Voting Classifier as a Balanced Framework for Fraud Detection in Imbalanced Credit Card Transactions

Abstract

Credit card fraud is a critical concern for financial institutions, as it leads to significant economic losses and compromises customer confidence. Detecting fraudulent activity remains challenging due to the extreme imbalance between legitimate and fraudulent transactions in real-world datasets. In this study, the publicly available dataset is analyzed to investigate the effectiveness of machine learning algorithms for fraud detection. The dataset exhibits a severe skew, with fraudulent cases representing only a small fraction of all transactions. To address this imbalance, the Synthetic Minority Oversampling Technique (SMOTE) is employed, enabling models to better learn discriminatory patterns. Several machine learning approaches, including Logistic Regression, Random Forest are implemented and evaluated. Performance is measured using accuracy, precision, recall, F1-score, and the area under the receiver operating characteristic curve (ROC-AUC), ensuring a comprehensive assessment of classification capability. Experimental results demonstrate that resampling combined with ensemble methods significantly improves the detection of minority fraud cases while minimizing false positives. This work emphasizes the importance of handling imbalanced data in fraud detection and provides insights into the potential of machine learning to enhance the security and reliability of electronic payment systems.