Algorithmic Fairness in Credit Scoring: An Analysis of Age-Based Disparities in Logistic Regression Error Rates
This independent quantitative research study investigates age-based disparities in machine learning credit scoring models, examining whether prediction error rates differ between applicants under 30 and applicants aged 30 and above. Using the UCI German Credit Dataset, consisting of 1,000 applicant records, the study implements and compares two logistic regression models in Python: a baseline model incorporating age as a predictor and a second model excluding age to investigate the persistence of potential proxy-based disparities. Model performance is evaluated using confusion matrices, false positive rates, false negative rates, and classification accuracy across age groups. Statistical analysis identified significant differences in prediction error rates (p = 0.0031). The findings suggest that excluding age as an explicit predictor does not necessarily eliminate demographic disparities, as correlated financial variables may retain age-related information. The study explores the implications of proxy discrimination and the limitations of protected-attribute removal as a standalone algorithmic fairness intervention. This research contributes to the broader discussion of fairness, accountability, and demographic disparities in automated financial decision-making. Methodology: Quantitative analysis, logistic regression, statistical significance testing, and comparative model evaluation. Tools: Python, NumPy, Google Colab. Dataset: UCI German Credit Dataset.
Authors
- Rachel Dunn
- Iniya Raghupathi
Institutions
- Eugene O'Neill Theater Center (US)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-14
- DOI
- https://doi.org/10.5281/zenodo.24233279
- Primary Topic
- Ethics and Social Impacts of AI
- Type
- article
- Field-Weighted Citation Impact
- 0.00