Modeling travel mode and timing decisions: Comparison of artificial neural networks and copula-based joint model

The field of travel demand analysis has traditionally been dominated by statistical models. Conversely, Machine Learning (ML) techniques have been rapidly emerging in the past few years, and several studies have demonstrated instances where ML outperformed statistical models, notably in their forecasting potential. In this article, we compare the performance of discrete, continuous, and joint discrete-continuous statistical models with the performance of the neural networks (NN), recognized as a popular ML technique. Specifically, we model two critical trip-related decisions of travel mode and departure time. Overall, we find that in addition to having a much easier and faster implementation process, the NN model offers better predictions for both decision variables. Nonetheless, critiques of NN usually typecast it as a black box due to difficulty of assessing the role of explanatory variables in estimating the target variables. To tackle this issue, we further investigate the contribution of exploratory variables in two steps: (1) estimating the relative importance of each exploratory variable, and (2) conducting sensitivity analysis on the most important variables. The results indicate that beside superior prediction accuracy, the NN is capable of capturing nonlinearities in travel demand, which suggests that it can also be more accurate to capture asymmetrical and non-linear responses for policy analysis purposes.