A Robust Methodology for Stroke Disease Prediction using a Hard Voting Classifier
Nouralhuda Ali Abdulsamad, Ali Adil Yassin, Zaid Ameen Abduljabbar, Mohammed S. Hashim, Vincent Omollo Nyangaresi · Engineering Technology & Applied Science Research · 2025
The vast majority of strokes are caused by an unexpected occlusion of the blood vessels that supply the brain and the heart arteries. Early detection of the many warning symptoms of stroke can help reduce the severity of the stroke and save the patient's life. Although researchers have proposed a variety of diagnostic methods to detect this disease, the methods currently in use still need further improvement. In this paper, we propose an effective methodology that utilizes a hard voting classifier based on three Machine Learning (ML) models, namely, Random Forest (RF), K-Nearest Neighbors (KNN), and Extra Trees Classifier (ETC). First, a series of data quality improvement procedures were performed using the Synthetic Minority Oversampling Technique (SMOTE) approach for data balancing to ensure an unbiased training process without majority class dominance. Next, we divided the dataset into two parts, a training part and a testing part, and these data were fed to the models used. In the last phase, we implemented four ML algorithms to evaluate their effectiveness and then selected the three most effective models for integration into our proposed hard voting classifier. The hard voting outperformed the results of modern studies with an accuracy of 97.48%, a precision of 0.9802, a recall of 0.9691, and an F1 score of 0.9747. Furthermore, we applied K-fold cross-validation (K=10), which systematically partitions the dataset into multiple subsets, preventing overfitting and providing a robust estimate of model performance across different data splits, where a mean accuracy of 97.1% was achieved.