ADVANCED STROKE PREDICTION WITH MACHINE LEARNING ALGORITHM VIA ENSEMBLE STACKING

Abstract

Abstract—Stroke is a significant global health issue, and accurately predicting the risk of stroke can aid healthcare providers in implementing timely interventions to reduce its incidence. This study employs an ensemble stacking approach to develop a stroke prediction model using a publicly available healthcare dataset. Feature selection techniques were used to identify the important predictors of stroke risk. The optimal parameters for the classifiers were determined using hyperparameter tuning. To construct the proposed stacking classifier, KNN, logistic regression, decision tree, and XGBoost were used as base classifiers, with Random Forest serving as the meta learner. The stacking model's performance was compared to that of traditional algorithms using various metrics such as precision, recall, F1-score, accuracy, confusion matrix, and MCC. The results indicate that the proposed stacking model has achieved the highest accuracy of 98.6%, outperforming the other traditional algorithms. Additionally, the precision, recall, f1 score, and MCC of this model are 0.98, 0.98, 0.99, and 0.97, respectively.