EEG-Driven Machine Learning for Stroke Detection in High-Risk Patients

Strokes continue to be a major reason for disability and mortality around the globe, necessitating the development of effective tools for early detection and intervention. In recent years, there has been an interest in utilizing bio-signals generated by the human body as potential detectors of stroke occurrence. Various types of bio-signals, including Electroencephalography (EEG), are employed in stroke detection studies. The use of bio-signals as a tool for stroke detection holds promise as a non-invasive, cost-effective, accurate, and transportable approach. Therefore, this paper employs Machine Learning (ML) techniques with EEG data to detect stroke occurrence. Four experimental setups were devised to investigate different feature engineering techniques. These included utilizing all features, selecting features with a Decision Tree (DT) based on different thresholds, as well as Principal Component Analysis (PCA) and Independent Component Analysis (ICA) for feature reduction. Empirical findings demonstrate that the fourth experimental setup, which comprises AdaBoost and XGBoost with ICA, yielded the best results, achieving an 89% accuracy, 86% precision, 100% recall, and 92% F1-score with an MCC value of 0.76.