Article: Dynamic identification model of financial fraud of listed companies based on XGBoost and graph neural network Journal: International Journal of Information and Communication Technology (IJICT) 2026 Vol.27 No.37 pp.51 - 65 Abstract: This paper addresses the challenge of identifying financial fraud in listed companies by proposing a dynamic identification model that integrates XGBoost and graph neural networks (GNNs). The model employs a dual-channel architecture: GNN mines heterogeneous graph structure features formed by corporate equity relationships, supply chain networks, and executive appointments to capture the 'risk contagion effect'; XGBoost analyzes structured features such as corporate financial indicators, governance structure, and market performance to identify individual abnormal signals. The dual-mode features are dynamically fused through an attention mechanism, and a rolling time window update strategy is designed to achieve periodic iterative parameter training. Experimental results on the 2015-2023 A-share dataset show that the model achieves an accuracy of 92.37%, a recall of 88.21%, and an AUC of 0.941, significantly outperforming single models and improving early warning capabilities for fraud by 23.6%. Inderscience Publishers - linking academia, business and industry through research

Title: Dynamic identification model of financial fraud of listed companies based on XGBoost and graph neural network

Authors: Dongwu Lin

Addresses: Guangzhou College of Technology and Business, Guangzhou, 510800, China

Abstract: This paper addresses the challenge of identifying financial fraud in listed companies by proposing a dynamic identification model that integrates XGBoost and graph neural networks (GNNs). The model employs a dual-channel architecture: GNN mines heterogeneous graph structure features formed by corporate equity relationships, supply chain networks, and executive appointments to capture the 'risk contagion effect'; XGBoost analyzes structured features such as corporate financial indicators, governance structure, and market performance to identify individual abnormal signals. The dual-mode features are dynamically fused through an attention mechanism, and a rolling time window update strategy is designed to achieve periodic iterative parameter training. Experimental results on the 2015-2023 A-share dataset show that the model achieves an accuracy of 92.37%, a recall of 88.21%, and an AUC of 0.941, significantly outperforming single models and improving early warning capabilities for fraud by 23.6%.

Keywords: financial fraud identification; eXtreme gradient boosting; XGBoost; graph neural network; GNN; dynamic model; attention mechanisms; risk contagion.

DOI: 10.1504/IJICT.2026.153264

International Journal of Information and Communication Technology, 2026 Vol.27 No.37, pp.51 - 65

Received: 21 Oct 2025
Accepted: 15 Dec 2025
Published online: 29 Apr 2026 *

Title: Dynamic identification model of financial fraud of listed companies based on XGBoost and graph neural network

Keep up-to-date