Deng Pan

Deng Pan

Postdoctoral Researcher at the University of Notre Dame specializing in Responsible AI. My research integrates interpretability, safety, fairness and robustness in both traditional machine learning and modern large language models, aiming to build transparent and accountable AI systems that advance responsible and trustworthy AI.

Automatic Moderator Discovery via SHAP Interaction Values

Published:

Abstract: Identifying moderator variables—factors that influence the relationship between predictors and outcomes—is crucial for understanding complex phenomena in social sciences, healthcare, and many other domains. Traditional approaches to moderator discovery rely heavily on domain expertise and hypothesis-driven analysis. We propose an automated framework for discovering moderator variables using SHAP (SHapley Additive exPlanations) interaction values. Our method leverages the theoretical foundations of cooperative game theory to systematically identify and quantify interaction effects between variables, enabling data-driven discovery of meaningful moderators without requiring extensive prior knowledge. Experiments on diverse datasets demonstrate that our approach effectively uncovers both known and novel moderator relationships, providing interpretable insights into complex multivariate systems.

Status: Under review at the International Conference on Learning Representations (ICLR 2026)

Recommended citation: Pan, D., Peng, J., Chawla, N. (2025). "Automatic Moderator Discovery via SHAP Interaction Values." Under review at the International Conference on Learning Representations (ICLR 2026).