DFG-JointGNN: A Data-Flow Graph Neural Network for Unified Sales Order Fraud Detection and Sales Forecasting
DOI:
https://doi.org/10.71222/y7xf9q04Keywords:
Sales Order Fraud Detection, Sales Forecasting, Data-Flow Modeling, Graph Neural Networks, Multi-Task Learning, Transaction GraphAbstract
Sales order systems in e-commerce platforms, enterprise information systems, and supply chain finance generate large-scale transactional data streams that exhibit strong process dependency and complex entity interactions. Fraudulent behaviors such as fake orders, brushing transactions, and refund abuse not only cause direct economic losses but also distort historical sales data, leading to biased and unreliable sales forecasting. Existing studies typically address sales order fraud detection and sales forecasting as independent tasks, overlooking the intrinsic coupling between fraud risk and sales demand in real transaction flows. In this paper, we propose DFG-JointGNN, a data-flow graph neural network that unifies sales order fraud detection and sales forecasting within a single learning framework. We model the full lifecycle of sales orders-including order creation, payment, logistics fulfillment, and settlement-as a temporal heterogeneous data-flow graph, where customers, orders, merchants, payment accounts, and logistics nodes are jointly represented. A relation-aware temporal graph attention network is employed to capture both structural dependencies and temporal dynamics of transaction flows. Fraud detection is performed at the order level, while sales forecasting is conducted at the merchant or product level through a fraud-aware aggregation mechanism that explicitly suppresses the influence of high-risk orders. Experiments on real-world and semi-synthetic sales order datasets show that DFG-JointGNN consistently outperforms state-of-the-art baselines. For fraud detection, it achieves an AUC of 0.956, improving approximately 6.3% over the strongest baseline (Temporal GAT, 0.893). For sales forecasting, it reduces RMSE to 78.4, a 15.2% improvement compared to the fraud-agnostic Temporal GAT model. These results confirm that jointly modeling fraud detection and sales forecasting with transaction data-flow graphs enhances both predictive accuracy and operational robustness.References
1. B. Liu, Q. Sun, and L. Wei, "Multimodal Forgery Recognition Algorithm and System Design for AI Frauds," In Proceedings of the 2nd International Symposium on Integrated Circuit Design and Integrated Systems, September, 2025, pp. 156-160. doi: 10.1145/3772326.3774725
2. B. H. Kim, "Development of Online Fraud Detection and Sales Prediction Model using Supply Chain Dataset," Journal of System and Management Sciences, vol. 13, no. 2, pp. 501-514, 2023.
3. T. Chen, and C. Guestrin, "Xgboost: A scalable tree boosting system," In Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining, August, 2016, pp. 785-794.
4. Y. Xie, G. Liu, M. Zhou, L. Wei, H. Zhu, R. Zhou, and L. Cao, "A spatial-temporal gated network for credit card fraud detection by learning transactional representations," IEEE Transactions on Automation Science and Engineering, vol. 21, no. 4, pp. 6978-6991, 2023. doi: 10.1109/tase.2023.3335145
5. R. Chalapathy, and S. Chawla, "Deep learning for anomaly detection: A survey," arXiv preprint arXiv:1901.03407, 2019.
6. C. Liu, L. Sun, X. Ao, J. Feng, Q. He, and H. Yang, "Intention-aware heterogeneous graph attention networks for fraud transactions detection," In Proceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining, August, 2021, pp. 3280-3288. doi: 10.1145/3447548.3467142
7. G. E. Box, G. M. Jenkins, G. C. Reinsel, and G. M. Ljung, "Time series analysis: forecasting and control," John Wiley & Sons, 2015.
8. A. Graves, "Long short-term memory," Supervised sequence labelling with recurrent neural networks, pp. 37-45, 2012. doi: 10.1007/978-3-642-24797-2_4
9. A. K. Soni, and P. Jain, "Decoding Sales Order Anomalies: Advanced Predictive Modeling and Discrepancy Resolution Utilizing Machine Learning Algorithms," International Journal of Advanced Computer Science & Applications, vol. 16, no. 7, 2025. doi: 10.14569/ijacsa.2025.0160767
10. S. Xu, L. Jiang, and B. Gu, "Design and Validation of a Smart Neuromorphic System Architecture for Algorithmic Trading," In Proceedings of the 2nd International Symposium on Integrated Circuit Design and Integrated Systems, September, 2025, pp. 127-136. doi: 10.1145/3772326.3774721
11. T. Ma, and G. Ke, "Multi-task learning for financial forecasting," arXiv preprint arXiv:1809.10336, pp. 5-17, 2018.
12. T. N. Kipf, and M. Welling, "Semi-supervised classification with graph convolutional networks," arXiv preprint arXiv:1609.02907, 2016.
13. G. Makhmetova, "Automated Anomaly Detection for Sales and Inventory in Data-Driven Industries," Universal Library of Engineering Technology, vol. 2, no. 1, 2025.
14. E. Rossi, B. Chamberlain, F. Frasca, D. Eynard, F. Monti, and M. Bronstein, "Temporal graph networks for deep learning on dynamic graphs," arXiv preprint arXiv:2006.10637, 2020.
15. R. Luo, N. Wang, and X. Zhu, "Fraud detection and risk assessment of online payment transactions on e-commerce platforms based on llm and gcn frameworks," arXiv preprint arXiv:2509.09928, 2025.
