Applying in machine learning and deep learning in finance industry: A case study on repayment prediction

Abstract

In the current era marked by the proliferation of peer-to-peer lending platforms, the imperative of ascertaining borrowers’ capacity to honor their financial obligations has assumed paramount significance. This endeavor transcends mere risk mitigation for individual investors, extending to the identification of judicious investment prospects. The present inquiry advocates for the adoption of sophisticated computational methodologies, including machine learning and deep learning, to analyze borrowers’ behavioral patterns, demographic profiles, and credit histories, thus facilitating the prognostication of loan repayment likelihood. Employed techniques encompass Logistic Regression (LR), K-Nearest Neighbor (KNN), Extreme Gradient Boosting (XGB), Light Gradient Boosting Machine (LGBM), in conjunction with deep learning architectures such as Long Short-Term Memory (LSTM) and Artificial Neural Network (ANN). Following methodological refinement, it becomes apparent that ensemble learning approaches, exemplified by XGB and LGBM, exhibit markedly superior predictive performance, surpassing conventional models with an accuracy rate exceeding 85%. Salient predictors include interest rates, credit ratings, and loan amounts. It is anticipated that the findings of this investigation will furnish investors with a potent analytical toolset for discerning and selecting loan portfolios, thereby fostering greater transparency and efficiency within the peer-to-peer lending ecosystem.