ARCHIVES

Credit card fraud detection is challenging due to class imbalance and subtle patterns in transaction data. This study evaluates the effectiveness of Random Forest and XGBoost using a publicly available dataset. Preprocessing involves feature scaling with StandardScaler, class balancing through SMOTE, and eliminating non-informative features to enhance model accuracy. Random Forest is trained using all relevant features with optimized hyperparameters, including limited tree depth and minimum sample splits, to reduce overfitting and improve generalization. It demonstrates solid baseline performance across F1- score, recall, accuracy, and precision. XGBoost is trained on a selected subset of high-impact features to reduce dimensionality and accelerate training. It outperforms Random Forest, particularly in identifying minority-class fraud cases, making it more effective for real-time fraud detection. The study underscores the importance of combining robust preprocessing techniques with ensemble learning models to develop reliable fraud detection systems.