Firefly-Optimized Cloud-Enabled Federated Graph Neural Networks for Privacy-Preserving Financial Fraud Detection

Abstract

Financial fraud detection remains a critical challenge in the era of digital transactions, where the complexity and volume of data make it difficult to identify fraudulent activities accurately and efficiently. Existing methods often face issues related to scalability, data privacy, and performance. This paper proposes a Firefly-Optimized Federated Graph Neural Network (GNN) framework for privacy-preserving financial fraud detection. The framework leverages federated learning to enable decentralized training of fraud detection models while ensuring that sensitive data remains secure. Graph Neural Networks (GNNs) are employed to capture complex transactional relationships and enhance fraud detection accuracy. Additionally, the Firefly optimization algorithm is used to fine-tune the model parameters, improving detection performance and model efficiency. The proposed framework was evaluated using the IEEE-CIS Fraud Detection Dataset, showing impressive results with an accuracy of 99%, precision of 98%, recall of 98.5%, and an F1-score of 97%. These results highlight the framework's ability to effectively detect fraudulent transactions with minimal errors. The AUC-ROC score further emphasizes the model's strong performance in distinguishing fraudulent from non-fraudulent transactions. This study demonstrates the potential of combining federated learning, GNNs, and Firefly optimization to create a scalable, efficient, and privacy-preserving solution for real-time fraud detection in financial systems.