RECENT ADVANCES AND CHALLENGES IN AI-BASED CREDIT SCORING MODELS: A SURVEY ON FAIRNESS AND BIAS MITIGATION IN FINANCIAL

Abstract

Data-driven models introduced by Artificial Intelligence (AI) have fundamentally transformed credit scoring, making
the process more accurate, faster, and fairer. AI-based credit scoring leverages machine learning, deep learning, and hybrid models
to uncover complex patterns across diverse data sources, including alternative and behavioral data, thereby surpassing traditional
methods that rely solely on limited historical financial data. Explainable AI (XAI) adds transparency and interpretability,
addressing the long-standing “black box” issue in automated credit decisions. However, AI-powered systems still face challenges
related to algorithmic bias, data privacy, and security concerns. Recent research advocates adopting fairness-aware frameworks
and bias-mitigation techniques, such as adversarial debiasing and continuous validation, to ensure equitable credit evaluations.
Furthermore, global regulatory standards like GDPR, ECOA, and FCRA promote ethical AI practices and safeguard consumer
rights. The success stories of fintech companies such as Tala and Lenddo illustrate AI’s transformative potential to promote
financial inclusion for the unbanked and underbanked, while highlighting the need for responsible, explainable systems.
Ultimately, integrating fairness-by-design principles ensures that lending decisions remain unbiased and sustainable, marking AIbased
credit scoring as a pivotal advancement toward a transparent, inclusive, and ethically governed financial ecosystem.