Enhancing Fairness and Performance in Machine Learning Models: A Multi-Task Learning Approach with Monte-Carlo Dropout and Pareto Optimality

TL;DR

This work addresses bias in machine learning arising from data and design choices by introducing a bias-mitigation framework that combines multi-task learning (MTL) with Monte Carlo Dropout to quantify uncertainty in protected-label predictions and uses Pareto-optimality to balance fairness and performance. By training an MTL model and sampling predictions via MC Dropout, the approach identifies uncertain regions linked to protected attributes and guides bias mitigation through a Pareto-front of models. The method is validated on ADULT, MIMIC-III, and SNAPSHOT, showing substantial improvements in fairness metrics such as the Disparate Impact Ratio (DIR) with only modest or negligible losses in accuracy, and it enhances explainability through saliency maps that reveal shifted feature importance away from protected attributes. The results demonstrate transferability across domains and offer a practical, tunable framework for deploying fair and robust models in real-world settings, with explicit Pareto-front guidance for domain-specific priorities.

Abstract

Bias originates from both data and algorithmic design, often exacerbated by traditional fairness methods that fail to address the subtle impacts of protected attributes. This study introduces an approach to mitigate bias in machine learning by leveraging model uncertainty. Our approach utilizes a multi-task learning (MTL) framework combined with Monte Carlo (MC) Dropout to assess and mitigate uncertainty in predictions related to protected labels. By incorporating MC Dropout, our framework quantifies prediction uncertainty, which is crucial in areas with vague decision boundaries, thereby enhancing model fairness. Our methodology integrates multi-objective learning through pareto-optimality to balance fairness and performance across various applications. We demonstrate the effectiveness and transferability of our approach across multiple datasets and enhance model explainability through saliency maps to interpret how input features influence predictions, thereby enhancing the interpretability of machine learning models in practical applications.

Figures (24)

• Figure 1: Proposed Method Structure
• Figure 2: ADULT: Pareto Front Plot for Age, Race, and Sex Labels
• Figure 3: ADULT: DIR and Accuracy Scores of Models
• Figure 4: ADULT: Feature Importance Comparison Between Protected and Target Labels
• Figure 5: MIMIC-III: Pareto Front Plot for AUROC vs DI Ratio Score