Enhancing Structured Fraud Detection in Trade Operations through a Multi-Aspect Tabular Transformer-Based Architecture

• Douglas B. Silva UFG https://orcid.org/0000-0003-2708-5801
• Nadia F. F. da Silva UFG https://orcid.org/0000-0002-3875-2211
• Sergio T. Carvalho UFG https://orcid.org/0000-0002-4191-5173

DOI: https://doi.org/10.5753/lasdigov.2026.23889

Resumo

Structured tax fraud supported by shell companies remains a persistent challenge in goods and services trading, as valid invoices used to simulate non-existent transactions enable undue tax advantages. This paper proposes a multi-aspect shell company detection approach based on Transformer models applied to tabular data. The models capture complex dependencies between heterogeneous attributes and improve performance over traditional ensemble methods, particularly in recall, which is critical for highly imbalanced fraud detection scenarios. Results using real tax data indicate that Transformer-based tabular modeling is an effective and scalable alternative for supporting fiscal intelligence in identifying structured fraud.

Referências

Gomes, G. S. L. et al. (2023). Identificação de práticas de evasão fiscal utilizando aprendizagem de máquina: o caso das empresas de fachada e os créditos ilegais de icms. Master’s thesis.

Gorishniy, Y., Rubachev, I., Khrulkov, V., and Babenko, A. (2021). Revisiting deep learning models for tabular data. Advances in neural information processing systems, 34:18932–18943.

Gutheil, J. and Donsa, K. (2022). Saintens: self-attention and intersample attention transformer for digital biomarker development using tabular healthcare real world data. In dHealth 2022, pages 212–220. IOS Press.

Huang, X., Khetan, A., Cvitkovic, M., and Karnin, Z. (2020). Tabtransformer: Tabular data modeling using contextual embeddings. arXiv preprint arXiv:2012.06678.

Mehta, P., Mathews, J., Kumar, S., Suryamukhi, K., Sobhan Babu, C., Kasi Visweswara Rao, S. V., K., C., S., S., and J., Z. L. (2019). Big data analytics for nabbing fraudulent transactions in taxation system. pages 95–109.

Melo, R. R. (2022). Graph neural network approach to detect shell companies in the brazilian state tax system. PhD thesis.

Mustafa, F. M., Al-Hussainy, A. F., et al. (2025). Tabnet and tabtransformer: Novel deep learning models for chemical toxicity prediction in comparison with machine learning. Journal of Applied Toxicology.

Reis, R. d. S. (2024). Predição e identificação de empresas noteiras utilizando machine learning na secretaria de fazenda do ceará.

Saito, T. and Rehmsmeier, M. (2015). The precision-recall plot is more informative than the roc plot when evaluating binary classifiers on imbalanced datasets. 10(3).

Silva, D. and Carvalho, S. (2021). Data science methods and techniques for goods and services trading taxation: a systematic mapping study.

Silva, D., Carvalho, S., and Silva, N. F. F. (2023). On identifying early blockable taxpayers on goods and services trading operations. In 24th International Conference on Digital Government Research, pages 405–413.

Silva, D., Carvalho, S. T., and Silva, N. (2022). Comparative analysis of classification algorithms applied to circular trading prediction scenarios. In EGOVIS 2022, Proceedings, pages 95–109. Springer.

Silva, D., Felix, N., and Carvalho, S. (2024). Detection of structured fraud supported by shell companies on goods and services trading operations. In International Conference on Electronic Government and the Information Systems Perspective, pages 168–183. Springer.

Somepalli, G., Goldblum, M., Schwarzschild, A., Bruss, C. B., and Goldstein, T. (2021). Saint: Improved neural networks for tabular data via row attention and contrastive pre-training. arXiv preprint arXiv:2106.01342.

Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, Ł., and Polosukhin, I. (2017). Attention is all you need. Advances in neural information processing systems, 30.

Wang, W., Xiao, X., Liu, M., Lan, Q., Huang, X., Tian, Q., Roy, S. K., and Wang, T. (2024). Multi-dimension transformer with attention-based filtering for medical image segmentation. In IEEE 36th International Conference on Tools with Artificial Intelligence (ICTAI), pages 632–639. IEEE.