Predicting and Reducing Peer 2 Peer Late Payments using Large Numerical Models (LNMs)

Abstract

The inability of borrowers to repay loans poses a significant challenge to the sustainability of the Peer-to-Peer (P2P) lending sector. This study leverages predictive modeling techniques to analyze historical applicant data from Lending Club, focusing on reducing late payments through the proprietary Sriya Expert Index (SXI) Artificial Intelligence-Machine Learning (AI-ML) algorithm. SXI serves as a super feature, synthesizing the outputs of 5–10 machine learning algorithms into a simplified score/index, enabling accurate prediction of late payments. The model dynamically adjusts algorithmic weights to optimize precision, considering critical features such as credit history, income, and repayment behavior. In comparison to traditional machine learning models, the Sriya Expert Index (SXI) algorithm significantly outperforms established approaches in predicting late payments. Models such as Random Forest and XGBoost achieved accuracies of 78.80% and 84.52 %, respectively, while the Mixture of Experts (MOE) neural network reached 92.10%. The Support Vector Machine (SVM) with a linear kernel delivered an AUC of 0.935, slightly higher than the 0.92 AUC of XGBoost. However, SXI surpasses all these models with a near flawless accuracy of 99.80% and an AUC score of 0.998. This demonstrates the model’s superiority in identifying late payment risks and its potential to guide effective intervention strategies. One of the standout features of SXI enabled AI-ML is to improve business outcomes with actionable insights. The study outlines a phased methodology for reducing late payments (desired business outcome), achieving an initial 20% reduction and further improvements to 50% and 80% in the mid-term and long-term, respectively. These findings highlight the transformative potential of SXI in enhancing risk management in P2P lending, offering a scalable, data-driven solution to improve the financial health of the sector.