Prediction of Roadway Accident Frequencies: Count Regressions versus Machine Learning Models

Prediction of accident frequency based on traffic and roadway characteristics has been a very significant tool in the field of traffic management. The accident frequencies on 185 roadway segments of the City of Mashhad, Iran for the year 2007 were used to develop accident prediction models. Negative Binomial Regression, Zero Inflated Negative Binomial Regression, Support Vector Machine and Back-Propagation Neural Network models were used to fit the accident data. Both fitting and predicting abilities of the models were evaluated through computing errors values. Results show the NBR model, because of its low prediction and fitting error values, is the most effective model to predict the number of accidents. Although, the BPNN model has high fitting capability, but it does not have the prediction ability of the NBR model. Furthermore, the NBR is easy to develop and interpret the role of effective variables, in comparison with the machine learning models which have a black-box form.Marginal effect values for the NBR and ZINBR models and sensitivity analysis of the SVM and BPNN models reveal that volume to capacity ratio (V/C), Vehicle-Kilometers Travelled (VKT) and roadway width are the most significant variables. Increase in V/C and roadway width will decrease the number of accidents, however, increase in VKT and permission of parking on the right lane of the roadway can increase the crash frequency.