Data-driven choice set generation and estimation of route choice models

This paper proposes a novel combination of machine learning techniques and discrete choice models for route choice modeling. The data-driven choice set generation method identifies routes characteristics by clustering, and implicitly generates the choice set by sampling route characteristic attributes from the clusters. Important features are selected by random forests for route choice model development. With the selected features, the methodological-iterative approach is applied to specify the utility functions and to find significant explanatory variables automatically. Results show that the proposed data-driven method produces a discrete route choice model not only with strong explanatory power, but also with high prediction accuracy compared to models estimated with conventional choice set generation methods.