Abstract The rapid growth of FinTech systems requires efficient fraud detection mechanisms capable of processing large-scale financial transactions in real time. Traditional rule-based and machine learning approaches often suffer from limited scalability, adaptability, and high false-positive rates. This paper proposes a big data–driven framework that integrates deep learning with lightweight blockchain technology designed to reduce inference complexity and computational overhead during real-time deployment . The integration is implemented by embedding a compact Gated Recurrent Unit (GRU)-based fraud detection model within the transaction processing pipeline, where each transaction is first analyzed for fraud likelihood and then recorded on a permissioned Hyperledger Fabric blockchain with cryptographic hashing and smart contract validation , ensuring that only verified transactions are immutably stored, reducing redundant re-evaluation, preventing post-detection data tampering, and enabling consistent decision synchronization across distributed nodes—capabilities not achieved by standalone deep learning or blockchain-based approaches. The blockchain layer ensures data integrity and transparency with minimal overhead, while the deep learning model captures complex transactional patterns for adaptive detection. Experimental results demonstrate that the proposed model achieves 94.1% accuracy and 89.6% F1-score, outperforming traditional machine learning methods by 8–10% in accuracy, while reducing the false positive rate to 3.2% compared to 12.5% in rule-based systems. In addition, the framework enables efficient real-time processing with a latency of 4.3 ms per transaction, significantly faster than conventional deep learning models. These results confirm the effectiveness of the proposed approach in delivering accurate, scalable, and low-latency fraud detection for FinTech systems.
Darwish et al. (Fri,) studied this question.