Behavioral and Traffic-Related Determinants of Road Crash Severity: Supporting the Safety Dimension of Sustainable Transportation Systems

Abstract

Road crashes undermine the sustainability of transportation systems by generating substantial human, social, and economic costs. Identifying the factors associated with severe crash outcomes is therefore essential for supporting safer and more sustainable mobility, in accordance with Vision Zero and Sustainable Development Goals 3.6 and 11.2. This study investigates the influence of behavioral, environmental, and infrastructural factors on road crash severity through a comparative evaluation of three modeling approaches: Ordinal Logistic Regression (OLR), eXtreme Gradient Boosting (XGBoost), and Categorical Boosting (CatBoost). The analysis was based on 5194 police-reported crash records from Cyprus covering the period 2015–2024. The dataset included crash severity outcomes together with traffic violations and roadway and environmental conditions. The results identified speeding, substance use, and improper turning maneuvers as influential predictors of crash severity. Among the evaluated models, CatBoost achieved the strongest overall predictive performance, obtaining the highest accuracy (0.52), weighted area under the curve (AUC) (0.72), weighted precision (0.50), and weighted recall (0.52), while CatBoost and XGBoost tied for the highest weighted F1-score (0.48). CatBoost’s overall performance reflects its ability to efficiently process categorical variables and capture nonlinear relationships without extensive preprocessing. The findings highlight the dominant role of driver behavior in determining crash severity and demonstrate how the comparative modeling framework can serve as an analytical decision-support tool for identifying and monitoring crash-severity risk factors, prioritizing targeted interventions, and supporting progress towards safer and more sustainable transportation systems.