FairUDT: Fairness-aware Uplift Decision Trees

2025-01-01

Knowledge-Based Systems

AI Summary

To address population-level unfairness in uplift modeling for causal inference—such as systematic underestimation of treatment effects for minority groups due to gender or racial biases—this paper proposes the first method that explicitly incorporates group fairness constraints into the uplift decision tree construction process. Our approach innovatively extends the splitting criterion by integrating causal-forest-inspired gain estimation with a fairness-aware regularization term, and introduces an adaptive-threshold pruning mechanism that jointly optimizes both causal identification accuracy and group fairness during tree growth and pruning. Evaluated on multiple real-world datasets, our method reduces Δ-ATE (the absolute deviation from the true average treatment effect) by 42%–68% compared to state-of-the-art baselines, while maintaining stable uplift AUC. This work achieves, for the first time, end-to-end unification of fairness constraints with interpretable uplift tree modeling.

Technology Category

Machine Learning: Causal LearningReasoning under Uncertainty: CausalityGame Theory and Economic Paradigms: Fair Division

Application Category

User Modeling, Personalization and Recommendation: Fairness-aware retrieval and rankingEconomics, Online Markets and Human Computation: Fairness, privacy, and diversity in economic environmentsSearch and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Problem

• Machine Learning Bias
• Unfairness in Data
• Minority Underestimation

Innovation

• FairUDT
• Bias Mitigation
• Transparency