Peer-to-peer (P2P) lending platforms have transformed modern financial services by enabling direct borrower and lender interactions, but they also introduce significant credit risk challenges due to borrower heterogeneity and class imbalance. This study proposes an explainable ensemble learning framework for credit risk prediction using LendingClub data as the primary dataset and Prosper data for external validation. A comprehensive preprocessing pipeline involving missing value handling, feature encoding, mutual information-based feature selection, and SMOTE balancing was applied to improve data quality and class distribution. The proposed framework integrates XGBoost, LightGBM, and CatBoost using a weighted soft voting strategy based on ROC-AUC performance. Experimental results demonstrate that the proposed ensemble achieved superior performance with 99.77% accuracy, 99.62% F1-score, 0.9194 ROC-AUC, and 0.9941 MCC on the primary dataset, while maintaining strong generalization on the external dataset. Furthermore, SHAP and LIME explainability techniques were incorporated to provide global and local interpretations of model predictions. The findings confirm that the proposed framework delivers both high predictive capability and transparent decision support for financial risk assessment systems.

Credit Risk Prediction,Explainable Artificial Intelligence,Ensemble Learning,Peer-to-Peer Lending,SHAP,LIME