期刊名称：International Journal of Statistics and Applications
印刷版ISSN：2168-5193
电子版ISSN：2168-5215
出版年度：2018
卷号：8
期号：4
页码：173-188
DOI：10.5923/j.statistics.20180804.03
语种：English
出版社：Scientific & Academic Publishing Co.
摘要：Objective: Typically, intersections that carry more traffic will have more crashes; however, these crashes might not be severe. On the other hand, low-volume intersections might have lower number of crashes; however, can be more severe than their high-volume counterparts. Since the geometric and traffic characteristics of signalized and stop-controlled intersections are different, the significant factors affecting crash severity at both intersection types will be also different. This paper identifies and compares those significant factors affecting crash severity at high- and low-volume signalized and stop-controlled intersections in Alabama using five-year crash data from 2010 to 2014. A cut-off value of 1,000 vehicles/day was used to classify intersections as high-volume vs. low-volume. Method: A random forest model was used to rank variable importance and a binary logit model was applied to identify the significant factors at both high- and low-volume signalized and stop-controlled intersections. Four discrete models (high-volume signalized, low-volume signalized, high-volume stop-controlled, and low-volume stop-controlled) were developed. Roadway, traffic, vehicle, driver, and environmental characteristics were used as independent variables in the models. Results: In all four models, crashes in rural areas showed higher severity compared to urban areas and right-turning maneuver showed relatively lesser severity. Rear-end crashes showed lower severity compared to side impacts at high- and low-volume stop-controlled and high-volume signalized intersections. Head-on crashes, driving under influence (DUI) of alcohol/drugs, and increase in driver age showed higher severity at high- and low-volume signalized intersections. Motorcycles were associated with higher severity at high- and low-volume signalized intersections, as well as high-volume stop-controlled intersections. Conclusions: Most of the factors with the highest ranking from the random forest model were found significant in the binary logit models. Strategies to alleviate crash severity at different intersections are suggested. Practical Applications: Since the left-turning vehicle maneuver showed higher severity likelihood at high-volume signalized intersections, providing enough sight distance and protected left turn phase (with no permitted phase) in busy intersections is suggested. Also, education programs should be designed disseminating the dangerous effect of DUI on crash severity while crossing signalized intersections.
