Object-Oriented Bayesian Networks for Activity-Based Model with Deep Generative Graph Averaging

This study focused on enhancing the use of Bayesian networks (BN) in activity-based models (ABM) for transportation by integrating domain-specific knowledge and stabilizing structure learning. An object-oriented Bayesian network (OOBN) and a novel graph averaging method were used to construct a BN graph that captures an individual's decision-making in a data-oriented manner. In these methods, a deep generative model was used to improve explanatory power and learning stability. The effectiveness of this approach was validated through numerical experiments, which demonstrated the ability of the proposed method to rapidly and consistently generate plausible BN graph structures that fit the observed data better than existing BN-based ABMs did.