Investigating the influence of socioeconomic factors on the relationships between road characteristics and traffic crash frequency and severity-- A hybrid structural equation modelling - artificial neural networks approach.

Traffic crashes result from complex interactions between driver, roadway, and environmental factors, which traditional methods often fail to capture. This paper investigates the influence of road, weather, and socioeconomic factors on traffic crashes, using a two-stage hybrid Structural Equation Modelling (SEM)-Artificial Neural Networks (ANN) approach to capture the complex relationships between these factors and crash intensity, a variable that jointly captures the frequency and severity of crashes. A database from Ohio collector road segments served as the case study in this novel hybrid approach, which utilized SEM to analyze the complex and moderating relationships between different factors and crash intensity. SEM revealed significant relationships between crash intensity and factors such as "Horizontal Curve," "Road" (AADT and surface width index), "Segment Length," "Speed Limit," "Vertical Curve," and "Vehicle Possession." Based on the SEM results, "Vehicle Possession" significantly moderated the relationship between "Horizontal Curve" and crash intensity. In the next step, ANN further identified key predictors, including "Segment Length," "Road," the interactions of "Vehicle Possession-Speed Limit," "Vehicle Possession-Vertical Curve," and "Age-Road." The findings highlight the advantage of the complementary application of linear and nonlinear methods in providing invaluable theoretical and methodological insights for crash data analysis.