Loan Default Prediction based on Machine Learning (LightGBM Model)

Abstract

With the rapid development of Internet finance, the number of online loan platforms has been increasing, and the scale of loan business is gradually expanding. At the same time, the bad debt and non-performing loan ratios have also risen sharply, the reasons for which include incomplete information of lenders, imprudent management of capital chain, inaccurate default prediction, etc. Therefore, loan platforms need a set of efficient and accurate loan default prediction solutions to ensure the healthy operation of Internet finance and avoid huge losses for platforms and investors. Current loan default prediction models mainly contain traditional machine learning methods such as logistic regression and decision tree, as well as integrated models such as Random Forest and GBDT. In this paper, we use LightGBM machine learning model to process massive loan default information using the lender information dataset on Tianchi Platform, and the process includes data preprocessing, feature engineering, model training and evaluation. The experimental results show that the LightGBM algorithm has a strong ability to predict loan default, with a final AUC value reaching about 0.73. This method is helpful for banks and Internet loan platforms to conduct background investigation and default prediction of loan applicants, so as to strengthen the ability of loan risk management.