CREDIT LOAN DEFAULT RISK PREDICTION USING IMBALANCED CLASSIFICATION TECHNIQUES AND BORROWER FINANCIAL PROFILE FEATURES

Abstract

Accurate prediction of credit loan defaults is a crucial aspect of risk management in the financial sector, as it enables institutions to minimize losses and make informed lending decisions. However, building effective predictive models is challenging due to the inherent class imbalance in loan datasets, where non-defaulters significantly outnumber defaulters. This study presents a comprehensive machine learning framework designed to address this challenge and improve the accuracy of loan default prediction. The model utilizes key borrower financial features such as debt-toincome (DTI) ratio, credit utilization, payment history, and employment status to capture meaningful patterns related to credit risk. To handle the imbalance in the dataset, advanced techniques including Synthetic Minority Over-sampling Technique (SMOTE), class weighting, and cost-sensitive learning are employed. These methods help balance the data distribution and enhance the model’s ability to detect minority class instances, particularly high-risk borrowers. Multiple machine learning algorithms are implemented and evaluated using performance metrics suited for imbalanced data, such as Precision-Recall Area Under Curve (PR-AUC), F1-score, and recall for the minority class. The experimental results demonstrate that combining relevant financial features with appropriate resampling and weighting strategies significantly improves predictive performance. The proposed framework effectively increases the detection rate of potential defaulters while maintaining overall model reliability. This approach supports financial institutions in reducing credit risk, optimizing loan approval processes, and strengthening decision-making in credit management systems.