Inductive Tax Evasion Detection: A Hybrid Framework Integrating Community Discovery and GraphSAG

Document Type : Original Article

Authors
Fatemeh Ayoughi Tafti 1
Ahmad Ebrahimi 2
Babak Farhang-Moghaddam 3
Alireza Rashidi-Komijan 4
Roya Soltani 5
1 Ph.D. Candidate, Department of Information Technology Management- Business Intelligence, Faculty of Management and Economics, SR.C., Islamic Azad University, Tehran, Iran
2 Associate Professor, Department of Industrial Engineering, SR.C., Islamic Azad University, Tehran, Iran
3 Associate Professor, Department of Systems and Economics, Institute for Management and Planning studies, Tehran, Iran.
4 Associate Professor, Department of Industrial Engineering, SR.C., Islamic Azad University, Tehran, Iran.
5 Assistant Professor, Department of Industrial Engineering, Faculty of Engineering, Khatam University, Tehran, Iran
10.22034/jmaak.2026.78986.4669
Abstract
Abstract
Tax evasion is one of the most significant challenges in tax systems, with a considerable portion stemming from transactions among related parties, profit shifting, and networked behaviors between companies. The complexity of ownership structures, multi-layered transaction chains, and hidden relationships between companies render traditional tax evasion detection methods inadequate for analyzing organized patterns. To address this gap, this study presents a graph-based framework utilizing community detection with the Louvain algorithm and the GraphSAGE graph neural network to identify hidden collusive structures, anomalous clusters, and tax evasion probabilities within a network of over 162,000 Iranian companies. The research data includes 33 financial, banking, and ownership indicators, with over 300,000 transactional and structural relationships extracted among companies. Results indicate that the proposed model achieves an AUC of 0.9579, an accuracy of 0.8905, and an F1-score of 0.8944, demonstrating high discriminative capability in identifying high-risk communities and network patterns contributing to tax evasion. Community analysis further reveals that a significant portion of anomalous relationships occurs within dense clusters of affiliated companies and economic groups. Overall, the findings suggest that combining community detection and GraphSAGE provides an effective tool for detecting complex tax evasion patterns in large-scale data and can enhance risk analysis processes in tax administration.
Keywords
Graph Analysis
Community Detection
Graph Neural Networks
Tax Evasion
Financial Fraud Detection
Inductive Learning
  1. احمدپور، ا.، جعفری، س. م.، و صراف، ف. (۱۴۰۴). مدل‌سازی فرار مالیاتی معاملات اشخاص وابسته: رویکرد گراف‌کاوی و شبکه‌های عصبی عمیق. پژوهشنامه مالیات، 33(113)، 1–45.
  2. حیرانی، ف.، زارعیان، ح.، و معین‌الدین، م. (۱۴۰۰). شناسایی شاخصه‌های مؤثر بر ریسک حسابرسی مالیاتی با روش ترکیبی اکتشافی. فصلنامه دانش حسابداری و حسابرسی مدیریت، ۳۳، ۱۲۱–۱۳۹.
  3. زارع بهنمیری، م. ج.، رحیمیان امیری، م. ح.، و تیرگان، م. (۱۴۰۱). اولویت‌بندی عوامل موثر بر فرار مالیاتی. تحقیقات حسابداری و حسابرسی، 54، 187–202.
  4. مددی‌زاده، ا.، تائبی‌نقندری، ا.ح.، و زینلی، ح. (۱۴۰۱). تأثیر استراتژی‌های مالیاتی جسورانه و تقلب حسابداری بر خوانایی گزارشگری مالی شرکت‌ها. فصلنامه دانش حسابداری و حسابرسی مدیریت، ۴۳، ۳۷۹–۳۹۲.
  5. مرادی، م.، اسکندر، ه.، و گرشاسبی، ف. (۱۴۰۲). تأثیر گرایش مدیران به ریسک و انتظارات آن‌ها بر گزارشگری مالی تقلب‌آمیز: شواهدی از بازی نظریه‌ای. فصلنامه دانش حسابداری و حسابرسی مدیریت، ۵۳، ۱–۲۳.
  6. Blondel, V. D., Guillaume, J. L., Lambiotte, R., & Lefebvre, E. (2008). Fast unfolding of communities in large networks. Journal of Statistical Mechanics: Theory and Experiment, *2008*(10), P10008.
