Towards a learning-based mode shift model: a conceptual framework

Abstract

Random Utility Maximization (RUM)-based mode choice models are criticized for their poor characterization of several behavioral aspects including habit and inertia in mode choice. As a result, traditional models may end up with a misleading estimation of modal shift in cases where strong habits towards a specific mode exist. In addition, given the cross-sectional nature of traditional choice models, it is hard to address individuals' sensitivity to various policy scenarios and to determine the period of time required to reap the benefits of the proposed policies. In order to contribute to this critical issue, this paper presents a review of existing literature on mode choice modeling focusing on capturing behavioral aspects related to habit and inertia formation. The paper then proposes a novel conceptual framework for a microsimulation learning-based mode shift model that compiles both reinforcement learning techniques with random utility maximization concepts. The proposed approach utilizes various elements of reinforcement learning methods to simultaneously integrate habit formation, level of information provision and awareness limitations. A numerical example is provided to illustrate the differences between traditional and learning-based modeling frameworks. The simulation results are in line with the prior expectations which can be considered a first step towards the model's credibility.