A New Method in Machine Learning Adapted for Credit Risk Prediction of Bank Loans

Abstract

The recent global financial crisis has significantly impacted the financial system, leading to major bank failures and prompting a reevaluation of credit risk management models. Given its critical role in maintaining banking stability, effective credit risk forecasting methods are essential. In light of this, various studies have introduced techniques to analyze, detect, and prevent bank credit defaults. In this paper, we present a new approach for predicting credit risk, known as the “Method of Separating the Learning Set into Two Balls.” This method involves partitioning a learning set into two distinct categories: the "Performing Ball," which contains feature vectors of customers with non-defaulting credits, and the "Non-Performing Ball," which includes vectors of customers with defaulting credits. To predict a customer’s default risk, it is sufficient to determine which ball their feature vectors belong to. If a customer’s vectors do not fall into either category, additional analysis is required for making a credit decision. We evaluated the performance of this method through extensive experimental tests and a comparative analysis. The findings suggest that our approach shows considerable promise for enhancing credit risk prediction in the banking sector.