Discrete choice models to evaluate smart mobility strategies with traffic microsimulation models

The study of the mode choice for urban regions has increased, with a growing set of recent publications using joint methodologies of data gathering and modeling with machine learning models. We propose a 3-stage joint methodology that covers literature review for the data gathering and modeling techniques, the modal preference study, and the modeling of the modal choice. We apply our proposed methodology to a private urban university in the north of Mexico. We show that decision tree based, machine learning models for multiclass classification, with datasets in which categorical data predominates, have a better performance than the widely applied econometric models covered in literature. We take advantage of the interpretability of decision trees to identify relevant variables that influence modal choice. We conclude that, for our studied sample, people that have access to collective modes do not choose them because they are not aware of their existence.