Developing Artificial Neural Networks and Highway Safety Manual Models for Predicting Accidents at Intersections in Bahrain

Abstract

Intersections are among the places where the highest number of accidents occur, thus, studying their safety and considering countermeasures to increase their safety should improve the overall safety of a traffic system. Prediction models, such as Artificial Neural Networks, have not been used for planning purposes in terms of providing countermeasures for accidents. This shortcoming forces the practitioners to employ traditional statistical methods which may be less accurate and have restricted applications. Hence, the Artificial Neural Networks models of this study were developed with the application of suggested countermeasures. Their performance was also compared with the traditional method given in the Highway Safety Manual after calibrating the procedure for local conditions. In this study, the intersections with the highest reported accidents isn the Kingdom of Bahrain were analyzed. The data was taken for the years 2013–2016, courtesy of the data provided by the Bahrain General Directorate of Traffic. Using this data, two predictive Artificial Neural Networks models were developed and used to forecast the accident number and severity in these selected intersections. Four intersections were selected to showcase the findings and to study the potential countermeasures that can be applied to reduce the occurrence of accidents. The comparison between Artificial Neural Networks and Highway Safety Manual procedures showed that Artificial Neural Networks models were more convenient to use with generic applications to different types of intersections. Moreover, they also provided higher accuracy while the Highway Safety Manual model was found to be heavily dependent upon traffic demand, which greatly affected its accuracy. The countermeasures suggested in this study were shown to reduce the accidents at the selected locations.