Fraudulent Credit Card Activity Detection Using Adaptive Boosting and Aggregate Voting

Abstract

With the rapid digitization of financial transactions, credit card fraud has emerged as a major concern, posing serious risks to both individual users and financial institutions. This research addresses the challenge of detecting fraudulent credit card activities in a timely and accurate manner through a hybrid ensemble learning approach. The proposed framework integrates Adaptive Boosting (AdaBoost) and Aggregate Majority Voting to form a highly robust fraud detection system. AdaBoost, known for its ability to enhance the predictive performance of weak classifiers, is utilized as the core component. Multiple shallow decision trees are iteratively trained, with each successive model focusing on instances previously misclassified, thereby refining the detection of complex fraud patterns. To further strengthen the model's stability and decision reliability, the outputs of these classifiers are combined via a majority voting mechanism, where the final decision is determined by consensus among the classifiers. Experimental evaluation, conducted on both benchmark and real-world credit card datasets, demonstrates that this hybrid system outperforms traditional individual classifiers, standalone AdaBoost, and other conventional detection methods in terms of precision, recall, F1-score, and Receiver Operating Characteristic (ROC) curve analysis. The hybrid system also shows strong resilience in noisy environments, making it a viable solution for practical fraud prevention applications