%0 Journal Article %T Road Accidents Prediction with Multilayer Perceptron MLP modelling Case Study: Roads of Qazvin, Zanjan and Hamadan %J Journal of Civil Engineering and Materials Application %I PenPub %Z 2676-332X %A Shamsashtiany, Reza %A Ameri, Mahmoud %D 2018 %\ 12/01/2018 %V 2 %N 4 %P 181-192 %! Road Accidents Prediction with Multilayer Perceptron MLP modelling Case Study: Roads of Qazvin, Zanjan and Hamadan %K Safety improvement %K Accident Modeling %K Multilayer Perceptron MLP %R 10.22034/jcema.2018.91998 %X Demand growth this has increased the incidence of road accidents and the resulting casualties, including injuries and deaths. In this study, six of the rural two lane roads were selected as the study area and crash data was collected in the roads for 2013-2016 years. In this study, multi-la ered perceptron model was used for modeling crashes for different roads. The purpose of the multi-layered perceptron model training is to find the optimal value of weights and biases in such a way as to minimize network error. With this view, multi-layered perceptron modeling is an optimization issue with a number of specific parameters. Based on the collected data, the studied roads included Hamadan to Avaj, Hamedan to Qorveh, Hamadan to Malayer and Hamadan to Bijar in the area of the protection of Hamadan province, as well as Abhar to Qeydar in the area of protection of Zanjan province and the old road to Abeek to Qazvin, in Qazvin province. An appropriate model for the roads of Qazvin, Zanjan and Hamadan was architectu e. Approximately good results were obtained from the network. The value of the r2 statistic that was calculated was 0.83. The value of the MSE parameter equals to 0.59, which indicates the accuracy of the results in the training phase. For the roads of the Qazvin region, the value of r2 was 0.94. The value of the MSE parameter was also 0.33, which was very good, and showed the accuracy of the results in the training phase. %U https://www.jcema.com/article_91998_1934a71b4d38f4a1815c26445d4ce688.pdf