CHARACTERISTICS AND ESTIMATION OF TRAFFIC ACCIDENT COUNTS USING ARTIFICIAL NEURAL NETWORK AND MULTIVARIATE ANALYSIS: A CASE STUDY IN TURKEY NORTH TRANSIT INTERURBAN

Abstract

By the World Health Organization, traffic accidents in 2013 have been identified as human-caused natural disaster. Since the parameters which cause traffic accidents are numerous, they have a complex structure and they have continuously been a source of interest for analysis by researchers. In this study, D100 (E80) State highway section (422 km) between Erzincan- Agri (Gurbulak) cities in Turkey was investigated by examining the statistical data obtained from Turkish Statistical Institute and Republic of Turkey General Directorate of Highways 12th Regional Directory. For each kilometer lane, widths, Annual Average Daily Traffic (AADT), the number of accident, injury and death, average velocity, distance (km), the number of links and junctions were determined. The methods of ANN (Artificial Neural Network), Poisson regression and multivariate regression were used to analyze above parameters. The methods used have been evaluated as per R-2, MSE, AIC criteria. It has been found for the ANN method that R-2 (0.93), MSE (0.004), in Poisson regression method R-2 (0.57), MSE (0.33), and in the multivariable regression method R-2 (0.61), MSE (0.26). The results showed that ANN models perform better than the multivariate regression in terms of traffic accidents.