Assessing traffic conflict/crash relationships with extreme value theory: Recent developments and future directions for connected and autonomous vehicle and highway safety research

With proactive safety assessment gaining significant attention in the literature, the rela-tionship between traffic conflicts (which form the underpinnings of proactive safety mea-sures) and observed crashes remains a critical research need. Such a need will grow significantly with the ongoing introduction of connected and autonomous vehicles where software and hardware improvements are likely to be determined from observed traffic conflict data as opposed to data from accumulated crashes. Extreme value theory has been applied for over two decades to study the relationship between traffic conflicts and crashes. While several advancements have been made in extreme value theory models over time, the need to continually evaluate the strengths and weaknesses of these models remains, particularly considering their likely use in improving the safety-critical elements of con-nected and autonomous vehicles. This paper seeks to comprehensively review studies on extreme value theory applications in traffic conflict/crash contexts by providing an in-depth assessment of alternate modelling methodologies and associated issues. Critical research needs relating to the further development of extreme value theory models are identified and include identifying efficient techniques for sampling extremes, determining optimal sample size, assessing and selecting appropriate traffic conflict measures, incorpo-rating covariates, accounting for unobserved heterogeneity, and addressing issues associ-ated with real-time estimations.(c) 2023 Elsevier Ltd. All rights reserved.

Automated summary

This paper comprehensively reviews studies on extreme value theory applications in the context of traffic conflicts and crashes. It highlights the need to continuously evaluate the strengths and weaknesses of these models, considering their likely use in improving the safety of connected and autonomous vehicles. The paper identifies critical research needs, including efficient techniques for sampling extremes, determining optimal sample size, selecting appropriate traffic conflict measures, incorporating covariates, accounting for unobserved heterogeneity, and addressing real-time estimation issues.