Graph Neural Network-Based Financial Fraud Detection and Cybersecurity Risk Analysis in Banking Networks

Md. Sumsuzzaman; Md Abdullah  Enayet; Sana Ullah; Masrufa Tasnim; Mst Anupa Nasrin  Khan; Tanaya  Jakir; Shakila  Sultana; Arat Hossain

doi:10.29284/ycdqpk10

Authors

Md. Sumsuzzaman Master's of Business Administration, International American University, USA.
Md Abdullah Enayet Master of Science in Cybersecurity, Washington University of Science and Technology, USA.
Sana Ullah MS in Management – Business Analytics, St. Francis College, Brooklyn, New York
Masrufa Tasnim Master of Business Administration (MBA) – Human Resource Management, University of Dhaka, Bangladesh.
Mst Anupa Nasrin Khan Master of Science in Zoology (Entomology Focus), The University of Rajshahi, Bangladesh.
Tanaya Jakir PhD in Information Technology, University of the Cumberlands, USA.
Shakila Sultana Master's in Accounting (M.A.), National University, Bangladesh.
Arat Hossain PhD in Information Technology University of the Cumberlands, USA.

DOI:

https://doi.org/10.29284/ycdqpk10

Keywords:

graph neural networks; financial fraud detection; anti-money laundering; network intrusion detection; explainable artificial intelligence; banking cybersecurity; class imbalance.

Abstract

Existing modern banking networks are vulnerable two classes of threat; typically scoped separately78: (a) financial fraud and money laundering (the transaction layer), and (b) cyber intrusions on the network-traffic layer. The two threats are relational, yet most detectors flatten this structure to per-record tabular features or use node-centric GNNs that neglect the wealth of attributes carried by transactions and flows. We propose a unified edge-centric GNN framework which models both fraud detection and intrusion detection as the same fundamental problem: binary edge classification on a directed multigraph. Trained on two strongly dissimilar graphs: an account-to-account money-laundering graph with 515,088 account nodes and 5,078,345 transaction edges (real data), and a NetFlow-based intrusion (derived from NF-UNSW-NB15-v2). A single backbone is trained to be able to learn on both graphs by fusing edge features directly into neighborhood aggregation. We compare the proposed model to strong tabular baselines (logistic regression, random forest, XGBoost and a multilayer perceptron) as well as structure-only GNNs (plain GCN and GraphSAGE) across three random seeds. Edge-feature fusion is decisive: for the money-laundering graph it raises area under the precision-recall curve (AUPRC) from approximately 0.004 for structure-only GNNs to 0166, and on the intrusion graph the proposed model achieves both best AUPRC (0985) and highest precision (0958) of any method evaluated. We additionally provide graph-native explanations through Integrated Gradients and edge-mask attribution along with a node level risk-scoring layer to transform edge probabilities to interpretable Low, Medium, and High-risk tiers. Results are reported transparently, including settings in which even extremely well-tuned tabular models remain competitive on this extreme-imbalance money laundering task, and we discuss the practical and regulatory value of a single relational, interpretable model over both banking threat surfaces.

References

[1] T. N. Kipf and M. Welling. Semi-Supervised Classification with Graph Convolutional Networks. International Conference on Learning Representations (ICLR), 2017.

[2] W. L. Hamilton, R. Ying, and J. Leskovec. Inductive Representation Learning on Large Graphs. Advances in Neural Information Processing Systems (NeurIPS), 2017.

[3] P. Velickovic, G. Cucurull, A. Casanova, A. Romero, P. Lio, and Y. Bengio. Graph Attention Networks. International Conference on Learning Representations (ICLR), 2018.

[4] K. Xu, W. Hu, J. Leskovec, and S. Jegelka. How Powerful are Graph Neural Networks? International Conference on Learning Representations (ICLR), 2019.

[5] Z. Wu, S. Pan, F. Chen, G. Long, C. Zhang, and P. S. Yu. A Comprehensive Survey on Graph Neural Networks. IEEE Transactions on Neural Networks and Learning Systems, 32(1):4-24, 2021.

[6] W. W. Lo, S. Layeghy, M. Sarhan, M. Gallagher, and M. Portmann. E-GraphSAGE: A Graph Neural Network based Intrusion Detection System for IoT. IEEE/IFIP Network Operations and Management Symposium (NOMS), 2022.

[7] M. Sarhan, S. Layeghy, N. Moustafa, and M. Portmann. NetFlow Datasets for Machine Learning-Based Network Intrusion Detection Systems. Big Data Technologies and Applications (BDTA/WiCON), Springer, 2021.

[8] N. Moustafa and J. Slay. UNSW-NB15: A Comprehensive Data Set for Network Intrusion Detection Systems. Military Communications and Information Systems Conference (MilCIS), 2015.

[9] Y. Dou, Z. Liu, L. Sun, Y. Deng, H. Peng, and P. S. Yu. Enhancing Graph Neural Network-based Fraud Detectors against Camouflaged Fraudsters. ACM International Conference on Information and Knowledge Management (CIKM), 2020.

[10] Y. Liu, X. Ao, Z. Qin, J. Chi, J. Feng, H. Yang, and Q. He. Pick and Choose: A GNN-based Imbalanced Learning Approach for Fraud Detection. The Web Conference (WWW), pp. 3168-3177, 2021.

[11] M. Weber, G. Domeniconi, J. Chen, D. K. I. Weidele, C. Bellei, T. Robinson, and C. E. Leiserson. Anti-Money Laundering in Bitcoin: Experimenting with Graph Convolutional Networks for Financial Forensics. arXiv:1908.02591, 2019.

[12] M. Weber, J. Chen, T. Suzumura, A. Pareja, T. Ma, H. Kanezashi, T. Kaler, C. E. Leiserson, and T. B. Schardl. Scalable Graph Learning for Anti-Money Laundering: A First Look. arXiv:1812.00076, 2018.

[13] T.-Y. Lin, P. Goyal, R. Girshick, K. He, and P. Dollar. Focal Loss for Dense Object Detection. IEEE International Conference on Computer Vision (ICCV), 2017.

[14] T. Chen and C. Guestrin. XGBoost: A Scalable Tree Boosting System. ACM SIGKDD International Conference on Knowledge Discovery and Data Mining (KDD), pp. 785-794, 2016.

[15] L. Breiman. Random Forests. Machine Learning, 45(1):5-32, 2001.

[16] R. Ying, D. Bourgeois, J. You, M. Zitnik, and J. Leskovec. GNNExplainer: Generating Explanations for Graph Neural Networks. Advances in Neural Information Processing Systems (NeurIPS), 2019.

[17] M. Sundararajan, A. Taly, and Q. Yan. Axiomatic Attribution for Deep Networks. International Conference on Machine Learning (ICML), 2017.

[18] E. A. Lopez-Rojas, A. Elmir, and S. Axelsson. PaySim: A Financial Mobile Money Simulator for Fraud Detection. European Modeling and Simulation Symposium (EMSS), pp. 249-255, 2016.

[19] Z. Liu, Y. Dou, P. S. Yu, Y. Deng, and H. Peng. Alleviating the Inconsistency Problem of Applying Graph Neural Network to Fraud Detection. ACM SIGIR Conference on Research and Development in Information Retrieval, pp. 1569-1572, 2020.