Classifier Comparison for Stroke Prediction Ensembling SMOTE+ENN using Machine Learning Approach

Abstract

Abstract Stroke is one of the deadliest diseases found in the world which is the second major reason for mortality rate. Early detection of stroke can reduce the mortality due to stroke. Inorder to diagnose it earlier several machine learning techniques are being utilized. This proves that machine learning can also be used for disease prediction for various diseases. Supervised machine learning algorithms has been used for stroke prediction. Important feature responsible for stroke prediction has been done. Inorder to balance the dataset hybrid sampling technique of SMOTE + ENN has been performed so that the results are much appreciable. After sampling, machine learning models has been used for stroke prediction using Logistic Regression (LR), KNearest Neighbour (KNN), Naïve Bayes (NB), Decision Tree (DT), Support Vector Classifier (SVC) and Random Forest (RF). With all the models Random Forest achieved higher performance with accuracy of 99%, recall of 100%, precision of 98% and f-score of 99%. The proposed work also extends the use of different classifiers like Voting, Bagging and Stacking methods. Among all the proposed classifiers stacking provides higher performance with all metrics. The stacking classifer achieved accuracy of 100%, recall of 100%, precision of 99% and f-score of 100%. From the results, it is known that Random forest algorithm perform better with data sampling of SMOTE + ENN than other models.