Providing model of accident prediction in urban squares using neuralnetwork (case study: Ardabil City)

Ten percent of all accidents in the study area arerelated to accidents in urban squares. The purpose of this study was todetermine the effective parameters of accidents in urban squares and to presenta model of accident prediction.

method of this research is descriptive-survey. In this study, 456accidents were studied that occurred in 26 squares of Ardabil city from 2014 to2016 and were collected the effective factors on each of those accidents.Factors were designed as a questionnaire to select the effective refining factorsand final parameters for modeling using Likert spectrum method and Delphimethod and polling of expert. Statistical sample was selected 102 persons. 40initial effective parameters of the accidents were selected. Questionnaires wereanalyzed by statistical tests, mean and differential power, which identified 16high-impact parameters for modeling. The analysis and comparison of themodeling results were done in two ways. The first method is using a linear -regression statistical model with MiniTab software 14 and the second methodusing artificial neural network model with matlab software.

The results of statistical analysis show that among the presentedregression models, the best model of accident frequency prediction is consistedof three parameters of square traffic volume, number of passages leading to thesquare, existence of a production site and trip absorption. as well as the resultsof neural network analysis show that the above model in predicting the numberof accidents, is model with four main traffic volume input parameters, thepresence of the taxi station and bus, speed bump on the main pathways, thenumber of passages leading to the intersection.