Modeling freeway traffic conflict intervals considering unobserved heterogeneity: a latent-class Weibull AFT approach

Traffic conflict intervals provide a duration-based measure of short-term risk recurrence. Using 424 conflict intervals from the highD trajectory dataset, this study develops a latent-class accelerated failure time (AFT) framework. Six parametric distributions are compared before Weibull-based variable selection and latent-class estimation. A two-class Weibull model is selected based on information criteria, convergence, and multi-start stability, with prior class probabilities of 0.198 and 0.802. Wald tests confirm significant cross-class heterogeneity. Higher acceleration variance and traffic flow are associated with shorter intervals in both classes, with stronger effects in Class 2, whereas mean speed is significant only in Class 1. A random-parameter Weibull benchmark provides complementary evidence of continuous heterogeneity. Accounting for unobserved heterogeneity therefore improves characterization of freeway conflict recurrence.