Abstract
Credit card fraud detection remains a critical challenge in modern financial systems due to strict privacy regulations, institutional data silos, and increasingly adaptive adversarial behaviors. Centralized learning approaches, while effective within isolated environments, struggle to generalize across institutions and raise significant concerns regarding data security and trust. This paper proposes a Digital Twin–Enabled Blockchain-based Federated Learning framework for collaborative and privacy-preserving credit card fraud detection. The proposed architecture integrates federated learning to enable cross-institutional model training without raw data sharing, a permissioned blockchain layer to enforce trust, integrity, and auditability of model updates, and a digital twin layer to model evolving transaction behaviors and support adaptive analytics. Blockchain smart contracts verify model updates prior to aggregation, mitigating adversarial manipulation and ensuring accountable collaboration. Digital twins provide behavior-aware representations that enable anomaly detection, controlled adversarial experimentation, and system-level interpretability. By tightly coupling decentralized learning, trust enforcement, and behavioral modeling, the framework establishes a resilient and transparent collaborative intelligence ecosystem. While conceptual in nature, this work provides a formal system design foundation for trust-centric federated learning and digital twin–driven analytics, advancing the state of the art in secure and adaptive credit card fraud detection.