Optimization of Machine Learning Algorithms with Bagging and AdaBoost Methods for Stroke Disease Prediction

Abstract

Stroke is an acute neurologic disorder of blood vessels in brain due to blockage of blood flow to the brain resulting in less oxygen. Stroke remains one of the leading causes of death worldwide. Therefore, the development of Machine Learning is expected to help health professionals make early predictions of stroke disease. The purpose of this study is to compare the performance results of stroke classification modeling using Bagging and AdaBoost methods in Machine Learning algorithms (Naïve Bayes, SVM, Decision Tree, and KNN) using Stroke Prediction Dataset from Kaggle. The results show that Machine Learning algorithm that has the best performance is Decision Tree with 91% accuracy, followed by KNN, Naïve Bayes, and finally SVM. Optimization of Machine Learning algorithms with Bagging and AdaBoost only increases performance value of the Decision Tree algorithm, but does not increase performance value of other algorithms. The results of Decision Tree optimization with Bagging increased 1% accuracy and F1-score, as well as 4% precision in the missing value deleted scenario. Furthermore, in the missing value scenario using mean value increases 1% F1-score and 4% precision. While the results of Decision Tree optimization with AdaBoost increase 2% recall and 1% F1-score in the missing value deleted scenario. Then in the missing value scenario using mean value has same performance as without optimization. The conclusion is that the application of Bagging and AdaBoost methods only increases the performance value of Decision Tree algorithm, but the increase is still insignificant.

Keywords: Stroke Disease, Machine Learning, Bagging, AdaBoost