Citation

Alrukaibi F, Alkheder S, Sayed T, Alburait A. J. Transp. Health 2021; 20: 101025.

Copyright

(Copyright © 2021, Elsevier Publishing)

DOI

10.1016/j.jth.2021.101025

PMID

unavailable

Abstract

Traffic collisions are considered a serious problem worldwide that cause severe losses. Gulf Cooperation Council (GCC) Countries have high rates of collisions, which requires urgent proactive strategies and attention to solve such problem. This study aimed to enhance traffic safety conditions and reduce collisions' severity levels at several locations in Kuwait City, Kuwait. Consequently, the study importance raised in four folds; reducing the total number of collisions on Kuwait highways, predicting the future numbers of collisions, identifying and managing risk factors contributing to collision's severity, and developing new strategies to enhance traffic safety condition in Kuwait. Three-year crash dataset from 2016 to 2018 including 4028 road collisions in Kuwait City were used to analyse the driver injury severity influence factors and to predict the future collision counts. Statistical indices were used to evaluate the mixed logit model performance, which are the MCFadden Pseudo R-Squared statistics and the two-log likelihood. Eight covariates were tested for significance in the mixed logit model estimation.

RESULTS showed that female drivers, driving during night-time, driving outside the city, and not using seatbelt produced the highest possibility of having incapacitating injuries and fatalities by 46.6% and 31.6%, respectively. Contrastingly, male drivers, driving during daytime, driving inside the city, and using seatbelt had resulted in the lowest probability of getting incapacitating injuries and fatalities by 4.7% and 0.2%, respectively. Furthermore, Bayesian hierarchical model was used to investigate the future road collision counts and to identify the collision blackspots (CBS) on Kuwait City highways. Six covariates were considered in the model. Log-linear regression model (CPM) was used to estimate the vector of means μj(t). By applying regression to mean (RTM) and predicting the trend in the mean collision rate λj (t), results showed that the average model accuracy for year 2018 was 45.76% (for a 95% confidence interval estimation using the collision dataset of previous years). It was also found that the highest number of collisions had occurred on the second ring road. Basing the results on year 2018 helped in predicting more accurate future collision counts, and in justifying collision rates. Additionally, further covariates can be added to the models for any additional recent crash data to increase the model accuracy. This work is anticipated to help Kuwaiti decision makers in designing accurate countermeasures to improve traffic safety condition.

Language: en

Keywords

Bayesian statistics; Collision blackspot prediction; Collision prediction models; Driver injury severity; Mixed logit model; Regression-to-mean