A Road Safety Strategy for Cities Entrance Roads Based on Effective Factors of Accidents using Artificial Neural Network and Poisson Regression Models

Due to the fact that accidents at cities entrance roads have a significant role in road accidents. Accordingly, this issue has been one of the challenges of the last decade to reduce accidents and provide appropriate strategies for increasing the road safety. Thus, the objectives of the present study are first to identify and prioritize the factors affecting the accidents of cities entrance roads in 26 different types of roads in Tehran, Isfahan and Yazd provinces using artificial neural network - multilayer perceptron (ANN-MLP) and Poisson Regression (PR) models. Then, the study presents a road safety strategy model for controlling accidents regarding effective factors prioritized with various degrees of road performance. The results of the present study showed that the best model is the feed-neural network with Levenberg-Marquadt training function with 7 input variables, and 5 hidden neurons which the root mean square error (RMSE) of this model is 1.020 which includes longitudinal slope, operating speed, change the number of lanes, the percentage of heavy vehicles, degree of road performance, speed control camera, and road width. However, PR model indicated that operating speed, longitudinal slope, road width, change the number of lanes, degree of road performance, and the percentage of heavy vehicles. Further, the results of the proposed strategies to reduce the accident based on ANN-MLP, and PR models showed that longitudinal slope among all variables decreased number of accidents by 40%, and 45% for arterial roads class I, respectively.