Despite the increasing use of structural equation modeling (SEM) and machine learning (ML) techniques in traffic safety analysis, few studies have integrated these approaches to examine how socioeconomic factors moderate the relationships between crash severity and road/environmental characteristics. This study addresses that gap by applying a two-stage hybrid modeling framework that combines partial least squares structural equation modeling (PLS–SEM) with ML techniques to analyze the propensity of road segments to experience severe crashes, considering the moderating effects. A data set of lane departure crashes on Ohio collector roads and associated roadway, weather, and socioeconomic data was used as a case study. In the first stage, PLS–SEM was used to model latent constructs and investigate the moderation effects of socioeconomic variables. The second stage applied four ML methods: (1) random forest; (2) support vector machines; (3) artificial neural networks; and (4) eXtreme Gradient Boosting, to improve predictive performance and identify nonlinear relationships. The results indicate that the SEM–RF combination outperformed the others, achieving higher scores in macro average F1-score, precision and recall. Subsequent analysis with the Shapley additive explanation algorithm (SHAP) revealed that some moderators were among the top contributors to crash severity, insights not captured by SEM alone. By revealing how socioeconomic factors interact with roadway and environmental factors associated with severe crash outcomes in nonlinear ways, this hybrid approach enables tailored, data-driven safety interventions. The findings can support developing targeted countermeasures that reflect the unique socioeconomic conditions of each neighborhood, advancing more equitable and effective traffic safety strategies.
Jafari et al. (Tue,) studied this question.