Comparative Analysis of Balancing Techniques with Synthetic Data in the Classification of Police Reports
Resumo
Este trabalho investiga técnicas de balanceamento de dados para classificação de boletins de ocorrência de Altamira/PA (2019–2023), utilizando o BERTimbau em ajuste fino sobre 22.117 registros em dez classes criminais. Comparam-se Back-Translation, Paraphrasing, Class Weights e Geração Sintética via LLM (Sabiá 7B) em um conjunto com razão de desbalanceamento de até 25,4:1. Embora as classes majoritárias alcancem F1-score entre 0,93 e 0,96 sem balanceamento, a Geração Sintética via LLM obteve os melhores resultados globais (F1 Macro: 0,896; Weighted: 0,911), demonstrando que estratégias de geração de texto são eficazes para melhorar a robustez do modelo em PLN com dados desbalanceados em domínios sensíveis.Referências
Bencke, L. and Moreira, V. P. (2024). Data augmentation strategies to improve text classification: a use case in smart cities. Language Resources and Evaluation, 58(2):659–694.
Brasil (2024). Anuário brasileiro de segurança pública 2024. [link]. Acesso em: 01 abr. 2026.
Buda, M., Maki, A., and Mazurowski, M. A. (2018). A systematic study of the class imbalance problem in convolutional neural networks. Neural Networks, 106:249–259.
Cauas, P., Pastor, R., de Mattos Neto, P. S., and Duarte, F. C. d. L. (2025). Enhancing crime forecasting with hybrid approaches: Integrating machine learning and statistical models. In Encontro Nacional de Inteligência Artificial e Computacional (ENIAC), pages 297–308. SBC.
de Castro Oliveira, J. V., Rufino, N., Levcovitz, S., Utino, M. Y. R., Lobato, F., Turine, M., Rezende, S., Junior, A. F. L. J., and Marcacini, R. M. (2025). Agents4gov: Privacy-preserving web agents using open llms for public sector tasks. In Encontro Nacional de Inteligência Artificial e Computacional (ENIAC), pages 580–591. SBC.
Devlin, J., Chang, M.-W., Lee, K., and Toutanova, K. (2019). Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies, volume 1 (long and short papers), pages 4171–4186.
Henning, S., Beluch, W., Fraser, A., and Friedrich, A. (2023). A survey of methods for addressing class imbalance in deep-learning based natural language processing. In Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pages 523–540.
Kesgin, H. T. and Amasyali, M. F. (2024). Advancing nlp models with strategic text augmentation: A comprehensive study of augmentation methods and curriculum strategies. Natural Language Processing Journal, 7:100071.
Pires, R., Abonizio, H., Almeida, T. S., and Nogueira, R. (2023). Sabiá: Portuguese Large Language Models, page 226–240. Springer Nature Switzerland.
Schütze, H., Manning, C. D., and Raghavan, P. (2008). Introduction to information retrieval, volume 39. Cambridge University Press Cambridge.
Silva, G., Kuribayashi, H., Santos, R., and Santos, A. (2024). Automating the classification of crime reports in altamira, pará using bert-based architectures. In Encontro Nacional de Inteligência Artificial e Computacional (ENIAC), pages 823–833. SBC.
Sokolova, M. and Lapalme, G. (2009). A systematic analysis of performance measures for classification tasks. Information processing & management, 45(4):427–437.
Souza, V., Santos, A., Soares, A., et al. (2025). Sensitive data protection in police reports: De-identification techniques and applications in machine learning. In Encontro Nacional de Inteligência Artificial e Computacional (ENIAC), pages 105–116. SBC.
Wei, J. and Zou, K. (2019). Eda: Easy data augmentation techniques for boosting performance on text classification tasks. In Proceedings of the 2019 conference on empirical methods in natural language processing and the 9th international joint conference on natural language processing (EMNLP-IJCNLP), pages 6382–6388.
Yang, W., Li, J., Fukumoto, F., and Ye, Y. (2020). Hscnn: A hybrid-siamese convolutional neural network for extremely imbalanced multi-label text classification. In Proceedings of the 2020 conference on empirical methods in natural language processing (EMNLP), pages 6716–6722.
Zhang, D., Mi, R., Zhou, P., Jin, D., Zhang, M., and Song, T. (2024). Large model-based data augmentation for imbalanced text classification. In 2024 5th International Seminar on Artificial Intelligence, Networking and Information Technology (AINIT), pages 1006–1010. IEEE.
Zhao, H., Chen, H., and Yoon, H.-J. (2023). Enhancing text classification models with generative ai-aided data augmentation. In 2023 IEEE International Conference On Artificial Intelligence Testing (AITest), pages 138–145. IEEE.
