Improving Loan Approval Decisions: The Impact of Data Balancing on the Classifier’s Performance in Predicting Borrower Reliability

Creditworthiness prediction is critical for financial institutions to assess loan risks and improve lending efficiency. This study evaluates a comprehensive range of machine learning algorithms, including Decision Tree, Random Forest, Logistic Regression, and advanced ensemble methods, for their performance in credit risk prediction. Addressing data imbalance challenges through techniques such as random oversampling and SMOTE, the Stacking Ensemble model emerged as the most effective, achieving 93% accuracy with high precision, recall, and F1 score, validated by 10-fold cross-validation. The findings underscore the importance of selecting suitable models and preprocessing techniques for balanced and accurate predictions. Furthermore, the study highlights the potential for practical deployment and interpretability of machine learning models in real-world applications. It offers valuable insights for financial institutions to integrate these tools into their workflows. This research contributes to the field of credit risk management by demonstrating how advanced methodologies can overcome challenges and improve decision-making.