In-depth investigation of contributing factors of fatal/severe-injury crashes at highway merging areas using machine learning classification methods
Nischal Bhattarai, Ciyun Lin, Yibin Zhang, Hongchao Liu
Abstract
Highway on-ramp merging locations are vulnerable to traffic collisions inflicting fatal or serious injuries to drivers. Although numerous studies have uncovered the major contributing factors to crashes at on-ramp merging areas, none of these studies have focused on fatal/severe-injury crashes. This paper aims to provide an in-depth and systematic investigation on critical contributing factors of the high-severity crashes at highway merging areas. As part of the analysis, support vector machines (SVM) and random forest (RF) models were developed for a 10-year data set of crashes at more than 250 merging locations in Texas, United States, using 23 different crash attributes describing each incident to predict high-severity crashes. A sensitivity analysis was conducted to quantify the marginal effects of each contributing factor. The results indicate that there is an increased likelihood of fatal/severe-injury crashes when the number of highway lanes is high, and the number of lanes on the frontage roads/connector roads is low (<4). Likewise, presence of heavy vehicles seems to affect the occurrence of fatal injury crashes at merging areas. Additionally, longer ramp lengths, presence of auxiliary lanes, and the proximity of exit ramps are found to increase the likelihood of high severity crashes. These findings, either new or consistent with previous studies are helpful in enriching the literature of on-ramp related highway safety studies. • The proposed model can classify fatal/serious-injury crashes with higher accuracy at highway merging locations. • The rank of important variables influencing fatal/serious-injury crashes is identified and qualified. • The machine learning classification methods is used to examine the contribution of fatal/high-severity crashes.