Ensemble-based feature selection and classification for ischemic stroke subtyping

Abstract Strokes have become a major cause of death in recent years due to their effects on the central nervous system. Hemorrhagic and ischemic strokes (IS) are the two most types that severely harm the central nervous system. Subtyping ischemic stroke is crucial for both the prognosis of the condition and for efficient intervention and therapy. Despite advances in machine learning, reliable IS subtyping remains challenging due to class imbalance, feature redundancy, and limited predictive performance in existing studies. In this study, we present an ensemble-based machine learning framework for ischemic stroke subtype classification utilizing clinical patient data classified according to the Oxfordshire Community Stroke Project (OCSP) scheme. To address challenges in the dataset, such as class imbalance and feature irrelevance, we prepare the dataset by applying Random Forest-based feature selection and Synthetic Minority Oversampling Technique (SMOTE) for data balancing. Categorical variables are encoded to enable efficient model training, and multiple classifiers are combined within an ensemble learning strategy to enhance robustness and generalization. A comparative analysis has been conducted to verify the efficacy of the proposed framework using a dataset of 16,636 samples of patients’ clinical data, which are categorized into four OCSP subtypes. Compared to previous studies, which reported moderate classification performance on IS subtyping, our proposed ensemble achieved significantly improved results, achieving 98% accuracy. This indicates a significant improvement over previously reported methods. These findings demonstrate the efficacy of ensemble learning combined with feature optimization for IS subtyping and indicate a high potential for clinical decision support applications.