The increasing number of road accidents and the nonlinear complex relationship among various factors contributing to the occurrence of an accident call for up-to-date tools enabling to fully comprehend the casual processes and predict accident occurrence in road segments so that policy maker could eva‎luate various strategies aiming at improving the vehicular safety. The machine learning approach has emerged alternative for the classic statistical approach that can discover patterns in data and use them as new input The major challenge in using machine learning models is the availability of rich, high-quality data This challenge, however, has been recently addressed by the emerging sensors and high performance computation technologies empowering real-time data collection and fast pre-processing The present research develops algorithms for the collection of input data that can be used for the prediction of vehicular crashes in Iran. The study also develops algorithms for data hygiene and pre-processing processes. It then puts forwards neural network models with four layers, including the input layer, two intermediate layers, and the output layer, to predict the number of vehicular crashes at the segment level. The algorithms are then applied to the off-road accidents in occurred in Kerman province between 1392 and 1394. Geometric, weather information, daily traffic volumes and used as input variables. The results indicate that the machine learning approach has a high level of accuracy in predicting the number of accidents on road segments

احمدرضا طالبيان , حسين حق شناس

ميثم اكبرزاده، سينا صاحبي