Purpose: The purpose of this study is to develop a model for detecting bank loan fraud using data mining techniques at Bank Mellat. Methodology: The methodology of this study is based on the Saunders Onion Model and the post-positivist approach. This research adopts a deductive method and evaluates machine learning algorithms for fraud detection by analyzing 5, 000 real banking transactions from Bank Mellat during the period from 1400 to 1402 (Islamic calendar). The CRISP-DM framework is employed as a guideline for the data mining process, which includes data collection, preprocessing, feature selection, modeling, and algorithm evaluation. The research method combines quantitative and qualitative approaches, utilizing algorithms such as Decision Tree, KNearest Neighbors (KNN), Logistic Regression, Isolation Forest, Support Vector Machine (SVM), and ensemble methods like Bagging and Boosting. The performance of the models is evaluated based on Accuracy, Recall, and F1-score metrics. Findings: The results indicate that the Boosting ensemble algorithm achieves the best performance, demonstrating 94. 7% accuracy, which surpasses the performance of other algorithms. The Bagging algorithm ranks second with an accuracy of 94. 2%. These findings suggest that ensemble algorithms outperform other models in fraud detection. Originality/Value: Originality in research mean what you are doing is from your own perspective although you may draw arguments from other research work to back up your arguments.