Answers are grounded in this paper and cite their sources.
In clinical datasets, data augmentation is often avoided because it can remove important unusual values, known as outliers, which are important for understanding diseases accurately. However, clinical EEG datasets often suffer from two key limitations: small sample sizes and class imbalance, which significantly reduce classification performance. To overcome these challenges, this study applies generative AI-based data augmentation techniques with a carefully designed hybrid pipeline that effectively enhances performance even on limited EEG data. Using EEG signals from 50 brain stroke patients, we investigate both traditional oversampling methods (SMOTE and ADASYN) and advanced generative models, including WGAN and WGAN-GP. Our proposed pipeline involves preprocessing EEG recordings, extracting comprehensive statistical and frequency-domain features, applying feature reduction, and evaluating multiple machine learning classifiers. Experimental results show that the combination of SelectKBest feature selection, SVM classifier, and WGAN-GP augmentation achieves the highest classification accuracy of 72%, outperforming other methods. This work introduces a novel hybrid approach for EEG-based stroke severity classification, demonstrating that generative augmentation, when integrated properly, can significantly improve performance on small clinical datasets. To the best of our knowledge, this combination of techniques has not been explored before on any dataset, providing a solid foundation for future research in automated EEG-based stroke diagnosis.
No takes yet. Share an insight, caveat, or question.
Sarkar et al. (2026) studied this question.
Synapse has enriched 8 closely related papers on similar clinical questions. Consider them for comparative context: