Model building, inference and interpretation: developing discrete choice models in the age of machine learning

This chapter explores the combination of Machine Learning (ML) techniques with discrete choice modeling. We adopt Box's loop, a framework derived from George Box and collaborators' work, to facilitate iterative experimental design, data collection, model formulation, and model criticism. This framework serves as a guiding tool throughout the chapter to assist readers in making informed decisions about when and where to apply ML versus econometric models (or a combination of both). The chapter equips choice modelers with essential tools to explain and predict human choice behavior in the age of machine learning, without favoring data-driven over theory-driven approaches. Our intent is to foster a holistic approach to choice modeling, recognizing the significance of theory-driven approaches alongside data-driven methodologies. By arming choice modelers with a diverse set of tools, we aim to empower them to successfully explain and predict human choice behavior in the ever-evolving landscape of machine learning. Upon completion of this chapter, readers will possess the necessary knowledge and resources to embrace the powerful combination of ML and choice models, unlocking new avenues for understanding human decision-making processes.