LAUSR

The study aimed to utilize Structural Equation Modeling (SEM) and K-means clustering for predicting real-time crash risks on freeways. The SEM was used to transform a number of correlated traffic variables into four independent latent traffic factors, and to establish the interrelationships among the traffic variables and crash risks. The segmentation analysis based on K-means clustering was then conducted to investigate the main traffic factors affecting crash risks in various traffic regimes. It was found that: (a) The measurement equations in SEM can effectively account for the correlations among traffic variables by transforming numerous correlated traffic variables into several latent traffic variables; (b) The SEM can both capture the direct and indirect effects of traffic flow variables on crash risks. This promotes a better understanding how traffic conditions affect crash risks; (c) The SEM produces more accurate estimates of crash risks than existing modeling technique. It can increase the crash prediction accuracy by an average of 7.6% compared with the commonly used logistic regression; and (d) Segmentation analysis results suggested that the traffic factors contributing to crash risks are various across different traffic regimes. The proactive crash prevention strategies for different traffic regimes were discussed based on the findings in the segmentation analysis.