Modeling behavioral patterns of commuters and their decision-making process is crucial to develop sustainable and effective transport policies, predict and forecast the travel mode choices of a certain population with respect to changes in some attributes or components of the transportation system, and determine the different sources of heterogeneity in tastes and preferences.
Econ-ML is about developing hybrid frameworks that combine several machine learning techniques with econometric discrete choice models to better account for different aspects of unobserved heterogeneity within a population such as systematic and random taste variations in addition to market segmentation. The proposed models would abide by McFadden’s vision of an appropriate econometric choice model in order to maintain the behavioral interpretability while improving the prediction and forecasting capabilities. Moreover, this project will focus on estimating the proposed models using Bayesian Variational Inference (VI) techniques and on providing solutions to overcome the corresponding limitations of such methods. In addition, a comparison of traditional estimation techniques such as Maximum Simulated Likelihood Estimation (MSLE) and Expectation-Maximization (EM) with Bayesian Variational Inference techniques will be conducted, with the aim of providing recommendations on when each estimation method should be used. The ultimate goal is to apply the proposed framework to real-world case studies (e.g., shared mobility, biking behavior in Copenhagen, adoption of electric vehicles, etc.) and provide the authorities and operators with forecasts and recommendations for new policies that might mitigate the negative impacts of the transportation system.