DEEP LEARNING-BASED REAL-TIME CREDIT CARD FRAUD DETECTION IN FINANCIAL TRANSACTIONS

Abstract

In the rapidly evolving financial industry, credit card fraud remains a critical threat, with substantial financial and reputational impacts on both institutions and customers. Real-time fraud detection systems are essential to mitigate risks and protect transaction integrity. Traditionally, fraud detection relied on rule-based systems and basic machine learning models, which often struggle with high-dimensional data and the adaptability required for emerging fraud patterns. Recent research has explored deep learning approaches, yet limitations such as high false-positive rates and computational inefficiency continue to challenge system effectiveness. This paper proposes a novel deep learning-based model designed to enhance real-time detection accuracy while reducing false-positive instances in financial transactions. Building on previous models, our contribution includes an optimized architecture that leverages recurrent neural networks (RNNs) and attention mechanisms, enabling better temporal pattern recognition and adaptability to changing fraud tactics. To implement and evaluate the proposed model, we use the IEEE-CIS Fraud Detection dataset, a large, anonymized transactional dataset that provides a realistic and comprehensive basis for assessing model performance in diverse fraud scenarios. Our experimental results demonstrate a significant improvement in fraud detection rates and a reduction in false-positive alerts compared to conventional deep learning models, confirming the model's potential for enhancing security in high-stakes, real-time financial environments. The findings underscore the value of advanced deep learning architectures in addressing the pressing challenges of fraud detection, ultimately contributing to more robust and reliable transaction systems.