Modeling highway-rail grade crossing (HRGC) crash severity using statistical and machine learning methods.

Journal: International journal of injury control and safety promotion
Published Date:

Abstract

A principal safety issue at highway-rail grade crossings (HRGCs) is the severity of crashes. Although many studies have analyzed crash severity at HRGCs, they often rely on national datasets or a narrow set of variables, frequently overlooking region-specific factors such as roadway design, driver behavior, and local environmental conditions. However, this study contributes to the existing body of literature by providing additional insights into the factors associated with injury severity in HRGC crashes. This study aimed to model HRGC crash severity using statistical and machine learning methods, specifically Ordinal Logistic Regression (OLR) and Random Forest (RF) algorithms, to determine significant factors associated with severe injury HRGC crashes. The statistical modeling and analyses were based on five years of HRGC crash data (2017-2021) at state-maintained HRGCs in Florida. Based on the OLR statistical model, ten variables were significant at a 95% confidence interval: crashes that occurred in the morning peak hours, no lighting condition, adverse weather conditions, railway vehicle (i.e. train or train engine), driver action (i.e. disregarded signs, signals, markings as well as other contributing actions), a speed limit of greater than 45 mph, four-lane highways, driver younger than 25, female drivers, crashes that occurred at the railroad crossings, and estimated vehicle damage of more than $1,000. Results from the OLR model indicate that all significant variables increase the likelihood of an HRGC crash being more severe, except for the time of crash occurrence (morning peak), adverse weather conditions, and drivers under 25 years of age. According to the RF model, the most important (top five) factors affecting the injury severity of HRGC crashes include estimated vehicle damage, posted speed limit, type of shoulder, driver action, and crash type. Except for the type of shoulder and crash type, the RF model results are consistent with those of the OLR model. Finally, based on the model results, potential countermeasures to mitigate fatalities and injuries at HRGCs were presented.

Authors

  • Mostafa Soltaninejad
    Department of Civil and Environmental Engineering, Florida International University, Miami, FL, USA.
  • Jimoku Salum
    SRF Consulting Group, Tallahassee, FL, USA.
  • Abdallah Kinero
    Department of Civil and Environmental Engineering, Florida International University, Miami, FL, USA.
  • Priyanka Alluri
    Department of Civil and Environmental Engineering, Florida International University, Miami, FL, USA.

Keywords

No keywords available for this article.