Motorcycles Accident Severity Prediction Models in Highways Using Logistic Regression and Neural Networks Approach

The study of motorcyclists’ vulnerability in roads and highways has not received adequate attention and it seems there would be a need for extensive study in this field. That is why in this paper we are about to investigate factors influencing accident severity of motorcyclists and to predict severity level of accidents due to the human, road, weather and vehicle factors as predictor factors. Motorcycle accidents data for developing models were extracted from database of Tehran Traffic Police Department for the years 2003 to 2007. Then, the data associated with traffic characteristics, such as volume and flow speed and highways geometrical characteristics, were collected from Tehran Traffic and Trans-portation Organization and Tehran Comprehensive Traffic and Transportation Studies Company, respectively. In the present paper, the severity of accidents was classified into two levels, “injury or fatality” and “property damage” and, then, to model, analyze and compare the results, they were processed through two methods; first, the mixed logistic method was used by means of SPSS software and then other models were developed by a series of artificial neural networks. The results of the two methods, i.e. the logistic and artificial neural networks methods, were compared, showing that artificial neural networks are better estimators than logistic models. Comparing the results prove the reliability of new methods based on natural evolutionary algorithms, like neural networks, when dealing with irregular objective functions.