IDENTIFYING FRAUDULENT CREDIT CARD TRANSACTIONS USING ENSEMBLE LEARNING

Abstract

Credit card fraud detection has become increasingly complex due to the rapid growth of digital transactions and evolving cyber-attack strategies. Traditional detection systems often rely on static rules or single machine learning models, which struggle to generalize across diverse and highly imbalanced transaction datasets. This project proposes an advanced ensemble learning-based fraud detection framework centered on the XGBoost algorithm to enhance predictive reliability and scalability. The proposed mechanism integrates data pre-processing, imbalance handling techniques, and gradient boosting optimization to effectively distinguish fraudulent activities from legitimate transactions. By combining multiple weak decision learners into a unified predictive model, the system captures nonlinear feature interactions and subtle anomaly patterns that conventional methods may overlook. Experimental evaluation on both real-world and synthetic transaction datasets demonstrates that the ensemble-based approach improves precision, recall, and F1-score while maintaining low false positive rates. The findings indicate that structured real transaction data enables stronger pattern learning compared to highly randomized synthetic datasets. The proposed system not only strengthens fraud detection accuracy but also enhances operational efficiency, making it suitable for real-time financial monitoring environments.