Predicting mode choice using boosted trees in a multi-level panel effect model

Predicting travel mode choice is a crucial aspect of transportation planning and research. It involves developing models and methodologies to anticipate the mode of transportation individuals are likely to choose for a given trip. Both discrete choice models and machine learning techniques are often used to analyze historical travel behavior data and derive patterns that can be used for prediction. These models help urban planners and policymakers make informed decisions about transportation infrastructure, public transit services, and sustainable mobility options. Both discrete choice models and machine learning models have strengths and weaknesses. In this paper, we present a method that is able to harness the strengths of advanced gradient boosted decision trees while accounting for the panel structure in the data and estimating random effects, which - in machine learning studies - are otherwise often ignored. The models are tested on a travel mode choice case study and show improved predictive performance compared to a plain gradient boosted decision trees model.