Investigating Machine Learning Algorithms for Stroke Occurrence Prediction

Stroke is the leading cause of death and the principal cause of long term disability. Accurate prediction of stroke is highly valuable for early intervention of treatment. In this study, six (6) machine learning (ML) algorithms namely: Random Forest (RF) classifier, Decision Tree (DT) classifier, K-Nearest Neighbour (KNN) classifier, Support Vector Classifier (SVC), Logistic Regression (LR) and Stacking Classifier (SC) were trained on 10 stroke risk factors to determine the most precise model for predicting the risk of stroke occurrence. The primary contribution of this work is the development of a stacking method that achieves high performance, as measured by various metrics such as Area under Curve (AUC), precision, recall, F1-score, and accuracy. The experimental results indicate that the stacking classification outperforms other methods, with an AUC of 98.80%, F1-score of 95.18%, precision of 95.08%, recall of 95.41%, and accuracy of 95.25%. The results revealed that the stacking classifier achieves a high performance and outperforms the other methods. With the rapid evolution of machine learning, the clinical professionals, and decision-makers can use the established models to assess the corresponding risk likelihood.