Estimation Model of Two-Lane Rural Roads Safety Index According to Characteristics of the Road and Drivers’ Behavior

Vehicle crashes are amongst the major causes of mortality and results in losses of lives and properties. A large number of the vehicle crashes occur on rural roads. Accidents become more noteworthy in two-lane roads due to going and coming traffic. Therefore, prediction of crashes and their causes are considerably important to reduce the number and severity of the accidents. The safety index is a suitable quantity for determination of road safety degree. It informs us to study the number of accidents in a specific road and time. In this study, safety index of two-lane rural roads is predicted by Artificial Neural Network (ANN), Radial Basis Function Neural Networks (RBFNN) and Adaptive Neuro-Fuzzy Inference System (ANFIS) algorithms using MATLAB software. The number of causes which ends to an accident is related to some parameters. We chose seven new parameters as inputs to the ANN, RBFNN and ANFIS methods that are geometric and statistical values of the roads and one output variable that is the safety index of segments of two-lane rural roads. 5 roads in Ilam Province, Iran, were selected for the case study to train, validate and test the proposed estimation models. Finally, the results show that, it is possible to predict the safety index of two-lane rural roads with a high correlation coefficient and a low mean square error (MSE) in relation to real values. The ANN method has a higher correlation coefficient and lower MSE in comparison to RBFNN and ANFIS methods. The achieved correlation coefficient and MSE for validation of the ANN approach are 0.94 and 0.0086 respectively, and correlation coefficient of 0.845 and MSE of 0.019 for all data.