A machine learning fairness analysis of wage inequality across demographic groups in the United States, using 2022 American Community Survey (ACS) data.
The author thanks Professor Yaser S. Abu-Mostafa of the California Institute of Technology for his encouragement and support.
- Education is the single strongest predictor of income (feature importance = 0.241)
- A graduate degree earns 2.18× the median income of a high school diploma
- 100% of the gender wage gap is unexplained by observable characteristics
- 102.5% of the White–Black wage gap is unexplained by observable characteristics
- Gradient Boosting achieves R²=0.213, outperforming OLS baseline
- 4 ML models: OLS, Ridge, Random Forest, Gradient Boosting
- Fairness audit: demographic parity, residual bias, equalized R² by group
- Oaxaca-Blinder decomposition of gender and racial wage gaps
- Dataset: 85,000 records from 2022 ACS PUMS (5 states)
pip install -r requirements.txt python download_acs.py python run_analysis.py
Naomi Danganan — Independent Researcher