A Comparative Analysis of Classification Algorithms for Loan Eligibility on Imbalanced Data

The increasing complexity and diversity of modern loan applications demand a structured, unbiased predictive framework capable of addressing class imbalance while ensuring robust model evaluation. Traditional classification methods often falter in such settings, producing biased outcomes and compromising decision quality. To overcome these limitations, this study applies advanced preprocessing techniques and employs the Synthetic Minority Over-sampling Technique (SMOTE) to effectively balance class distributions. A comprehensive evaluation of diverse classification algorithms—including linear models, distance-based methods, tree-based learners, boosting techniques, and probabilistic classifiers—is conducted to determine the most suitable approach for loan eligibility prediction. To further enhance performance, ensemble learning strategies such as Bagging and Voting are integrated with extensive hyperparameter optimization using Random Search and Grid Search. The use of 5-fold cross-validation strengthens model reliability by mitigating overfitting and promoting generalization. The proposed ensemble framework achieves a test accuracy of 87.18 %, setting a strong benchmark for loan eligibility prediction and underscoring the effectiveness of structured ensemble modeling in financial risk assessment.