Stroke Prediction Using a Stacked Ensemble Classifier

A stroke occurs when there is a disruption or decrease in blood flow to various brain regions. As a result, the brain cells in those areas die from a lack of oxygen and nutrients. A stroke is a medical emergency that requires immediate attention. Early detection and effective therapy are necessary to prevent additional harm to the damaged area of the brain and its potential consequences in other sections of the body. This study presents a novel stroke risk prediction model utilizing a stacked ensemble approach that integrates multiple machine learning classifiers, including Adaptive Boosting (AdaBoost), Random Forest (RF), Extreme Gradient Boosting (XGBoost), Support Vector Machine (SVM), Categorical Boosting (CatBoost), Light Gradient Boosting Machine (LGBM), and Multilayer Perceptron (MLP). Addressing the challenges of imbalanced datasets, we employ a cost-sensitive learning technique and introduce logistic regression as a meta-learner to aggregate predictions from the base models, enhancing overall accuracy. Evaluating the proposed model across ten distinct medical datasets, our results demonstrate exceptional performance in terms of accuracy, recall, F1 score, precision, Matthews Correlation Coefficient (MCC), and Kappa, significantly outperforming all the traditional classifiers. This research underscores the potential of ensemble methods in improving stroke prediction accuracy and highlights the importance of leveraging diverse algorithms to achieve robust predictive capabilities. Our finding contributes to the ongoing efforts in healthcare analytics, paving the way for more effective stroke risk assessment and timely intervention strategies.