Identification of contributing factors for interchange crashes based on a quasi-induced exposure method
Xin Gu, Mohamed Abdel‐Aty, Jaeyoung Lee, Qiaojun Xiang, Yongfeng Ma
Abstract
Interchanges are important highway facilities by connecting two or more freeways/expressways; however, it has been considered more dangerous than basic segments because of drivers’ decision-making to stay or exit, weaving, variations in speeds, etc. In this study, we aim at contributing to the literature by using a quasi-induced exposure method and logistic regression modeling approach to identify contributing factors associated with the risk of causing an interchange crash. To this end, the 2014 traffic crash data, roadway and drivers’ characteristics were collected from Florida. The modeling results indicate that drivers’ age, gender, distraction, alcohol, and other factors have statistically significant effects. In addition, the finding suggests that drivers are more likely to cause crashes at cloverleaf and direct connection interchanges than at diamond interchanges. Furthermore, a support vector machine (SVM) model was applied to compare its predictability with the logistic regression model, and a sensitivity analysis was conducted. From the comparison of the areas under the receiver operating characteristic curve (AUC) of the two approaches, it shows that the SVM model outperforms the logistic regression model. It is expected that the findings would help establish effective strategies to reduce traffic crashes at interchanges by targeted education, engineering, and enforcement.