Machine learning for credit risk analysis across the United States

Abstract

Money-Lending financial institutions face risk, which necessitates adopting a robust framework to manage it effectively. While traditional methods have been applied across the financial industry, the advent of artificial intelligence offers organizations the opportunity to utilize advanced methods to manage credit risk. This paper focuses on the application of machine learning techniques for credit risk analysis. Secondary data on information related to borrowers was extracted from Kaggle database simulating Credit Bureau data. Two ensemble models in random forest and gradient boosting were adopted for this study. The findings showed that percentage of income for loan repayment, borrower’s income, and interest rates on loans are the most important features for determining defaulters. Furthermore, the evaluation results revealed that both the random forest and the gradient boosting algorithms performed well, with F1 scores of 92.9% and 93% respectively. It was recommended that financial institutions should priorities the verification and comprehensiveness of their data, as precise data is essential for developing resilient models.