ENHANCING CREDIT CARD FRAUD DETECTION FOR FINANCIAL SECURITY THROUGH ADVANCED ENSEMBLE LEARNING

Abstract

The increasing volume of digital financial transactions has made credit card fraud detection a critical challenge for financial institutions, particularly due to severe class imbalance and evolving fraudulent patterns. This study extends a baseline machine learning framework by evaluating conventional machine learning and deep learning models, including ANN, KNN, GBDT, SVM, CNN, and LSTM, alongside advanced ensemble methods such as XGBoost, LightGBM, CatBoost, Voting, Stacking, and Optuna-optimized LightGBM. Experiments were conducted on the European Credit Card Fraud and AMEX datasets using preprocessing techniques including feature selection, scaling, missing-value handling, and random undersampling. Model performance was evaluated using accuracy, precision, recall, F1-score, ROC-AUC, precision-recall analysis, and confusion matrices. The Optuna-optimized LightGBM achieved the best accuracy of 97.02%, improving upon the baseline LightGBM result of 96.92%. Overall, the findings demonstrate that ensemble learning combined with feature selection and hyperparameter optimization can improve fraud detection performance and provide an effective approach for enhancing the security of digital payment systems.