<![CDATA[Credit card fraud is a significant threat in the financial industry, causing significant financial losses annually, posing a challenge to both businesses and the financial sector. This requires research and development to identify fraud models that significantly improve over time. The purpose of this research is to develop a machine learning (ML)-based credit card fraud detection system with an ensemble approach to address the challenges of imbalanced data in digital financial transactions. The method used includes four main stages: data collection; SMOTE; Hyperparameter Tuning; and model evaluation. The dataset used is from Kaggle Credit Card Fraud Detection, which has a very low fraud proportion (0.17%). The increase in data volume was carried out using SMOTE on the training data. Three main models (Logistic Regression, Support Vector Classifier, Random Forest) and ensembles (hard and soft voting) were tested with hyperparameter tuning for optimal results. Random Forest performed best with an F1-Score of 0.8482 and an ROC-AUC of 0.9684. This model was able to detect 84% of fraudulent transactions with high precision, surpassing other models in handling imbalanced data. The combined advantages of RF and SMOTE are effective for fraud detection which is relevant for real-time systems in the financial sector.]]>
Copyrights © 2025
