Citation

Wu W, Jiang S, Liu R, Jin W, Ma C. Transportmetrica A: Transp. Sci. 2020; ePub(ePub): ePub.

Copyright

(Copyright © 2020, Informa - Taylor and Francis Group)

DOI

10.1080/23249935.2020.1711543

PMID

unavailable

Abstract

This paper explores the joint effect of economic development, demographic characteristics and the road network on regional road safety. Although extensive efforts have been undertaken to model and predict the safety effects of different influential factors using statistical regression or machine learning models, little evidence is provided on the relative importance of explanatory variables by accounting for their mutual interactions and non-linear effects on traffic accidents. We present an innovative gradient boosting decision tree (GBDT) model to explore the joint effects of these comprehensive factors on four traffic accident indicators (i.e., the number of traffic accidents, injuries, deaths, and the economic loss). A total of 27 elaborated influential factors associated with the economic, demographic and road network conditions in Zhongshan, China for the period of 2000-2016 are collected. The results show that, compared to other traditional machine learning methods, the GBDT not only presents a higher prediction accuracy, but can also better handle the multicollinearity between the explanatory variables; more importantly, it can rank the influential factors on traffic accident prediction. The results also show that there are both similarities and differences in the key influential factors for the four traffic accident indicators. In particular, we also investigate the partial effects of the key influential factors. Based on the key findings, we highlight the practical insights for planning practice.

Language: en

Keywords

Demographics; Gradient boosting decision tree; Partial effect; Relative importance; Socio-economic; Traffic accidents