<![CDATA[Detecting fraudulent transactions remains a major challenge in digital financial systems due to the severe imbalance between legitimate and fraudulent records. This study aims to develop a classification model capable of identifying fraudulent transactions with high sensitivity to minority classes, while ensuring performance stability suitable for operational deployment. The methodology includes data preprocessing through outlier removal, feature normalization, and stratified data partitioning. To address class imbalance, the Synthetic Minority Over-sampling Technique (SMOTE) is applied to generate representative synthetic samples for the minority class. Multiple machine learning algorithms are evaluated, including Random Forest, Decision Tree, Bagging, Gradient Boosting, Logistic Regression, Neural Network, K-Nearest Neighbors, and Support Vector Machine. Model performance is assessed using Precision, Recall, F1-Score, AUC, and G-Mean. The results show that the proposed approach achieves stable and reliable performance, with an AUC of 0.89 and a G-Mean of 0.81, demonstrating its effectiveness for operational fraud detection and error minimization.]]>
Copyrights © 2026
