1667080 - Clarifying Rear-end Crash Propensity with Spatial Spillover Effect: Artificial Bayes-GLM Method

Rear-end crashes often result in severe injuries. The reasons may be attributed to a bunch of factors, so clarifying crash mechanisms is crucial to develop effective countermeasures. Statistical analysis of read-end crash event data is a conventional approach in an attempt to interpret vehicle-to-vehicle interactions at microscopic level. However, the resulted judgment always involves biases due to inaccurate pre-crash data with lack of relevant causation mechanism. On the other hand, many studies found out that crashes also are highly affected by adjacent geographic unit of the crash location in terms of geometric, traffic and environment, which is termed as spatial spillover effect. To clarify rear-end crash propensity for developing safety countermeasures at a larger scale such as corridor or roadway network, it is crucial to incorporate the spillover effect into the microscopic rear-end crash analysis. This paper innovates a systematic Bayes-GLM based model to address this issue. Hotspot analysis with K-means clustering is integrated with crash causation model. This is a multi-layered approach aiming to enhance the functionality and scalability via systematically screening potential factors contributing to rear-end crashes, and then identify relationships between the spillover effect and microscopic impacting factors. The predominant benefit from such an integrated macro- and microscope approach lies in quick identification of critical areas, on top of which, a heuristic vehicle-to-vehicle interactions reflecting pre-crash behaviors is systematically connected to the spillover effect, so that a larger scale of the crash influencing factors could be integrated to enhance the rear-end crash occurrence mechanism.