Design and Implementation of an Ensemble Learning–Based Credit Card Fraud Detection System

Abstract

Fraudulent transactions using credit cards is a huge issue for financial institutions as hackers can pose as authorised cardholders. Several resampling techniques such as oversampling, undersampling and SMOTE are used to solve the class imbalance problem on European Data and Sparkov Data. Various approaches are adopted to improve the classification performance through ensemble learning to increase the accuracy and robustness of the classification. In this paper, a framework for ensemble is proposed using advanced resampling techniques during model training and prediction to optimise the process. Through a comprehensive evaluation of different classification models and by using Stacking Classifier with the smart combination of many base models, it is observed that it achieves a better accuracy, precision, recall and F1 score for all the methodologies. This method could be a powerful tool that helps enhance fraud detection systems by detecting fraudulent transactions accurately without raising too many false alarms. The proposed solution emphasizes the importance of incorporating ensemble methods and data balancing techniques for solving financial fraud detection problems. Index Terms - Fintech, credit card fraud detection, ensemble learning, machine learning, simulated dataset, real-world data set.