Explainable Machine Learning Framework for Credit Risk Assessment in Consumer Loan Portfolios

Abstract

Accurate and interpretable credit risk assessment is vital for maintaining the stability of financial systems and preventing loan default losses in consumer lending. This paper proposes an explainable machine learning framework for credit risk prediction in consumer loan portfolios, integrating advanced predictive models with post-hoc interpretability techniques. We leverage gradient boosting decision trees (GBDT) and ensemble learning methods to capture complex nonlinear relationships in borrower attributes, transactional behavior, and macroeconomic indicators. To address the black-box nature of these models, we incorporate SHapley Additive exPlanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME) to provide both global and instance-level explanations of risk predictions. We evaluate the framework using real-world anonymized loan data from a U.S.-based fintech platform, and compare its performance against traditional logistic regression and credit scoring models. Experimental results show significant improvements in prediction accuracy and stability, while providing actionable insights into risk drivers such as debt-to-income ratio, credit utilization, and employment stability. The proposed framework demonstrates the potential of combining predictive performance and interpretability in machine learning-based financial risk analysis.