Modeling Crashes Frequency at Urban Three Leg Intersections

One of the intersection safety analysis methods is creating crash prediction models using the negative binomial and Poisson regression models. However, the use of negative binomial regression is preferable due to its more data dispersion. But, the research evaluates both models at first for modeling urban intersections crashes. After testing the models by criteria of Akaike information, deviation and log-likelihood, the negative binomial regression model showed a better fit and was chosen as the better model. For modeling, the data of a 5-year period of crashes occurred at the three leg intersections of Boroujerd city has been used. The input variables of the model were selected after a significance evaluation test. These variables included the number of passing lanes, the number of right turn with traffic island lanes and the skew angle. After selecting the variables, the accuracy of the model was also studied. To validate the model, the criteria of R2, root mean square error, mean absolute error, and mean absolute percentage error were used. After evaluating the model accuracy in the prediction of accidents, it was found that the model provides acceptable results for evaluating crashes, but the results are not very accurate for predicting them (R2 = 0.52). The R2 value for the model shows that the predictions will not be precise enough, but reasonable predictions can still be provided.