Enhancing Credit Card Fraud Detection in Banking Using Neural Networks

Abstract

Credit card fraud has become one of the most critical issues in modern banking systems due to the rapid increase in digital transactions and online payment platforms. Traditional fraud detection techniques often fail to identify sophisticated fraudulent activities because of highly imbalanced transaction data and evolving fraud patterns. This paper presents an intelligent fraud detection framework using the TabNet deep learning architecture for efficient classification of fraudulent and legitimate transactions. TabNet utilizes sequential attention mechanisms to identify the most important transaction features while maintaining interpretability and computational efficiency. The proposed system includes data preprocessing, class balancing, model training, performance evaluation, visualization, and deployment through a Flask-based web application. Experimental results demonstrate that the proposed TabNet model achieves high accuracy, precision, recall, and ROC-AUC performance while reducing false positives compared to traditional machine learning approaches. The system also supports real-time prediction and visualization for practical banking applications.