Explainable Ensemble Learning for Loan Approval Prediction Using XGBoost, LightGBM, and Random Forest with SHAP Analysis

Abstract

Loan approval prediction is a critical task in financial institutions, as it directly impacts risk management and decision-making processes. However, challenges such as class imbalance and lack of model interpretability often limit the effectiveness and reliability of machine learning approaches. This study proposes an explainable ensemble learning framework for predicting loan approval using Random Forest, XGBoost, and LightGBM models. To address the issue of imbalanced data, the Synthetic Minority Over-sampling Technique (SMOTE) was applied to the training dataset, ensuring a balanced class distribution and improving model performance. The models were evaluated using multiple metrics, including Accuracy, Precision, Recall, F1-score, and ROC-AUC. The results show that all models achieved strong performance, with Random Forest outperforming the others by achieving the highest accuracy of 0.959 and ROC-AUC of 0.9386. Furthermore, SHapley Additive exPlanations (SHAP) were employed to enhance model interpretability by providing both global and local explanations of feature contributions. The analysis revealed that CreditScore, EmploymentType, and Income are the most influential factors in determining loan approval decisions. In addition, SHAP dependence and interaction analysis demonstrated the presence of non-linear relationships and feature interactions, offering deeper insight into the model’s decision-making process.