Enhancing Brain Stroke Prediction Using Machine Learning for Early Intervention

Abstract

Stroke is still a significant global health concern that requires sophisticated predictive methods for early detection and treatment. In this work, a novel machine learning (ML) framework for automated stroke prediction is presented, and its accuracy and generalization skills are evaluated against those of six well-known classifiers. In order to guarantee clear decision-making in clinical applications, SHAP and LIME approaches are also used to highlight model interpretability. By combining local and global analytical approaches, the suggested framework improves the standardization of intricate machine learning models. Notably, Random Forest routinely achieves higher predicting accuracy than other algorithms. An enhanced ensemble strategy that uses a voting mechanism to leverage numerous classifiers and incorporates CATBOOST and a Stacking Classifier is presented in order to further increase performance. This study offers a thorough and trustworthy approach to early stroke diagnosis and treatment, which will ultimately lessen the serious health and financial effects of this common illness.