EXPLAINABLE AI FOR FAIR AND ACCOUNTABLE LOAN APPROVALS

Abstract

Improvements in computing power and optimization algorithms have revolutionized sectors like lending, hastening the use of AI-driven automated decision-making. Machine learning (ML) applications automate crucial choices like loan approvals; nonetheless, consumers often lack faith in these systems due to the models' complexity and opaqueness. This work presents an explainable AI system for loan underwriting based on the belief-rule-base (BRB) paradigm to address this issue. Incorporating heuristic and factual concepts into a hierarchical framework, this method integrates supervised learning with human expertise. It not only finds a happy medium between precision and clarity, but it also gives reasons (in writing) for decisions (like a loan denial). This paper uses a mortgage underwriting business case to show how the BRB system makes decision-making more transparent, intelligible, and trustworthy by drawing attention to the role of antecedent attributes and active rules.