Authors: R A Shasank
Abstract: In the rapidly evolving landscape of financial technology, the imperative for model interpretability often conflicts with the pursuit of predictive accuracy. Financial institutions heavily rely on automated credit scoring models; however, the lack of transparency in conventional "black-box" approaches—such as deep neural networks and complex ensemble methods—poses significant regulatory and ethical risks. This paper introduces a hybrid credit risk assessment framework that bridges the gap between performance and interpretability. By leveraging First-Order Inductive Learners (specifically the RIPPER algorithm), the proposed model transforms raw financial data into a structured set of human-auditable domain rules. Furthermore, we implement a novel "Abstention-Driven Human Audit" layer, which identifies cases with marginal prediction confidence and redirects them for manual expert review. The experimental analysis, conducted on standard benchmark datasets, demonstrates that this architecture maintains competitive predictive power while providing a clear, logical rationale for every automated decision. The results highlight that the integration of rule-based logic not only fosters regulatory compliance but also enhances stakeholder trust in automated financial systems. This study contributes a scalable, transparent, and robust alternative for modern credit risk management.
