (CART) and Bayesian Network (BN) framework. CART identified key predictors, while BN models examined their interactions. Analyses distinguished rural crash severity by roadway class and crash type from urban injury severity by road-user group. Rural severity was mainly associated with vehicle type, poor lighting, and lower driver education. Urban injury severity was strongly linked to vulnerable road users, particularly unlicensed or unhelmeted motorcyclists and older pedestrians wearing dark clothing. Recommended rural interventions include average-speed enforcement, flexible cable barriers, periodic overtaking lanes, and vehicle-specific daytime speed management. Urban measures include license-verification campaigns, smart crossing signals, and digital crash recording. Despite reasonable predictive performance, the framework’s main contribution is identifying interpretable, context-specific risks to support targeted road-safety interventions.
Method in one line
Screen, then explain.
CART first screened influential variables using a 70/30 train-test split and a normalized-importance threshold above 20%. Tree-Augmented Naive Bayes models then mapped probabilistic dependencies among the selected variables and severity outcomes.
- Six rural models separated two road classes and three crash types.
- Five urban models separated motorcyclists, motorcycle passengers, drivers, occupants and pedestrians.
- The networks describe probabilistic dependencies—not causal relationships.
Interactive findings
Change the road-safety lens
Select a context to see which evidence pattern and policy direction the paper reports.
Policy direction:
This interaction summarizes reported associations and recommendations. It is not a causal calculator or a reproduction of the fitted Bayesian networks.
Access and rights
Copyright-aware discovery
This page shares the bibliographic record, abstract, and an authored summary of the findings. The version of record remains with Taylor & Francis and is accessed through the DOI.