Ziyaden, A., Yelenov, A., Hajiyev, F., Rustamov, S., and Pak, A. (2024). Text data augmentation and pre-trained language model for enhancing text classification of low-resource languages. PeerJ Computer Science, 10:e1974.
Brasil (2024). Anuário brasileiro de segurança pública 2024. [link]. Acesso em: 01 abr. 2026.
Buda, M., Maki, A., and Mazurowski, M. A. (2018). A systematic study of the class imbalance problem in convolutional neural networks. Neural Networks, 106:249–259.
Cauas, P., Pastor, R., de Mattos Neto, P. S., and Duarte, F. C. d. L. (2025). Enhancing crime forecasting with hybrid approaches: Integrating machine learning and statistical models. In Encontro Nacional de Inteligência Artificial e Computacional (ENIAC), pages 297–308. SBC.
de Castro Oliveira, J. V., Rufino, N., Levcovitz, S., Utino, M. Y. R., Lobato, F., Turine, M., Rezende, S., Junior, A. F. L. J., and Marcacini, R. M. (2025). Agents4gov: Privacy-preserving web agents using open llms for public sector tasks. In Encontro Nacional de Inteligência Artificial e Computacional (ENIAC), pages 580–591. SBC.
Devlin, J., Chang, M.-W., Lee, K., and Toutanova, K. (2019). Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies, volume 1 (long and short papers), pages 4171–4186.
Henning, S., Beluch, W., Fraser, A., and Friedrich, A. (2023). A survey of methods for addressing class imbalance in deep-learning based natural language processing. In Proceedings of the 17th Conference of the European Chapter of the Association for Computational Linguistics, pages 523–540.
Kesgin, H. T. and Amasyali, M. F. (2024). Advancing nlp models with strategic text augmentation: A comprehensive study of augmentation methods and curriculum strategies. Natural Language Processing Journal, 7:100071.
Pires, R., Abonizio, H., Almeida, T. S., and Nogueira, R. (2023). Sabiá: Portuguese Large Language Models, page 226–240. Springer Nature Switzerland.
Schütze, H., Manning, C. D., and Raghavan, P. (2008). Introduction to information retrieval, volume 39. Cambridge University Press Cambridge.
Silva, G., Kuribayashi, H., Santos, R., and Santos, A. (2024). Automating the classification of crime reports in altamira, pará using bert-based architectures. In Encontro Nacional de Inteligência Artificial e Computacional (ENIAC), pages 823–833. SBC.
Sokolova, M. and Lapalme, G. (2009). A systematic analysis of performance measures for classification tasks. Information processing & management, 45(4):427–437.
Souza, V., Santos, A., Soares, A., et al. (2025). Sensitive data protection in police reports: De-identification techniques and applications in machine learning. In Encontro Nacional de Inteligência Artificial e Computacional (ENIAC), pages 105–116. SBC.
Wei, J. and Zou, K. (2019). Eda: Easy data augmentation techniques for boosting performance on text classification tasks. In Proceedings of the 2019 conference on empirical methods in natural language processing and the 9th international joint conference on natural language processing (EMNLP-IJCNLP), pages 6382–6388.
Yang, W., Li, J., Fukumoto, F., and Ye, Y. (2020). Hscnn: A hybrid-siamese convolutional neural network for extremely imbalanced multi-label text classification. In Proceedings of the 2020 conference on empirical methods in natural language processing (EMNLP), pages 6716–6722.
Zhang, D., Mi, R., Zhou, P., Jin, D., Zhang, M., and Song, T. (2024). Large model-based data augmentation for imbalanced text classification. In 2024 5th International Seminar on Artificial Intelligence, Networking and Information Technology (AINIT), pages 1006–1010. IEEE.
Zhao, H., Chen, H., and Yoon, H.-J. (2023). Enhancing text classification models with generative ai-aided data augmentation. In 2023 IEEE International Conference On Artificial Intelligence Testing (AITest), pages 138–145. IEEE.
Ziyaden, A., Yelenov, A., Hajiyev, F., Rustamov, S., and Pak, A. (2024). Text data augmentation and pre-trained language model for enhancing text classification of low-resource languages. PeerJ Computer Science, 10:e1974.
Publicado
19/10/2026
Como Citar
SILVA, Luan; HENRIQUE, Gustavo; ARAÚJO, Leandro; SANTOS, Adam; SOARES, Anderson.
Comparative Analysis of Balancing Techniques with Synthetic Data in the Classification of Police Reports. In: ENCONTRO NACIONAL DE INTELIGÊNCIA ARTIFICIAL E COMPUTACIONAL (ENIAC), 23. , 2026, Cuiabá/MT.
Anais [...].
Porto Alegre: Sociedade Brasileira de Computação,
2026
.
p. 37-48.
ISSN 2763-9061.
DOI: https://doi.org/10.5753/eniac.2026.27188.
