Exploring Data Augmentation for Stroke Severity Prediction: How Generative AI Enhances EEG Signal Classification

A hybrid EEG pipeline using SelectKBest feature selection, an SVM classifier, and WGAN-GP augmentation achieves 72% classification accuracy for stroke severity in 50 brain stroke patients.

• Why it matters: Clinical EEG datasets are limited by small sample sizes and class imbalance, and common augmentation can discard informative outliers that help characterize disease.
• What they did: The study compares SMOTE and ADASYN oversampling with generative models WGAN and WGAN-GP, using preprocessing, statistical and frequency-domain feature extraction, feature reduction, and multiple machine learning classifiers.
• The result: The SelectKBest + SVM + WGAN-GP combination yields the best accuracy (72%), suggesting generative augmentation can improve EEG-based stroke severity classification when integrated carefully on limited data.

AI summary of the abstract · may contain errors · not medical advice · read the paper.