Fairness and Bias Auditing of Decision Models in Credit and Service Delivery: A Review
Gyanshauryam International Scientific Refereed Research Journal · 2023
DOI: 10.32628/gisrrj236341 ↗Kurz gefasst
Fachartikel in der Zeitschrift Gyanshauryam International Scientific Refereed Research Journal, 2023. Verfasst von Uchechi Mary-Linda Unamma Uchechi Mary-Linda Unamma und 1 weiteren Autor:innen. Die Arbeit nennt 350 Literaturangaben.
Abstract (Originalsprache)
Automated decision models increasingly govern access to credit and to public services, mediating outcomes such as loan approval, credit limits, benefit eligibility, and the prioritisation of citizens for scarce resources. Because these systems learn from historical data that encode existing social disparities, they can reproduce and amplify discrimination while presenting a veneer of neutrality. This review synthesises the multidisciplinary literature on fairness definitions and bias auditing of decision models, with an applied emphasis on credit and lending and on public service delivery. We organise the field around six themes: the competing formalisations of fairness and their mathematical tensions; the mechanisms through which bias enters the modelling pipeline; auditing methods and open-source tooling; the credit and lending domain; the public service delivery domain; and mitigation techniques spanning pre-, in-, and post-processing interventions. A central finding, established by impossibility results, is that group calibration, equalised false-positive rates, and equalised false-negative rates cannot in general be satisfied simultaneously when base rates differ across groups, so auditing necessarily involves normative choices rather than a single technical optimum. We describe a comparative table of the principal fairness metrics, their formal criteria, and the contexts in which each is defensible. We further examine documentation practices such as model cards and datasheets, auditing toolkits including AI Fairness 360, Aequitas, and Fairlearn, and the emerging legal and regulatory context surrounding disparate impact and algorithmic accountability. We conclude that credible auditing must be sociotechnical, contextual, and continuous rather than a one-off statistical certification, and we identify gaps in intersectional evaluation, causal reasoning, and post-deployment monitoring as priorities for future work.
Zitieren
APA
Unamma, U. M. U. U. M., & Ashore-Onisemo, F. A. F. (2023). Fairness and Bias Auditing of Decision Models in Credit and Service Delivery: A Review. Gyanshauryam International Scientific Refereed Research Journal. https://doi.org/10.32628/gisrrj236341
Vancouver
Unamma UMUUM, Ashore-Onisemo FAF. Fairness and Bias Auditing of Decision Models in Credit and Service Delivery: A Review. Gyanshauryam International Scientific Refereed Research Journal. 2023. doi:10.32628/gisrrj236341
BibTeX
@article{Unamma2023,
author = {Unamma, Uchechi Mary-Linda Unamma Uchechi Mary-Linda and Ashore-Onisemo, Funmilayo Ashore-Onisemo Funmilayo},
title = {Fairness and Bias Auditing of Decision Models in Credit and Service Delivery: A Review},
journal = {Gyanshauryam International Scientific Refereed Research Journal},
year = {2023},
doi = {10.32628/gisrrj236341}
}
EUVIMED