Forecasting of railway passenger transport demand with artificial neural network

Railway passenger planning and determining the transportation rate of the projects that are planned to be made to the required points are effective in choosing financially apropriate methods. Considering the factors affecting the passenger demand, the creation of forecasts has an important effect on making the right choices an decisions. In the study, 9 independent variables that are effective for determining the demand for railway passengers by regression analysis and artificial neural networks. In order to evaluate the performance of the models, determination coefficient and mean square eror (MSE) were taken in to consideration. With correlation analysis, the relationship between variables was examined first and the explanation rate of the independent variables for the dependent variables was found to be sufficient. As a result of the regression analysis, it was determined that population and railway line length were more significiant on the dependent variable. The appropriate network structure was determined by using Levenberg-Marquardt algorithm to train the network in analysis with artificial neural network. The effect of independent variables was analyzed with sensitivity analysis and it was determined that the most important variable was GDP. The determination coefficient within the two models are sufficient to explain the models and MSE has low values. However, it has been understood that artificial neural networks perform better in determining the number of rail passengers. Therefore, ANN creates better results in the forecast of passerger demand as it creates realistic values.