Mitigating Risk in P2P Lending Network: Enhancing Predictions with GenAI and SMOTE

Peer-to-peer (P2P) lending is a major revolution in the field of finance where it transformed the market by eliminating the need for middlemen or conventional intermediaries such as banks, connecting borrowers directly with investors. This transformation offers several advantages, including potentially lower interest rates for borrowers and higher returns for investors. However, it also introduces risks, particularly the possibility of borrowers defaulting on their loans which could lead to significant losses. Research indicates that classification models can be leveraged to address this risk. However, the real-world datasets available are heavily skewed which could lead to bias in the prediction and model over-fitting. Existing research utilize conventional approaches such as Synthetic Minority Over-sampling Technique (SMOTE) for balancing data and ensemble models. This study addresses these challenges by implementing a comparative study between SMOTE and generative AI for data synthesis to rationalize the effects of modern approaches. Further it also explores the inclusion of additional features as compared to existing research. Ensemble modeling approaches were adopted for the purpose of this study. Logistic Regression, Support Vector Machine (SVM), KNN, and Random Forest were selected to determine the best base model to be used for stacking. XGBoost, LightGBM, and AdaBoost were the three selected models for stacking. XGBoost outperformed all other models, achieving an average accuracy of 99.4% and average F1-score of 97.4% using SMOTE synthesis. GenAI synthesis obtained similar performance.