ML Models Enhance Complex Discrete Choice Decision-Making

Sheng Lun Christine Cao, Destenie Nock, Alex Davis· August 3, 2026 View original

Key takeaways

  • ML models can significantly enhance discrete choice modeling for policy and preference elicitation.
  • Semi-parametric and non-parametric ML models generally outperform parametric ones.
  • Model performance improves with more training data and higher choice rule determinism.
  • Twinned Neural Networks showed strong performance in a real-world energy policy case study.

Who benefits

Government/PolicyMarket ResearchRetailAutomotiveHealthcare

Summary

This study analyzes machine learning models for improving decision-making in complex discrete choice tasks, particularly for policy-based preference elicitation. It evaluates four ML models (multinomial logistic regression, generalized additive model, twinned neural network, and Gaussian process) against five behavioral choice rules, demonstrating that semi-parametric and non-parametric models generally outperform parametric ones, especially with more training data and higher choice rule determinism.

This research investigates the effectiveness of various machine learning (ML) models in enhancing decision-making for complex discrete choice tasks, a common challenge in policy-making and preference elicitation. Traditionally, these tasks rely on parametric models, but this study explores how ML can advance the field by adopting data-driven approaches and learning individual preferences. The study conducted Monte Carlo experiments to evaluate four distinct ML models—multinomial logistic regression, generalized additive model, twinned neural network, and Gaussian process—against five important behavioral choice rules. The findings consistently show that semi-parametric and non-parametric models generally surpass parametric models across different choice rules and experimental conditions. Performance significantly improves with increased training data and higher determinism in the choice rules. A real-world case study using energy policy preference data further confirmed these results, with the Twinned Neural Network (TNN) performing optimally.

Why it matters

For professionals involved in policy design, market research, or product development, this research offers insights into leveraging advanced ML techniques to more accurately understand and predict individual preferences, leading to better-informed decisions and more effective strategies.

How to implement this in your domain

  1. 1Review current methods for discrete choice modeling in policy-making, market research, or product design.
  2. 2Explore the application of semi-parametric and non-parametric ML models (e.g., Generalized Additive Models, Twinned Neural Networks, Gaussian Processes) for preference elicitation.
  3. 3Conduct pilot studies using these ML models on existing discrete choice datasets to compare their performance against traditional parametric models.
  4. 4Develop strategies for collecting richer and more diverse training data to maximize the benefits of these advanced ML approaches.
  5. 5Integrate the findings into decision-making processes to inform policy, product features, or marketing strategies based on more accurate preference predictions.

Original post by Sheng Lun Christine Cao, Destenie Nock, Alex Davis

"arXiv:2607.28854v1 Announce Type: new Abstract: Discrete choice modeling is a common tool used for preference elicitation during policy-making, but this is typically done through parametric models. Machine learning can push the boundaries of discrete choice modeling for policy-ba…"

View on X

Originally posted by Sheng Lun Christine Cao, Destenie Nock, Alex Davis on X · view source

Want to go deeper?

Turn these trends into skills with Learnijoy's hands-on AI & tech courses.

Explore courses