Credit Card Fraud Detection: Optimizing Performance with Supervised and Ensemble Learning

With the increasing trend of digital payments, credit card fraud has become a major challenge for financial institutions. The existing rule-based and knowledge-based traditional systems are not effective at detecting complex and changing fraud patterns, often leading to high false positive rates and lower efficiency that leads to customer dissatisfaction. This study aims to solve the issues with an efficient approach to detecting fraud in credit card transactions. This experiment investigates several machine learning models, such as Logistic Regression, Naive Bayes, XGBoost, Decision Trees, Random Forest, and Artificial Neural Networks, to build the detection systems and develop an ensemble Voting Classifier model combining these models to achieve better performance. This study provides strong evidence to support the advantage of ensemble methods over a single model. Previous studies paid little attention to single-model methods. The Voting Classifier with the ensemble method shows the best result in terms of both detection capability and false positive reduction among all other tested models, with accuracy reaching 1.00, which indicates the highest performance and robustness in the detection of fraud transactions. The proposed framework performs better than the conventional fraud detection system and other approaches due to the advantages of ensemble learning, such as heterogeneity, redundancy, and synergy. However, as we are aware, one single model is not enough to capture all known and unknown threats, especially for credit card fraud detection. We will, therefore, concentrate on this in future work to improve the trustworthiness of our system and, therefore, more effectively neutralize any emerging threats.