AI Fairness and Bias in Employment Dataset

Abstract

This project evaluated bias mitigation in the regulated context of employment, where model behavior is only part of the story and measurement discipline matters just as much.

The core workflow used preprocessing-based reweighting across protected attributes and then compared the original and mitigated datasets with explicit fairness metrics, turning the work into a concrete methodology-and-evaluation case study instead of a purely conceptual ethics discussion.

What I Built

• Intersection-aware reweighting created a concrete way to test bias mitigation across protected employment attributes.
• Disparate Impact and Statistical Parity Difference provided a measurable before-and-after fairness evaluation story.

Impact

• Shows the ability to handle AI problems where governance and metric rigor matter as much as model behavior.
• Turned fairness concerns into a measurable evaluation workflow for a regulated employment context.