CCFD-SSL: Optimizing Real-Time Credit Card Fraud Detection Using Self-Supervised Learning and Contrastive Representations

Credit card fraud poses an ongoing challenge for financial entities, especially within real-time transaction systems. This paper presents an innovative application of Self-Supervised Learning (SSL) aimed at tackling the difficulties associated with highly skewed credit card fraud detection (CCFD) datasets. Using transaction data from European cardholders, our SSL-based method improves the model’s ability to differentiate between fraudulent and legitimate transactions, with minimal reliance on extensive labeled datasets. The approach utilizes contrastive learning to extract useful representations of transaction data, followed by fine-tuning with labeled samples. The model delivers state-of-the-art results, including 93.7% precision, 92.3% recall, an F1-score of 92.99%, and an AUPRC of 92.45%. These out-comes underscore the effectiveness of SSL in enhancing detection rates while reducing false positives in real-time environments. Additionally, we explore the scalability and adaptability of SSL-based frameworks in responding to evolving fraud patterns, highlighting their potential to bolster financial security.