Accident Severity Prediction at HRGCs Using Artificial Neural Network
Despite considerable advances in railway safety, a lot of accidents in high severity still occur in railways. Specifically, Highway Railway Grade Crossings (HRGCs) are among the most hazardous points for railway safety. Accidents not only results in casualties of rail and road users but also cause stops in rail and road services and demolish equipment. Investigation of influential factors on accidents enables decision makers to choose the countermeasures in order to alleviate accident severity.
The purpose of this study is to predict the severity of accidents and illustrate the simultaneous effect of road features, car/train driver characteristics, weather conditions and crossing specs including angle and light on the accidents severity at HRGCs.
In this study, Artificial Neural Network (ANN) based on genetic input variable selection has been used for this purpose. To overcome the problem of designing the architecture of network by a trial and error procedure, parameters of the ANN model, are calibrated by dynamically-dimensioned search (DDS) algorithm as promising and efficient global optimizer that requires no algorithm parameter tuning. Additionally, input-variable-selections, were determined automatically using a binary genetic algorithm. The dataset used in this study consists of more than 8700 accidents that occurred from 2009 to 2013 on the U.S. HRGCs due to the fact that the remaining records were not suitable for this study. Finally, this method is compared with Basic ANN results which show the significant effects due to the calibration and the input variable selection.
- Frank F. Saccomanno, Congming Ren, Liping Fu., “Collision prediction models for highway-rail grade crossings in Canada,” Transportation Research Board, 2002.
- Shou-Ren Hu, Chin-Shang Li, Chi-Kang Lee., “Investigation of key factors for collision severity at railroad grade crossings by using a Logit model,” Safety Science, p. 186–194, 2010.
- Seyed Ali Jafaria, Sepideh Jahandideh, Mina Jahandideh, Ebrahim Barzegari Asadabadi, “Prediction of road traffic death rate using neural networks optimised by genetic algorithm,” International Journal of Injury Control and Safety Promotion, pp. 37-41, 2013.
School of Railway Engineering
Iran University of Science and Technology
Narmak, Tehran, IRAN, 1684613114
Weekday: 8am to 4pm