Incorporating Domain Knowledge in Deep Neural Networks for Discrete Choice Models

TL;DR

This paper introduces a framework that combines domain knowledge with machine learning to create interpretable discrete choice models, enhancing flexibility while maintaining behavioral interpretability.

Contribution

It proposes a novel method to incorporate domain knowledge into ML-based discrete choice models using constraints and pseudo data, improving interpretability and reducing model sensitivity.

Findings

Framework effectively integrates domain knowledge into ML models.

Enhanced interpretability of discrete choice models.

Case study demonstrates practical applicability.

Abstract

Discrete choice models (DCM) are widely employed in travel demand analysis as a powerful theoretical econometric framework for understanding and predicting choice behaviors. DCMs are formed as random utility models (RUM), with their key advantage of interpretability. However, a core requirement for the estimation of these models is a priori specification of the associated utility functions, making them sensitive to modelers' subjective beliefs. Recently, machine learning (ML) approaches have emerged as a promising avenue for learning unobserved non-linear relationships in DCMs. However, ML models are considered "black box" and may not correspond with expected relationships. This paper proposes a framework that expands the potential of data-driven approaches for DCM by supporting the development of interpretable models that incorporate domain knowledge and prior beliefs through constraints. The proposed framework includes pseudo data samples that represent required relationships and a loss function that measures their fulfillment, along with observed data, for model training. The developed framework aims to improve model interpretability by combining ML's specification flexibility with econometrics and interpretable behavioral analysis. A case study demonstrates the potential of this framework for discrete choice analysis.