A Grammar-Based Framework for Utility Function Specification: Validation with Synthetic Data

Abstract

This study introduces a grammar-based framework for utility function specification in discrete choice models (DCMs), combining interpretability and domain knowledge with data-driven automation. Using Grammatical Evolution (GE), the framework systematically explores variables, transformations, and interactions while enforcing theoretical constraints. Tested on synthetic data generated from a predefined “true” model, the approach demonstrates its ability to recover accurate utility specifications and prioritize key variables like travel time and cost. By balancing complexity and simplicity through parsimony penalties, this framework offers a robust, interpretable alternative to traditional trial-and-error methods, enhancing consistency and precision in DCM applications.