The Impact of Data Imbalance on the Explainability of Text Classification Models: A Comparative Analysis with SVM and LSTM
Resumo
This paper investigates the systematic impact of data imbalance and sampling strategies on the explainability of text classification models. By confronting a linear SVM with a LSTM network, we analyze the mathematical consistency of local post-hoc explanations generated by LIME and SHAP, alongside feature relevance and deep attention mechanisms across five datasets. Non-parametric statistical evaluations via Friedman’s test reveal that while SHAP maintains high stability due to its axiomatic foundation, LIME and attention mechanisms exhibit severe structural vulnerabilities. Notably, oversampling text sequences for LSTM triggers localized redundancy, causing attention weights to oscillate erratically and undermining explanation reliability.
Referências
ARAÚJO, J. P. B. d. (2020). Interpretabilidade de modelos de machine learning: aplicação no mercado de crédito. UFC.
BADJATIYA, P. e. a. (2017). Deep learning for hate speech detection in tweets. In INTERNATIONAL WORLD WIDE WEB CONFERENCES STEERING COMMITTEE, page 759–760. Proceedings of the 26th International Conference on World Wide Web Companion.
CHANDRA, Mayank Arya; BEDI, S. S. (2021). Survey on svm and their application in image classification. In International Journal of Information Technology, pages 1–11.
Chen, X. et al. (2023). Interpretable machine learning for imbalanced credit scoring datasets. European Journal of Operational Research, 308(1):356–370.
CORTES, Corinna; VAPNIK, V. (1995). Support-vector networks. In Machine learning, pages 273–297.
DABLAIN, D. e. a. (2024). Understanding imbalanced data: Xai& interpretable ml framework. In Machine learning, pages 3751–3769. DBLP.
Fujiwara, T. (2024). Knowledge distillation with resampling for imbalanced data classification: Enhancing predictive performance and explainability stability. Results in Engineering, 24:103406.
Hochreiter, S. (1997). Long short-term memory. Neural Computation MIT-Press.
INDRAHIMAWAN, M. R. e. a. (2023). Handling data imbalance using text augmentation for classifying public complaints. In 2023 International Conference on Computer, Control, Informatics and its Applications (IC3INA), pages 284–289. IEEE.
MENDES, T. M., Ferreira, D. D., SILVA, L. R. M., and DUQUE, C. A. (2019). Aplicação de subamostragem para análise de distorções supraharmônicas. In Congresso Brasileiro de Automática-CBA, volume 1.
MINH, D. e. a. (2022). Explainable artificial intelligence: a comprehensive review. In Artificial Intelligence Review, pages 1–66. SPRINGER NATURE.
MOHAMMED, R. e. a. (2020). Machine learning with oversampling and undersampling techniques: overview study and experimental results. In 2020 11th international conference on information and communication systems (ICICS), pages 243–248. IEEE.
PERES, Vitor, e. a. (2019). Análises de sentimentos: abordagem lexical de classificação de opinião no contexto mercado financeiro brasileiro. In Workshop of Artificial Intelligence Applied to Finance. ITA.
POCEBON, M. (2023). Ética na ia: Desafios e considerações éticas em aplicativos de ia e machine learning. In Jornada Acadêmica, pages 2–2. FAQI.
Rai, A. et al. (2025). Xai techniques under class imbalance. Frontiers in Artificial Intelligence, 8:1682919.
Roth, A. E. (1988). The Shapley value: essays in honor of Lloyd S. Shapley. Cambridge University Press.
SILVA, A. B. e. a. (2025). Evaluating how different balancing data techniques impact on prediction of premature birth using machine learning models. In PloS one, page e0316574.
VILONE, Giulia; LONGO, L. (2021). Notions of explainability and evaluation approaches for explainable artificial intelligence. In Information Fusion, pages 89–106. ELSEVIER.
WANG, S. e. a. (2018). Densely connected cnn with multi-scale feature attention for text classification. In IJCAI, pages 4468–4474. COAI.
