Abstract
Stroke is considered one of the most universal diseases. To understand this disease in medical sciences, large and complex datasets are collected and analyzed. The analysis of this data has been recognized as a big challenge in modern life. Therefore, there is a need to find effective techniques to deal with such huge datasets. To predict the stroke, researchers study the impact of various risk factors on the onset of stroke in an individual. Then, they use the analyzed data to predict the probability of stroke occurrence using machine learning algorithms and techniques like neural networks, decision tree, random forest, linear regression, etc. In this research, recent studies that proposed stroke prediction frameworks using data mining approaches have been reviewed, and a new hybrid framework is proposed to predict stroke disease using two main steps, clustering and classification. Enhanced Hierarchal Clustering is applied on the dataset, then five classifiers are evaluated and compared. The used algorithms are Logistic Regression, Random Forest, Support Vector Machine, Neural Network and XGBoost. All of them show good results according to accuracy and AUC. The best result which is (97%) has been achieved by Random Forest classifier.