H.W. Cho, C.-L. Lan

Pages: 3-16

Abstract
This study evaluates the practicality and efficacy of implementing the Highway Safety Manual recommended sliding window method for systemically identifying high-risk segments in the Virginia roadway network. The research proposes a homogeneous segmentation network that maintains consistency in segment characteristics, based on annual average daily traffic and safety performance function types. The sliding window method, executed in Python, was applied to the newly generated homogeneous segments. The evaluation of this method’s performance encompassed multiple aspects, including assessment of potential for safety improvement (PSI) values, segment rankings, and ranking of segments with roadway attributes. Further, the study investigated the sensitivity of window size selection to the inherent stochastic nature of crash occurrences. Specifically, smaller window sizes were more effective in identifying localized crash “hotspots,” and larger window sizes delivered a more general overview of the entire segment. The study also advises against the use of a single year’s ranking for determining high-risk PSI segments, owing to this stochastic variation. The study found that the sliding window method does not exhibit inherent bias toward two roadway attributes: segment length and median presence. The finding mitigates the existing segment length variation problem that is present in the current approach.
Keywords: network screening; sliding window; new segmentation