Utility Function Specification: A Grammatical Approach

Abstract

This work addresses the challenges in specifying utility functions for discrete choice models (DCMs), which are essential in understanding and forecasting travel behavior. Traditionally, utility functions are manually specified by modelers through a subjective process, based on intuition and experience. While this approach benefits from maintaining an analytical form for easy interpretation, it suffers from inconsistency and potential inaccuracies due to the lack of a systematic framework. To overcome these limitations, this study proposes a method that combines machine learning’s automation with the interpretability of analytical forms using grammar. The goal is to develop an automated approach for defining variables, transformations, and interactions in utility specification, while ensuring alignment with domain knowledge. This leads to analytically expressive and interpretable utility functions for DCMs. The proposed framework’s potential is demonstrated through a case study.