Machine learning for activity-based models: Toward personalised preferences and mobility tastes

This study aims to develop a next-generation activity-based model by leveraging different machine learning approaches to address the limitations of current activity-based models. The goal is to improve the activity-based model which can generate more reliable individual activity schedules. Specifically, we develop an activity generation framework that can generate both continuous activity start time and duration. We also improve the predictive capability of flexible location choice by developing a novel choice-set generation approach. Furthermore, our proposed mode choice model could improve flexible mobility tastes and predictability, as well as maintain the interpretability of discrete choice models. Thus, the study enhances the capability of activity-based models, which allows transport planners to investigate the impact of transport policies on individuals and different groups based on their demographic characteristics.