A comparative analysis of machine learning classifiers for stroke prediction: A predictive analytics approach

Abstract

Stroke is the third leading cause of death in the world. It is a dangerous health disorder caused by the interruption of the blood flow to the brain, resulting in severe illness, disability, or death. An accurate prediction of stroke is necessary for the early stage of treatment and overcoming the mortality rate. This study proposes a machine learning approach to diagnose stroke with imbalanced data more accurately. Random Over Sampling (ROS) technique has been used in this work to balance the data. Eleven classifiers, including Support Vector Machine, Random Forest, K-nearest Neighbor, Decision Tree, Naïve Bayes, Voting Classifier, AdaBoost, Gradient Boosting, Multi-Layer Perception, and Nearest Centroid, are analyzed in this study. Ten classifiers show more than 90% accurate results before balancing the data and four classifiers display more than 96% accurate results after data-balancing using the oversampling method. The Hyperparameter tuning and cross-validation are performed in each model to enhance the results. Moreover, Accuracy, F1-Measure, Precision, and Recall are used to measure the performance of machine learning models. The results show the Support Vector Machine has the highest accuracy of 99.99%, with recall values of 99.99%, precision values of 99.99%, and F1-measure of 99.99%. Random Forest achieves the second-highest accuracy of 99.87%, with a 0.001% error. In addition, a user-friendly web app and a user-friendly mobile app are built based on the most accurate model.