FAIRPLAI: A Human-in-the-Loop Approach to Fair and Private Machine Learning

TL;DR

FAIRPLAI addresses the fairness–privacy–accuracy paradox in high-stakes ML by embedding human judgment into a frontier-based framework. It constructs privacy–fairness frontiers across varying privacy budgets ε and fairness constraints, paired with a bidirectional policy tuple that translates stakeholder requirements into technical constraints and back. A differentially private auditing loop is included to review explanations without compromising data security. Empirical validation on five public datasets shows that FAIRPLAI preserves strong privacy protections while reducing fairness disparities relative to automated baselines, supported by a robust translation layer that communicates trade-offs to stakeholders. The approach offers a practical, interpretable path to deploying effective, responsible, and trustworthy ML systems in domains such as finance, healthcare, education, and public policy.

Abstract

As machine learning systems move from theory to practice, they are increasingly tasked with decisions that affect healthcare access, financial opportunities, hiring, and public services. In these contexts, accuracy is only one piece of the puzzle - models must also be fair to different groups, protect individual privacy, and remain accountable to stakeholders. Achieving all three is difficult: differential privacy can unintentionally worsen disparities, fairness interventions often rely on sensitive data that privacy restricts, and automated pipelines ignore that fairness is ultimately a human and contextual judgment. We introduce FAIRPLAI (Fair and Private Learning with Active Human Influence), a practical framework that integrates human oversight into the design and deployment of machine learning systems. FAIRPLAI works in three ways: (1) it constructs privacy-fairness frontiers that make trade-offs between accuracy, privacy guarantees, and group outcomes transparent; (2) it enables interactive stakeholder input, allowing decision-makers to select fairness criteria and operating points that reflect their domain needs; and (3) it embeds a differentially private auditing loop, giving humans the ability to review explanations and edge cases without compromising individual data security. Applied to benchmark datasets, FAIRPLAI consistently preserves strong privacy protections while reducing fairness disparities relative to automated baselines. More importantly, it provides a straightforward, interpretable process for practitioners to manage competing demands of accuracy, privacy, and fairness in socially impactful applications. By embedding human judgment where it matters most, FAIRPLAI offers a pathway to machine learning systems that are effective, responsible, and trustworthy in practice. GitHub: https://github.com/Li1Davey/Fairplai

Figures (5)

• Figure 1: Fairness--Privacy--Accuracy Paradox in Machine Learning Models
• Figure 2: Demographic Parity Difference across datasets under varying privacy budgets (epsilon). Higher values indicate greater disparities in fairness as privacy constraints are relaxed.
• Figure 3: Accuracy across privacy budgets ($\epsilon$) for five datasets. Most datasets (COMPAS, Diabetes, Adult, AIDS) show improved accuracy with weaker privacy, while Student Performance peaks at $\epsilon=1.0$ before declining.
• Figure 4: Accuracy across fairness settings by dataset. Baseline achieves slightly higher accuracy for Adult, Student Performance, and COMPAS, while results are identical for AIDS and Diabetes. Higher values indicate better performance.
• Figure 5: Accuracy comparison between baseline and Fair-Private ML across five datasets. Baseline consistently achieves higher accuracy over Fair-Private ML, highlighting the need for a balanced approach.