In this paper, a statistical model has been offered to predict accident number in the segments of Iranian urban roads. The model was constructed on the base of traffic, geometric, environmental data and also the accident number for one hundred sixty five segments of urban road in Zahedan city, one of the biggest cities in Iran. Producing of a general model for several urban road types for evaluation of safety performance was the final aim of the applying of the model in this area. For production of the model a new methodology have been used. The most important features of this methodology are: selecting type of model on the base of characteristics statistical of data, applying of integral functions for selecting of variables functional form, the optimization on the base of improved gradient vector method, the goodness of fit models on the base of plot of cumulative residuals, and also no using of software statistical for running of modeling processes. On the other hand, this methodology can be extended to the other engineering subjects.
The results showed that the independent variables, on the base of their character, imitate special functional form related to their data characteristics. The functional form can be liner, nonlinear or complex of them occasionally. Although the generalized linear models (GLMs) commonly use in develop accident prediction models, they are not able to model probability of nonlinear relationship between dependent variables and independent variables, and on the other hand they usually divide independent variables and study their ...
)