Optimal Time Series of Temporal Traffic Accidents using Artificial Neural Network, based on Spatial Information System (Case Study: Karaj - Qazvin Freeway)

The present study attempts to optimally model the daily temporal traffic accidents on Karaj-Qazvin freeway, as one of Iran's accident-prone roads between 2009-2013, using two methods of Time Series and Artificial Neural Network, based on spatial information system. Temporal patterns of road hazards are obtained from temporal traffic accident data (sample size equals to 1097) by taking into account trend and periodic components, model type and order, Autocorrelation and Partial Autocorrelation Functions. In the former method, the dependence of time domain and the order of temporal model were calculated and in the latter method, different architectures of Multilayer perceptron (MLP) (a Feedforward Artificial Neural Network) were implemented to determine the most optimum diagnostic performance. Characteristic indices, coefficient of determination and accuracy were used to evaluate the network and ARCGIS and MATLAB software were used to calculate the two models in different scenarios. The results indicate that the Artificial Neural Network model, with coefficient of determination and root mean square error 10.71, can estimate the rate of daily accidents somewhat better than the Time Series method and Partial Autocorrelation with coefficient of determination and root mean square error 14.31. It should be noted that the modeling of daily traffic accident data using Artificial Neural Network model and Partial Autocorrelation, has not yet been presented in similar studies and research.