This paper presents a two-step methodology, involving application of nonparametric (Decision Tree - DT) and parametric models (Multiple Linear Regression - MLR) to forecast modal choice. The DT application allows finding relationships between socioeconomic variables and modal choice and discretizing the numerical and categorical variables for the construction of linear models in the later stage. The data used are from the Origin-Destination Survey carried out in 2007/2008 in the city of São Carlos (SP). The accuracy of the nonparametric model is around 70%, with high association between estimated and observed categorical values of the travel mode. After discretizing all independent variables using DT, stepwise linear models, with a good accuracy for the three categories of travel mode, were obtained. Moreover, the validation of linear models with 30% of remaining sample showed low average of errors and variance of residuals. Finally, the proposed and reasonable method could be a good alternative to traditional approaches.
modal choice; Decision Tree; Multiple Linear Regression
Thumbnail
Thumbnail
