Uncertainty-Aware Fairness-Adaptive Classification Trees

TL;DR

A new classification tree algorithm is introduced using a novel splitting criterion that incorporates fairness adjustments into the tree-building process and effectively reduces discriminatory predictions compared to traditional classification trees, without significant loss in overall accuracy.

Abstract

In an era where artificial intelligence and machine learning algorithms increasingly impact human life, it is crucial to develop models that account for potential discrimination in their predictions. This paper tackles this problem by introducing a new classification tree algorithm using a novel splitting criterion that incorporates fairness adjustments into the tree-building process. The proposed method integrates a fairness-aware impurity measure that balances predictive accuracy with fairness across protected groups. By ensuring that each splitting node considers both the gain in classification error and the fairness, our algorithm encourages splits that mitigate discrimination. Importantly, in penalizing unfair splits, we account for the uncertainty in the fairness metric by utilizing its confidence interval instead of relying on its point estimate. Experimental results on benchmark and synthetic datasets illustrate that our method effectively reduces discriminatory predictions compared to traditional classification trees, without significant loss in overall accuracy.

Figures (4)

• Figure 1: Relationship between the $\lambda$ parameter and the two evaluation metrics, Statistical Parity and Accuracy, for the Syntethic Dataset.
• Figure 2: Data generating process for the synthetic data
• Figure A1: Decision tree generated by the CART algorithm applied to the Synthetic Dataset.
• Figure A2: Decision tree generated by the Uncertainty-Aware Adaptive Fair-CART algorithm applied to the Synthetic Dataset.

Theorems & Definitions (1)

• Definition 1