FairAgent: Democratizing Fairness-Aware Machine Learning with LLM-Powered Agents

TL;DR

FairAgent tackles the difficulty of implementing fairness-aware ML in practice by leveraging LLM-powered automation to perform data analysis, bias detection, preprocessing, and end-to-end model training that jointly optimize accuracy and fairness. The system comprises an automated bias mitigation pipeline and a user-friendly, no-code interface, enabling practitioners to specify fairness objectives and monitor trade-offs. Extensive experiments on the Adult and Law datasets demonstrate substantial reductions in fairness disparities (DP and EO) while preserving predictive performance, across multiple LLM backends and mitigation strategies. By lowering the technical barrier and providing precise fairness control, FairAgent advances the democratization of responsible AI in real-world settings.

Abstract

Training fair and unbiased machine learning models is crucial for high-stakes applications, yet it presents significant challenges. Effective bias mitigation requires deep expertise in fairness definitions, metrics, data preprocessing, and machine learning techniques. In addition, the complex process of balancing model performance with fairness requirements while properly handling sensitive attributes makes fairness-aware model development inaccessible to many practitioners. To address these challenges, we introduce FairAgent, an LLM-powered automated system that significantly simplifies fairness-aware model development. FairAgent eliminates the need for deep technical expertise by automatically analyzing datasets for potential biases, handling data preprocessing and feature engineering, and implementing appropriate bias mitigation strategies based on user requirements. Our experiments demonstrate that FairAgent achieves significant performance improvements while significantly reducing development time and expertise requirements, making fairness-aware machine learning more accessible to practitioners.

Figures (4)

• Figure 1: Overview of FairAgent.
• Figure 2: System Walkthrough of FairAgent
• Figure 3: Fine-Grained Fairness Control at a Given Threshold for Demographic Parity and Equalized Odds
• Figure : Performance of FairAgent with Adversarial Training Configuration and Varying LLM Backends