Loan default prediction is a critical challenge for financial institutions. On one hand, misclassifying high-risk borrowers can severely impact liquidity and solvency, and on the other hand, it may negatively impact the connection with good clients and ultimately lead to bad results. This study proposes a complete preprocessing framework that combines bootstrapping and random under-sampling to address the inherent class imbalance in credit datasets. Minority-class instances are replicated through bootstrapping, and the over-representation of non-default cases is reduced via under-sampling. The resulting training set be-comes more balanced, which enables models to better capture rare default pat-terns. We evaluate this approach using various machine learning models, including deep learning architectures (CNN and LSTM), support vector machines (SVM), and Random Forest (RF). The results are assessed through key evaluation metrics for imbalanced data classification, such as the F1 score and ROC AUC, which demonstrate that our sampling techniques significantly enhance the performance of the models. The findings underscore the ability of the proposed preprocessing strategy to mitigate class imbalance in training sets and open prom-icing directions for future research, including the integration of swarm intelligence optimization techniques for further model enhancement.