Scholarlink

<p>This study focuses on evaluating the role of loan approval rates in meeting community needs with data from the HMDA (Home Mortgage Disclosure Act), encoding strategies for discretizing continuous features through the use of varying thresholds. The computation of SAFE metrics (Sustainability (S, Sustainability means Robustness), Accuracy (A), Fairness (F), and Explainability (E) metric) involves three methods of handling continuous variables: (1) retaining continuous variables without binarization; (2) binarizing data by setting the top 1/3 of values to 1 and the rest to 0; (3) binarizing data by assigning 1 to the top 10% of values and 0 to the remainder. For SAFE metrics, binarizing continuous variables slightly boosts accuracy, but reduces RGE for continuous variables, while increasing it for binary ones. Gender fairness remains stable. Logistic regression reveals stark racial disparities in loan approvals, with Black applicants facing tougher barriers, especially for larger loans. Gender biases disproportionately impact Black males. Although higher income can improve approval rates for Black race, they still need more to qualify compared to White race. Therefore, we find that Black applicants may face stricter requirements from approval agencies. This implies the necessity for greater efforts and interventions to address disparities in loan approval processes and to foster equitable lending practices.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Evaluating credit lending fairness with binarized continuous variables in SAFE metrics

  • 期刊论文

  • 发表日期：

    2026年3月23日

查看全文
  • Lunshuai Wu

摘要

This study focuses on evaluating the role of loan approval rates in meeting community needs with data from the HMDA (Home Mortgage Disclosure Act), encoding strategies for discretizing continuous features through the use of varying thresholds. The computation of SAFE metrics (Sustainability (S, Sustainability means Robustness), Accuracy (A), Fairness (F), and Explainability (E) metric) involves three methods of handling continuous variables: (1) retaining continuous variables without binarization; (2) binarizing data by setting the top 1/3 of values to 1 and the rest to 0; (3) binarizing data by assigning 1 to the top 10% of values and 0 to the remainder. For SAFE metrics, binarizing continuous variables slightly boosts accuracy, but reduces RGE for continuous variables, while increasing it for binary ones. Gender fairness remains stable. Logistic regression reveals stark racial disparities in loan approvals, with Black applicants facing tougher barriers, especially for larger loans. Gender biases disproportionately impact Black males. Although higher income can improve approval rates for Black race, they still need more to qualify compared to White race. Therefore, we find that Black applicants may face stricter requirements from approval agencies. This implies the necessity for greater efforts and interventions to address disparities in loan approval processes and to foster equitable lending practices.