Description of Invention
The surge in digital transactions has intensified the need for effective credit card fraud detection, complicated by class imbalance. This study evaluates machine learning and deep learning methods—Random Forest, XGBoost, SVM, and CNNs—enhanced with SMOTE and stratified splitting to address imbalance. Using a Kaggle dataset, preprocessing included feature engineering on temporal and categorical data. A Streamlit dashboard tracked precision, recall, F1-score, and AUC-ROC. Random Forest reached 98% accuracy but only 2% precision for fraud, showing imbalance issues. With an AUC-ROC of 0.85, hybrid models proved superior in identifying complex fraud patterns, offering a promising solution for accurate, real-time fraud detection.