Exploring freeway crash duration by the latent class accelerated hazard model with heteroskedasticity effect

Managing traffic crashes on freeways presents a substantial challenge due to their unpredictable and nonrecurrent nature, often resulting in prolonged delays in traffic. To investigate the factors influencing freeway traffic crash duration in Taiwan, this study analyzed data encompassing 47,497 instances of freeway crashes between 2018 and 2021 to mitigate these detrimental effects. Additionally, this study introduced a novel approach, the generalized log-gamma-based latent class-accelerated hazard model with heteroskedasticity effects (HLCAH), to explore these factors. The HLCAH model categorizes crashes into three types based on crash variables: “medium duration with a long initial period,” “long duration with a short initial period,” and “short duration with a medium initial period.” Furthermore, to account for heteroskedasticity effects, the model parameterizes the probability distribution (location, scale, and shape parameters) with exogenous crash covariates. This approach not only contributes to a deeper understanding of freeway traffic crash duration but also has significant implications for enhancing freeway rescue efficiency. The model findings underscore the importance of incorporating latent classes and parameterizing distribution parameters. This research concludes with actionable insights aimed at reducing crash duration.