This talk explored the integration of machine learning with econometric principles, focusing on behavioral regularization in travel demand modeling. Two case studies will illustrate the need for combining behavioral regularization with advanced machine learning techniques to ensure that models are both theoretically sound and practically useful in travel decision-making contexts. The first case study examines monotonicity in travel demand forecasts from deep neural networks (DNNs). Despite the predictive strength of DNNs, predictions can fail to maintain behavioral consistency. New metrics for assessing monotonicity will be discussed, alongside a constrained optimization framework that exploits gradient regularizers. Applied to travel survey data from Chicago and London, the proposed approach improves DNN behavioral regularity while preserving predictive accuracy, with notable enhancements in smaller samples and out-of-domain scenarios. The second case study introduces a novel graph neural network (GNN) architecture designed to incorporate network effects into discrete choice models, while keeping parametric interpretability.
Read more here.