Research and Analysis of Explainable Machine Learning Methods for Credit Scoring and Loan Approval

Abstract

Financial institutions depend on credit scoring and loan approval systems to make essential decisions, as the accuracy of risk assessment directly affects asset performance, regulatory compliance, and fair treatment of consumers. Traditional statistical models remain widely used due to their simplicity and transparency, but their linear structure limits their ability to capture complex borrower behavior. Tree-based machine learning models, including Gradient Boosting Decision Trees, XGBoost, and Random Forest, offer stronger predictive performance; however, their explainability and transparency remain challenging in regulated credit settings. SHAP-based frameworks in Explainable Artificial Intelligence enable feature-level attribution that links model predictions to underlying financial characteristics. This survey reviews mainstream explainable tree-based credit scoring approaches, summarizes empirical findings, and discusses practical implementation challenges. It also outlines future directions related to regulatory requirements, fairness, drift adaptation, and scalable deployment. The results indicate that explainable tree-based scoring models can achieve strong performance while providing meaningful interpretability, although several unresolved challenges require further research.