Temporal Graph Neural Networks for Real-Time Fraud Detection in Cross-Border Transactions
DOI:
https://doi.org/10.63282/3050-9262.IJAIDSML-V6I4P111Keywords:
Temporal Graph Neural Networks, Fraud Detection, Real-Time Analytics, Cross-Border Payments, Financial Transactions, Streaming Graphs, Anomaly Detection, Time-Series Graph LearningAbstract
Cross-border financial transactions are increasingly vulnerable to sophisticated and coordinated fraud attacks that evolve rapidly over time. Traditional fraud detection systems struggle to capture temporal dependencies and relational patterns inherent in real-time transaction networks. In this paper, we propose a novel approach using Temporal Graph Neural Networks (TGNNs) to detect fraudulent behavior in streaming, multi-national payment data. By modeling dynamic transaction graphs with time-aware message passing and temporal node embeddings, our framework captures both short-term anomalies and long-range dependencies. We design a low-latency inference pipeline capable of real-time deployment in financial networks. Experimental evaluation on synthetic and real-world cross-border transaction datasets demonstrates that our TGNN-based model significantly outperforms traditional machine learning and static GNN baselines, achieving up to 18% higher AUC and reducing false positives by 25%. The results highlight the potential of temporal graph learning to enhance security, compliance, and trust in global financial systems
References
[1] W. Zhang, Y. Rong, and T. Huang, “A survey on graph neural networks in financial risk analysis,” IEEE Trans. Neural Netw. Learn. Syst., vol. 33, no. 11, pp. 6246–6264, Nov. 2022.
[2] J. Wu et al., “Detecting fraud in financial transactions using dynamic graph attention networks,” in Proc. AAAI, 2021, pp. 11918–11925.
[3] Z. Liu, X. Han, and J. Sun, “Temporal graph neural networks: A comprehensive survey,” IEEE Trans. Knowl. Data Eng., early access, doi: 10.1109/TKDE.2024.3330512.
[4] M. Kipf and M. T. Le, “Fraud detection with temporal graph convolutional networks,” in Proc. NeurIPS Workshop, 2021.
[5] D. Nguyen et al., “Real-time graph neural networks for streaming fraud detection,” in Proc. IEEE BigData, 2022, pp. 1804–1812.
[6] Y. Ma et al., “Learning on dynamic graphs with missing data,” in Proc. ICML, 2020, pp. 6245–6255.
[7] A. Kaur, S. K. Jha, and P. Gupta, “GNN-based detection of fraudulent accounts in digital banking,” IEEE Access, vol. 10, pp. 73280–73291, 2022.
[8] S. Xu and F. Wang, “A temporal-spatial GNN approach for financial fraud detection,” in Proc. KDD, 2021, pp. 2323–2331.
[9] M. S. Awan et al., “Cross-border payment fraud detection using graph-based anomaly learning,” IEEE Trans. Comput. Soc. Syst., vol. 9, no. 3, pp. 521–533, Jun. 2024.
[10] H. Jin, Y. Zhang, and C. Li, “Temporal graph attention networks for fraud event prediction,” in Proc. IJCAI, 2022, pp. 3456–3462.
[11] B. Rao and L. Wang, “Streaming GNNs for fast financial event detection,” IEEE Trans. Big Data, early access, doi: 10.1109/TBDATA.2023.3284907.
[12] T. Chen et al., “Time-aware graph neural networks for transaction fraud detection,” in Proc. IEEE ICDM, 2023, pp. 1050–1055.
[13] K. Mohan et al., “Deep temporal embeddings for cross-border money laundering detection,” in Proc. ACM CIKM, 2020, pp. 2351–2360.
[14] R. Bansal and D. Sinha, “Anomaly detection in temporal financial graphs using contrastive learning,” IEEE Trans. Ind. Inform., vol. 19, no. 2, pp. 1700–1711, Feb. 2023.
[15] Y. He et al., “Continual learning in graph neural networks for evolving fraud patterns,” in Proc. ICLR, 2024.
[16] X. Lin, P. Zhao, and L. Du, “Edge-centric temporal GNNs for scalable fraud detection,” in Proc. NeurIPS, 2023.
[17] M. Ali and H. Kim, “End-to-end system for GNN-based real-time fraud detection in global fintech,” IEEE Internet Things J., vol. 12, no. 1, pp. 689–701, Jan. 2025.
[18] Y. Luo, Z. Tang, and F. Zhu, “Graph learning in cross-border payments: A real-world benchmark and evaluation,” IEEE Trans. Serv. Comput., early access, doi: 10.1109/TSC.2025.3341785.
