DynSupplyNet: A Dynamic Graph Neural Network with Temporal Fusion for Supply Chain Risk Prediction and Propagation Analysis

Zheqi Hu; Mengdie Hu; Zhenghang Li

doi:10.63313/JCSFT.9066

Authors

Zheqi Hu University of Electronic Science and Technology of China, Chengdu, China Author
Mengdie Hu University of Pennsylvania, Philadelphia, PA, USA Author
Zhenghang Li Northeastern University, Boston, MA, USA Author

DOI:

https://doi.org/10.63313/JCSFT.9066

Keywords:

Dynamic Graph Neural Networks, Supply Chain Risk Prediction, Temporal Fusion, Credit Contagion, Risk Propagation, Supply Chain Finance, Network Analytics

Abstract

Supply chain disruptions and credit contagion propagate through complex inter-firm networks, posing severe challenges to financial institutions, supply chain finance platforms, and national regulators. Traditional risk assessment models typically treat firms as independent entities and ignore the structural and temporal dependencies intrinsic to real-world supply chains. To address these limitations, we propose DynSupplyNet, a unified framework that models the supply chain as a dynamic graph sequence and integrates a Dynamic Graph Neural Network (GNN) Encoder with a Temporal Fusion Layer to simultaneously capture cross-firm relational dependencies and evolving temporal dynamics. The model incorporates heterogeneous node features—including financial indicators, transaction patterns, and sector attributes—and edge features representing supplier–buyer relationships, dependency ratios, and credit exposures. DynSupplyNet generates time-aware firm-level embeddings and predicts disruption probability, credit default risk, and risk-propagation paths through a task-specific prediction head and a propagation simulator. Experiments conducted on the large-scale supply chain dataset show that DynSupplyNet consistently outperforms all baselines, including static GCNs, temporal GNNs, and sequence models. On the test set, DynSupplyNet achieves an Accuracy of 0.821, Precision of 0.781, Recall of 0.752, and an AUROC of 0.874, representing improvements of 4–7% over the strongest competing model. These gains confirm the effectiveness of dynamic graph modeling and temporal fusion in capturing evolving inter-firm dependencies, demonstrating DynSupplyNet’s clear advantage for real-world supply chain risk prediction and propagation assessment.

References

[1] Borgatti S P, Li X. On social network analysis in a supply chain context[J]. Journal of supply chain management, 2009, 45(2): 5-22.

[2] Carvalho V M, Nirei M, Saito Y U, et al. Supply chain disruptions: Evidence from the great east japan earthquake[J]. The Quarterly Journal of Economics, 2021, 136(2): 1255-1321.

[3] Acemoglu D, Akcigit U, Kerr W. Networks and the macroeconomy: An empirical exploration[J]. Nber macroeconomics annual, 2016, 30(1): 273-335.

[4] Kivelä M, Arenas A, Barthelemy M, et al. Multilayer networks[J]. Journal of complex networks, 2014, 2(3): 203-271.

[5] Inoue H, Todo Y. Firm-level propagation of shocks through supply-chain networks[J]. Nature Sustainability, 2019, 2(9): 841-847.

[6] Barrot J N, Sauvagnat J. Input specificity and the propagation of idiosyncratic shocks in production networks[J]. The Quarterly Journal of Economics, 2016, 131(3): 1543-1592.

[7] Sankar A, Wu Y, Gou L, et al. Dysat: Deep neural representation learning on dynamic graphs via self-attention networks[C]//Proceedings of the 13th international conference on web search and data mining. 2020: 519-527.

[8] Pareja A, Domeniconi G, Chen J, et al. Evolvegcn: Evolving graph convolutional networks for dynamic graphs[C]//Proceedings of the AAAI conference on artificial intelligence. 2020, 34(04): 5363-5370.

[9] Fathy A, Li K. TemporalGAT: Attention-based dynamic graph representation learning[C]//Pacific-Asia conference on knowledge discovery and data mining. Cham: Springer International Publishing, 2020: 413-423.

[10] Zhang C, Jiang M, Zhang X, et al. Multi-modal network representation learning[C]//Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining. 2020: 3557-3558.

[11] Baesens B, Van Gestel T, Viaene S, et al. Benchmarking state-of-the-art classification algorithms for credit scoring[J]. Journal of the operational research society, 2003, 54(6): 627-635.

[12] Fischer T, Krauss C. Deep learning with long short-term memory networks for financial market predictions[J]. European journal of operational research, 2018, 270(2): 654-669.

[13] Wei H, Wu Y, Li M. RAGN-IIoT: A Retrieval-Augmented NL2SQL Framework with Dynamic Sensor-Selection Guardrails for Industrial IoT Time-Series Data Warehouses[J]. Journal of Computer, Signal, and System Research, 2025, 2(7): 78-88.

[14] Wang L, Zhang X, Jiang L. Prediction Framework for E-Commerce Platform Sales Data Based on Informer: A Study on Furniture Sales on Amazon E-Commerce Platform[J]. Economics and Management Innovation, 2025, 2(6): 80-87.

[15] Xu S, Jiang L, Gu B. Design and Validation of a Smart Neuromorphic System Architecture for Algorithmic Trading[C]//Proceedings of the 2nd International Symposium on Integrated Circuit Design and Integrated Systems. 2025: 127-136.

[16] Li X, Li Z, Lin X. Automated Implementation of Machine Learning-Based Causal Inference in Product Operations Decision Making[C]//Proceedings of the 2nd International Symposium on Integrated Circuit Design and Integrated Systems. 2025: 204-209.

[17] Fan P, Li H, Hu M. Profit-Oriented Production and Pricing Optimization for Manufacturing Enterprises Using Proximal Policy Optimization[J]. Economics and Management Innovation, 2026, 3(2): 8-17.

[18] Li T, Li H, Zhou Y. E-commerce Sentiment Analysis Using Fine-tuned LLaMA3 Models: A QLoRA-based Approach[J]. Journal of Technology Innovation and Engineering, 2025, 1(4).