Hazards-Based Duration Time Model with Priorities Considering Unobserved Heterogeneity Using Real-Time Traffic and Weather Big Data

Abstract

Traffic crash-post management is very important for transportation agencies. Delays in clearing the scene after a crash can directly increase the likelihood of a secondary crash and cause more serious traffic congestion. To optimize the management strategies for non-recurrent congestion, it is important to understand the factors that affect incident clearance times. This paper develops a model to analyze the duration time on highways using various types of datasets, including real-time data at the time of or immediately before the crash, detailed time variables, and crash type, with an accelerated failure time model. The model includes the three parametric distributions and assumed randomness, which is called unobserved heterogeneity, and can parametrically estimate the time to hazard to provide the conditional probability that the crash will be resolved. The results show that the Weibull distribution model with random parameters was suitable for both injury and non-injury crashes. Specifically, factors such as whether a truck was involved, temporal speed difference, rain, and rollover status are related to the increase in the duration time. Also, when the weighted length of the response time and detection time are applied to the duration time, the shorter the response time, the shorter the duration time for injury crashes. If there are no injuries, the faster it will be detected and help arrive at the scene. On this result, it is expected that it will be possible to develop a highly accurate clearance time prediction model with artificial intelligence techniques by using more data samples or high-resolution vehicle trajectory data.