Learning-Based Model for Evaluating the Impact of Neighborhood Design on Travel Behavior

Abstract

This research developed an agent-based model that evaluates the impact of neighborhood design on travel behavior while accounting for habit formation, social interactions, various levels of information provision, and awareness of transport and land use system changes. The developed model employs a framework that integrates random utility maximization theory with reinforcement learning concepts to account for the bounded rationality and knowledge learning process. Moreover, the model utilizes the diffusions of innovations theory to simulate how agents propagate information across family members and co-workers. It also adds a time dimension to the modal shift process, which could be used to indicate the relative duration to reap the full benefits of proposed scenarios. The model was applied to a neighborhood in Kelowna, British Columbia, Canada, to assess the impact on travel behavior of the SMARTer growth principles. The results showed that retrofitting non-motorized networks has more impact on modal shift than retrofitting road networks. This implies that infrastructure investments related to providing more accessibility for non-motorized users may be more socially and sustainably profitable than investments in policies targeting auto users. In addition, the results revealed that land use policies led to higher modal shift to non-motorized modes compared to retrofitting the transportation network, which highlights the importance of integrating land use and transportation planning. Similarly, the results demonstrated that transportation demand management policies can provide a positive stimulus to commuters to maintain familiarity with active transportation (AT) modes, which led in the presented case study to an increase in AT modal share.