Citation

Khishdari A, Khani Sanij H, Zaker Harofteh J, Dehghan Banadaki M. J. Transp. Res. (Tehran) 2021; 18(1): 35-50.

Copyright

(Copyright © 2021, Iran University of Science and technology, Transportation Research Institute)

DOI

10.22034/tri.2021.120369

PMID

unavailable

Abstract

Numerous people have died and economically damaged due to the road accidents. One of the efficient ways of reducing crashes is to predict them before happening. This paper investigated the power of artificial-neural network (ANN) model to predict crash frequencies of Naein-Ardakan road, located in Yazd, Iran. To date, there seems no research done to compare the effects of ANN training functions on prediction performance. This research aimed to determine the proper ANN training algorithm for crash frequency prediction. In this regard, four different training algorithms were investigated. The results demonstrated the outperformance of 'trainlm' algorithm. Additionally, it was found that the average daily traffic per lane and gap lengths is the most influential factors in crash occurrences, respectively. The present study can be applied to more precisely explain the effects of independent variables on crash outcomes. An in-depth explanation of the effectiveness of independent variables can assist road safety experts in making better decisions for reducing accidents.

Language: en