Forecasting Passenger Volumes in Transit Systems Using Support Vector Machines: The Case of Istanbul
Basar Oztaysi, Seda Yanik and Cengiz Kahraman
The high demand of mobility and the growing population in the cities necessitated high capacity and environmental-friendly transportation modes such as metro. The planning of the operations of transportation systems is a challenge due to the temporal fluctuations in the demand. Accurate forecasts of the demand in transportation help to plan the resources efficiently as well as increase the service quality. The high degree of uncertainty in the metro transit demand cannot be incorporated into traditional models that have been limited with many assumptions such as linearity. When data are available, machine learning methods provide the ability to use the existing observations to learn the nature of the situation and apply the model to new observations. In this study, we conduct an empirical study to forecast the metro transit demand using a machine learning method, support vector regression. Using the data set of a metro line in city of Istanbul, we apply a SVR model and compare the accuracy of the forecasting with seasonal autoregressive integrated moving average (SARIMA), quadratic regression and linear regression.
Keywords: Support vector machine, forecasting, regression, SARIMA, transportation system