Pougala, J., Hillel, T., and Bierlaire, M.

Choice set generation for activity-based models

Speaker: Pougala Janody

STRC 2021

September 14, 2021

Activity-based models have seen a significant increase in research focus in the past decade. Based on the fundamental assumption that travel demand is derived from the need to do activities and time and space constraints. ABM offer a more flexible and behaviorally centered alternative to traditional trip-based approaches. Econometric – or utility-based – activity-based models postulate that the process of activity generation and scheduling can be modelled as discrete choices. Individuals derive a utility from performing activities, and they schedule them as to maximize the total utility. In classical discrete choice model applications, the parameters of the utility functions are estimated by deriving their maximum likelihood estimators. As the likelihood function is defined over a full enumeration of the alternatives in the choice set, this approach is limited for activity-based applications: the set of possible activities and their spatio-temporal sequence is combinatorial and not fully observed by either the decision-maker and the modeler. While discrete choice models can be estimated over samples of alternatives an appropriate definition of such sample is as crucial as it is challenging. This paper presents a methodology to sample a choice set of full daily schedules for a given individual and a list of activities. The Metropolis-Hastings algorithm allows us to explore the space efficiently and draw both high and lower probability alternatives for consistent estimation of the parameters. The methodology is tested on a sample of individuals from the 2015 Swiss Mobility and Transport Microcensus. Results show that the proposed methodology significantly improves the calibration of econometric activity-based models.