Xiyuan Ren considered a ubiquitous data setting where we can fit choice models at the individual level (using inverse optimization-based machine learning), and showed how this can better capture joint travel choices within an activity scheduling context. We demonstrated this with data from Shanghai, where the agent level mixed logit model (fitted individual coefficients) outperforms state of the art dynamic discrete choice models. We also show that this can better integrate with optimization models to support choice-based system design. The work is now being applied to our C2SMART Center project in collaboration with Replica to estimate constrained mode choice models across population segments from their synthetic population data for New York State to design equitable statewide mobility services.

The research was funded by NSF CMMI-1652735 and the C2SMART Center (USDOT #69A3551747124).
Paper here: https://www.sciencedirect.com/science/article/pii/S0191261522001862