Benchmark Study of Approval Disparities with Fairness Aware Classifiers in Consumer Lending Datasets

Abstract

The expansion of automated decision systems in consumer lending has introduced profound concerns regarding fairness, discrimination, and equitable access to credit. As financial institutions increasingly rely on complex machine learning models to predict default risks, the unintended consequence of disparate approval rates across demographic groups has become a pressing socio-economic issue. This paper investigates the predictability and mitigation of approval disparities through the implementation of fairness aware classifiers across benchmark consumer lending datasets. By systematically analyzing the trade-offs between predictive capability and demographic parity, the study aims to establish a comprehensive framework for evaluating algorithmic fairness in real-world credit scoring scenarios. Utilizing disparate impact metrics and equalized odds constraints, the research benchmarks several state-of-the-art fairness intervention techniques, including pre-processing methodologies, in-processing algorithms, and post-processing adjustments. The evaluation demonstrates that while fairness interventions successfully reduce approval disparities among historically marginalized groups, the degree of accuracy degradation varies significantly depending on the chosen methodology and the underlying dataset characteristics. The findings underscore the critical need for context-specific fairness applications in financial algorithms, offering actionable insights for regulators, data scientists, and practitioners in the financial technology sector. Ultimately, this benchmark study contributes to the broader discourse on ethical artificial intelligence by providing empirical evidence on the efficacy of fairness aware classifiers in fostering inclusive and equitable financial ecosystems.