Predicting transport mode choice preferences in a university district with decision tree-based models

Modeling and prediction of mode choice are essential to support more sustainable and safer transportation decisions. There is plenty of literature in this decade showing that machine learning (ML) models have been effective predicting techniques, although not easily interpretable. When these techniques are used, there is a lack of connection with the data-gathering step, which is crucial to the technique selection and appropriate analysis of results. Based on a systematic literature review on mode choice studies, we present a methodology that interconnects the data-gathering process as a fundamental part of the descriptive phase when ML classification methods are used to predict mode choice preferences. The case study presented occurs in a university context whose descriptive phase shows interesting behavior patterns and highly imbalanced data in terms of mode choice. We show how decision tree methods allow us to tackle this issue in a contextualized manner and permit sensitivity analysis to test policies promoting changes in the modal split that aim for more sustainable mobility for the community of the university.