Comparison of Six Machine Learning Algorithms for Stroke Risk Estimation

Abstract

Every year there are millions of people suffer from stroke around the world. In order to timely prevent stroke and effectively reduce the damage caused by stroke, this paper uses a variety of machine learning algorithms to predict stroke. The dataset used in this work contains 12 variables that provide some demographic, health and lifestyle information such as age, gender, hypertension, heart disease, marital status. First of all, this work analyzes dataset of 5110 patients by visualizing the distributions of numerical features and investigating the correlation between these different features. Six different algorithms, including decision tree, KNN, logistic regression, SVM, Native Bayes and random forest) are used to build models. Finally, use accuracy and classification report to evaluate these models performance. Among these models, the SVM classifier had the highest accuracy at 99.97%. Therefore, the SVM classifier model could be chosen to predict whether a person have a stroke and quickly screen out possible patients from the population.