Predicting Default Behavior in Peer-to-Peer Lending Using Gradient Boosting and SHAP Explainability

The growth of Peer-to-Peer (P2P) lending platforms has expanded financial inclusion but simultaneously increased the risk of borrower default due to asymmetric information and limited credit history. This study aims to develop an accurate and interpretable credit risk prediction model for P2P lending using the Gradient Boosting algorithm combined with SHapley Additive exPlanations (SHAP). The dataset, consisting of 48,920 borrower records from 2018 to 2023, underwent comprehensive preprocessing including missing value imputation, normalization, and outlier capping. Model optimization was conducted through Grid Search Cross-Validation, while SHAP analysis was applied to evaluate feature-level interpretability. Experimental results show that the Gradient Boosting model achieved 96.1% accuracy, 0.959 F1- score, and 0.975 ROC-AUC, outperforming baseline models such as Logistic Regression and Random Forest. The SHAP explainability analysis identified interest rate, loan amount, and past due counts as the most influential predictors of default. Temporal validation across three time-based splits confirmed model stability, with less than 1% degradation in predictive performance. The integration of Gradient Boosting with SHAP not only enhances classification accuracy but also provides transparent, interpretable insights into borrower risk profiles. This research contributes to the advancement of ethical and explainable artificial intelligence in FinTech by offering a data-driven yet transparent framework for decision-making in digital lending ecosystems.