Zafar, M. R. and Khan, N. (2021). Deterministic local interpretable model-agnostic explanations for stable explainability. Machine Learning and Knowledge Extraction, 3(3):525–541.
BADJATIYA, P. e. a. (2017). Deep learning for hate speech detection in tweets. In INTERNATIONAL WORLD WIDE WEB CONFERENCES STEERING COMMITTEE, page 759–760. Proceedings of the 26th International Conference on World Wide Web Companion.
CHANDRA, Mayank Arya; BEDI, S. S. (2021). Survey on svm and their application in image classification. In International Journal of Information Technology, pages 1–11.
Chen, X. et al. (2023). Interpretable machine learning for imbalanced credit scoring datasets. European Journal of Operational Research, 308(1):356–370.
CORTES, Corinna; VAPNIK, V. (1995). Support-vector networks. In Machine learning, pages 273–297.
DABLAIN, D. e. a. (2024). Understanding imbalanced data: Xai& interpretable ml framework. In Machine learning, pages 3751–3769. DBLP.
Fujiwara, T. (2024). Knowledge distillation with resampling for imbalanced data classification: Enhancing predictive performance and explainability stability. Results in Engineering, 24:103406.
Hochreiter, S. (1997). Long short-term memory. Neural Computation MIT-Press.
INDRAHIMAWAN, M. R. e. a. (2023). Handling data imbalance using text augmentation for classifying public complaints. In 2023 International Conference on Computer, Control, Informatics and its Applications (IC3INA), pages 284–289. IEEE.
MENDES, T. M., Ferreira, D. D., SILVA, L. R. M., and DUQUE, C. A. (2019). Aplicação de subamostragem para análise de distorções supraharmônicas. In Congresso Brasileiro de Automática-CBA, volume 1.
MINH, D. e. a. (2022). Explainable artificial intelligence: a comprehensive review. In Artificial Intelligence Review, pages 1–66. SPRINGER NATURE.
MOHAMMED, R. e. a. (2020). Machine learning with oversampling and undersampling techniques: overview study and experimental results. In 2020 11th international conference on information and communication systems (ICICS), pages 243–248. IEEE.
PERES, Vitor, e. a. (2019). Análises de sentimentos: abordagem lexical de classificação de opinião no contexto mercado financeiro brasileiro. In Workshop of Artificial Intelligence Applied to Finance. ITA.
POCEBON, M. (2023). Ética na ia: Desafios e considerações éticas em aplicativos de ia e machine learning. In Jornada Acadêmica, pages 2–2. FAQI.
Rai, A. et al. (2025). Xai techniques under class imbalance. Frontiers in Artificial Intelligence, 8:1682919.
Roth, A. E. (1988). The Shapley value: essays in honor of Lloyd S. Shapley. Cambridge University Press.
SILVA, A. B. e. a. (2025). Evaluating how different balancing data techniques impact on prediction of premature birth using machine learning models. In PloS one, page e0316574.
VILONE, Giulia; LONGO, L. (2021). Notions of explainability and evaluation approaches for explainable artificial intelligence. In Information Fusion, pages 89–106. ELSEVIER.
WANG, S. e. a. (2018). Densely connected cnn with multi-scale feature attention for text classification. In IJCAI, pages 4468–4474. COAI.
Zafar, M. R. and Khan, N. (2021). Deterministic local interpretable model-agnostic explanations for stable explainability. Machine Learning and Knowledge Extraction, 3(3):525–541.
Publicado
19/10/2026
Como Citar
SILVA, Luirys do Nascimento; RODRIGUES, Cleyton Mário de Oliveira.
The Impact of Data Imbalance on the Explainability of Text Classification Models: A Comparative Analysis with SVM and LSTM. In: ENCONTRO NACIONAL DE INTELIGÊNCIA ARTIFICIAL E COMPUTACIONAL (ENIAC), 23. , 2026, Cuiabá/MT.
Anais [...].
Porto Alegre: Sociedade Brasileira de Computação,
2026
.
p. 1400-1411.
ISSN 2763-9061.
DOI: https://doi.org/10.5753/eniac.2026.31095.
