Citation

Taamneh M, Alkheder S, Taamneh S. J. Transp. Saf. Secur. 2017; 9(2): 146-166.

Copyright

(Copyright © 2017, Southeastern Transportation Center, and Beijing Jiaotong University, Publisher Informa - Taylor and Francis Group)

DOI

10.1080/19439962.2016.1152338

PMID

unavailable

Abstract

Road traffic accidents are among the leading causes of death and injury worldwide. In Abu Dhabi, in 2014, 971 traffic accidents were recorded, which contributed to 121 fatalities and 135 severe injuries. Several factors contribute to injury severity, including driver-related factors, road-related factors, and accident-related factors. In this article, data-mining techniques were employed to establish models (classifiers) to predict the injury severity of any new accident with reasonable accuracy, based on 5,973 traffic accident records in Abu Dhabi over a 6-year period from 2008 to 2013. Additionally, the research aimed to establish a set of rules that can be used by the United Arab Emirates (UAE) Traffic Agencies to identify the main factors that contribute to accident severity. Using Waikato Environment for Knowledge Analysis (WEKA) data mining software, four well-known classification algorithms were employed to model the severity of injury. These algorithms included: Decision Tree (DT) (J48), Rule Induction (PART), Naïve Bayes (NB), and Multilayer Perceptron (MLP). The effectiveness of each method in predicting accident severity was evaluated in three different ways. First, the entire data set was used as a training set for the algorithm. Second, accuracy was evaluated using cross-validation with 10-fold. Third, to overcome the problems that resulted from the imbalanced distribution of accident severity in our data set, the data set was resampled to bias the accident severity distribution toward a uniform distribution, and then cross-validation with 10-fold was used again to evaluate the performance. Furthermore, to establish the main contributing factors for road accidents severity, rules generated by the DT J48 algorithm were further explored. The results showed that the overall accuracy of the DT J48 classifier, the PART classifier, and the MLP classifier in predicting the severity of injury resulting from traffic accidents, using 10-fold cross-validation, were similar. The NB classifier exhibited less accuracy. Additionally, the prediction accuracy of the classifiers was enhanced after resampling the training set. The results indicated that the most important factors associated with fatal severity were age, gender, nationality, year of accident, casualty status, and collision type. 18- to 30-year-olds were the most vulnerable age group to traffic accidents. There was a clear trend in accident reduction over the period of the study. Drivers were involved more frequently in traffic accidents than passengers and pedestrians. Male drivers were involved more frequently in traffic accidents than female drivers. UAE, Asian, and Arab nationalities had the highest traffic accident frequency; Gulf and other nationalities had lower traffic accident frequency. The highest number of traffic accidents occurred at right angles. Pedestrian-vehicle type collisions had the next highest number of traffic accidents, followed by rear-end collisions and sideswipe collisions.

Language: en