Comparison of regression methods and artificial intelligence techniques in accident analysis

Estimating the number of traffic accidents is one of the important studies that guide the development of transportation policies. At the macro level, there are many variables that affect the number of road accidents. In this study, population, number of vehicles, number of drivers, vehicle-kilometers which are the macro variables, are considered as independent variables. The number of accidents, the number of deaths, and the number of injureds, which are dependent variable,s are estimated by using these macro variables. The data used in this study are from the years 1988-2016 of the US states. In the study, the prediction model generated by multivariate regression methods is compared with logarithmic regression, artificial neural networks (ANN) and Fuzzy Logic (FL). Linear, interaction, quadratic and purequadratic types of multivariate linear regression model are applied. Comparisons between the models are made with the correlation coefficient (R), the mean of the percentage errors (MPE) and the mean square of the errors (MSE). As a result of the analyzes conducted for the estimation of the number of accidents, the R value of the linear regression model is 0.8577, the MPE value is 4.8027% and the MSE value is 14501376454,2255. Interaction regression model R value is 0.9692, MPE value is 2.0864%, MSE value is calculated as 3326475148,8991. Purequadratic regression model R value is 0.9672, MPE value is 2.1795%, MSE value is calculated as 3540099887,9246. The R value of the quadratic regression model is 0.9889, the MPE value is 1.4379% and the MSE value is 1214042257,7187. R value obtained as a result of logarithmic regression model is calculated as 0.8502, MPE value is 4,8007% and MSE value is calculated as 15243601485,0837. R value obtained from ANN analysis is calculated as 0.9816, MPE value is calculated as 1.6463% and MSE value is calculated as 2152002423,0350. The R value obtained by the FL analysis is 0.8869, the MPE value is 5.7766% and the MSE value is calculated as 16956785090.7348. As a result of the analyzes conducted for estimating the number of injured, R value of linear regression model is calculated as 0.9390, MPE value is 4.5540% and MSE value is calculated as 21634202561,6271. The R value of the interaction regression model is 0.9934, the MPE value is 1.4955%, the MSE value is calculated as 2420035918,9591. Purequadratic regression model R value is 0.9925, MPE value is 1.6821%, MSE value is calculated as 5340363074,3138. The R value of the Quadratic regression model is calculated as 0.9951, the MPE value is 0.6148% and the MSE value is 600590065,7349. R value obtained as a result of logarithmic regression model is calculated as 0.9481, MPE value is calculated as 4,1322% and MSE value is calculated as 18549631727,8743. The R value obtained from ANN analysis is calculated as 0.9860, MPE value is calculated as 2.4522% and MSE value is calculated as 6268315631,2724. The R value obtained by FL analysis is calculated as 0.9518, MPE value is 5.0739% and MSE value is calculated as 26260831083,4280. As a result of the analyzes made for the estimation of the number of deaths, the R value of the linear regression model was calculated as 0.9971, the MPE value was 0.66600% and the MSE value was calculated as 99919.4830. Interaction regression model R value is 0.9994, MPE value is 0.2926%, MSE value is calculated as 19416,1569. Purequadratic regression model R value is 0.9993, MPE value is 0.3437%, MSE value is calculated as 25040,2321. The R value of the quadratic regression model is 0.9996, the MPE value is 0.2071%, and the MSE value is 13087.7865. R value obtained as a result of logarithmic regression model is calculated as 0.9972, MPE value is 0.6375% and MSE value is calculated as 96023,6160. R value obtained from ANN analysis is calculated as 0.9972, MPE value is calculated as 0.3949% and MSE value is calculated as 98718,9991. The R value obtained by the FL analysis is 0.9437, the MPE value is 4.4421% and the MSE value is 4737707,4473. As a result of the analyzes, the model which gives the best results in the estimation of accident, injured and death numbers was found as quadratic regression model.