← Reference notes

Summary

Moldovan compares methods that try to make automated credit decisions less unequal. The abstract says machine learning has long been used to judge whether a loan applicant is creditworthy, and that those automated decisions can treat groups or individuals unequally. The paper evaluates 12 bias-mitigation methods on 5 fairness metrics, and also looks at accuracy and at profit for the lender. The abstract says fairness, accuracy, and profit pull against each other, and that some mitigation methods do better than others. It does not, on the abstract page, name a method a person should use to compute their own score. The abs page journal line is IEEE Access, vol. 11, pp. 59729-59743, 2023. This note follows that abstract page only. It is general information about the study, not financial advice, and not a score calculator.

When to apply

• The question is how published fair-ML methods behave on credit data, including the tradeoff with accuracy and lender profit.
• The question is how to calculate one person's credit score. This paper does not answer that.

Costs and limits

• The abstract does not name the 12 methods or the 5 metrics. This note does not invent the names and does not re-check the PDF.
• A journal version can differ from the preprint. The license checked here is the one on the arXiv abs page (CC BY 4.0), which is the source of this summary.

Evidence (detail)

• The abs page (v3, 22 Jun 2023) states the study compares 12 bias-mitigation methods across 5 fairness metrics, plus accuracy and potential profitability for financial institutions (Abstract).
• The abstract reports a tension between fairness, accuracy, and profit, and says both stronger and weaker mitigation methods showed up in the analysis (Abstract).
• Journal reference on the abs page: IEEE Access, vol. 11, pp. 59729-59743, 2023. Related DOI on the page: https://doi.org/10.1109/ACCESS.2023.3286018.

Facts

• id: F1 claim: The paper evaluates 12 bias-mitigation methods for algorithmic credit decisions on 5 fairness metrics, and also measures accuracy and potential profit for the lender. confidence: medium date: 2026-10-18 sources: [S1 Abstract] license: CC BY 4.0
• id: F2 claim: The abstract says fairness, accuracy, and profitability are in tension, and that the mitigation methods were not equally successful. confidence: medium date: 2026-10-18 sources: [S1 Abstract] license: CC BY 4.0
• id: F3 claim: The arXiv abs page shows CC BY 4.0 and a journal reference to IEEE Access, vol. 11, pp. 59729-59743, 2023. confidence: high date: 2026-10-18 sources: [S1 abs page] license: CC BY 4.0

Related

• This note is the open paper behind the library miss on calculating a credit score. It covers fairness of automated scoring, not the arithmetic of a consumer score.

Review notes

• Staging draft. Paraphrase of the arXiv abstract page checked 2026-10-18. The PDF was not re-read.
• License on the abs page: icon by-4.0.png, link text "view license", http://creativecommons.org/licenses/by/4.0/.

@misc{iil_note_00h8jkqn,
  author = {{iinventedlight Library}},
  title = {{Algorithmic decision making methods for fair credit scoring}},
  howpublished = {iinventedlight Library},
  year = {2026},
  date = {2026-10-18},
  url = {https://iinventedlight.com/library/reference/arxiv/2209-07912-fair-credit-scoring/},
  note = {Permanent ID iil:note:00h8jkqn. License: CC-BY-4.0. Summary of https://arxiv.org/abs/2209.07912.},
}

For language models and agents: start at /llms.txt. This page in Markdown: /library/reference/arxiv/2209-07912-fair-credit-scoring.md.