  7. Chen, Z., Wang, Y., & Liu, Q. (2023). Enhancing GraphSAGE with gated aggregation for financial fraud detection. Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 2455-2465.
  8. Grama, T.-I. (2025). Money Laundering Detection Using Graph Neural Networks Enhanced with Autoencoder Components. Studia Universitatis Babeș-Bolyai Informatica, 70(1–2), 88–105. https://doi.org/10.24193/subbi.2025.06
  9. Grover, A., & Leskovec, J. (2016). node2vec: Scalable feature learning for networks. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 855–864.
  10. Gounoue, S., Sao, A., & Gottschalk, S. (2025). TeMP-TraG: Edge-based Temporal Message Passing in Transaction Graphs.
  11. Hamilton, W., Ying, Z., & Leskovec, J. (2017). Inductive representation learning on large graphs. Advances in Neural Information Processing Systems, *30*.
  12. Huang, R. (2025). FinGuard-GNN: Dynamic Graph Neural Network Framework for Financial Fraud Detection. Frontiers in Business, Economics and Management, 19(3), 121–125. https://doi.org/10.54097/wh6hg844
  13. Kipf, T. N., & Welling, M. (2017). Semi-supervised classification with graph convolutional networks. International Conference on Learning Representations (ICLR).
  14. Kulkarni, O., & Chandra, R. (2025). DynBERG: Dynamic BERT-based Graph Neural Network for Financial Fraud Detection. arXiv.
  15. Lee, J., Lee, J., & Park, S. (2016). Detecting financial fraud using graph-based features. Expert Systems with Applications, *59*, 84–93.
  16. Liu, Y., Li, Z., Zhou, C., & Yang, J. (2022). Temporal Graph Neural Networks for Dynamic Fraud Detection in E-commerce Transactions. IEEE Transactions on Knowledge and Data Engineering, *35*(8), 7895-7908.
  17. Ngai, E. W. T., Hu, Y., Wong, Y. H., Chen, Y., & Sun, X. (2011). The application of data mining techniques in financial fraud detection: A classification framework and an academic review of literature. Decision Support Systems, *50*(3), 559–569.
  18. OECD. (2023). Tax Administration 2023: Comparative Information on OECD and Other Advanced and Emerging Economies. OECD Publishing.
  19. Perozzi, B., Al-Rfou, R., & Skiena, S. (2014). DeepWalk: Online learning of social representations. Proceedings of the 20th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 701–710.
  20. Rahmati, M. (2025). Real-Time Financial Fraud Detection Using Adaptive Graph Neural Networks and Federated Learning. International Journal of Management and Data Analytics (IJMADA), 5(1), 98–110.
  21. Shi, X., Ye, Y., & Zhang, Y. (2023). Financial fraud detection using graph neural networks: A comprehensive study. Journal of Financial Technology, *15*(2), 45–60.
  22. Takahashi, R., Nishimura, H., & Matsuda, K. (2025). A Graph Neural Network Model for Financial Fraud Prevention. Frontiers in Artificial Intelligence Research, 2(1).
  23. Velickovic, P., Cucurull, G., Casanova, A., Romero, A., Lio, P., & Bengio, Y. (2018). Graph attention networks. International Conference on Learning Representations (ICLR).
  24. Wang, C., Zhang, Y., & Liu, J. (2023). Heterogeneous Graph Attention Networks for Tax Evasion Detection. ACM Transactions on Knowledge Discovery from Data, *17*(4), 1–24.
  25. Wu, Z., Pan, S., Chen, F., Long, G., Zhang, C., & Yu, P. S. (2021). A comprehensive survey on graph neural networks. IEEE Transactions on Neural Networks and Learning Systems, *32*(1), 4–24.
  26. Zhang, S., Tong, H., Xu, J., & Maciejewski, R. (2019). Graph convolutional networks: A comprehensive review. Computational Social Networks, *6*(1), 1–23.
  27. Zhou, J., Cui, G., Hu, S., Zhang, Z., Yang, C., Liu, Z., ... & Sun, M. (2020). Graph neural networks: A review of methods and applications. AI Open, *1*, 57–81.