<![CDATA[Strokes remain one of the leading causes of disability and mortality worldwide, underscoring the need for effective early detection and intervention methods. Recently, researchers have shown a growing interest in harnessing bio-signals, natural indicators produced by the human body, as potential markers for stroke detection. Multiple types of bio-signals, such as electroencephalography (EEG), are currently being explored in stroke diagnostic studies. This approach is promising because it offers a non-invasive, cost-effective, accurate, and portable means of detecting strokes. The objectives of this research are to investigate the effectiveness of deep learning (DL) techniques, including Convolutional Neural Networks (CNN), Long Short-Term Memory networks (LSTM), Recurrent Neural Networks (RNN), CNN-LSTM hybrid models, and CNN-Gated Recurrent Unit (CNN-GRU) models, for detecting early-stage strokes based on EEG data. In addition, the impact of diverse feature extraction techniques, including utilizing all features, selecting features with a Decision Tree (DT) based on different thresholds, Principal Component Analysis (PCA), and Independent Component Analysis (ICA), is analyzed to evaluate their influence on model performance. A comparative discussion is conducted across multiple experimental setups to identify the most effective DL and feature engineering combinations for stroke detection. Across 35 different experiments, the CNN-LSTM model with seven selected features using the DT method yields the best results, achieving 86% accuracy, 99% precision, 81% recall, and an F1-score of 89%.]]>
