Estimating road accidents of Turkey based on regression analysis and artificial neural network approach

Abstract

This study proposes two new analytical models and an Artificial Neural Network (ANN) model to estimate the number of accidents, injuries and fatalities in Turkey utilizing historical data between 1986 and 2005. The data between the years 1986 and 2000 were used to develop the models and the rest of data (i.e., 2001-2005) were utilized for testing the developed models. The first of the analytical models is a modified form of the Smeed accident prediction model. The second one is an adapted form of the Andreassen model to Turkey. In the model development, the number of vehicles (N), fatalities (D), injuries (I), accidents (C), and population (P) were taken as model parameters. In the ANN model, the sigmoid and linear functions were used as activation functions with feed forward-back proportion algorithm. The model results were compared against the observations and it was found that the ANN model performed better than the other two analytical models. In order to investigate the performance of the models for future estimations, a fifteen year period from 2006 to 2020 was employed. Considering the fact that Turkey is likely to enter the European Union by 2020, road safety strategies were evaluated with two possible scenarios. In the first scenario, the annual average growth rates of the population and the number of vehicles are assumed to be 1.7% and 7.5% (average growth rates between 1986 and 2005) respectively. In the second scenario, the average number of vehicles per capita is assumed to reach 0.45 which represents a three-fold increase in fifteen years. The results obtained from both scenarios reveal the suitability of the current methods for the road safety applications.