A Comparative Study of Heart Disease Prediction Using Machine Learning Techniques

Abstract

Since early diagnosis of heart disease plays an important role in the survival of individuals, the classification method, which is one of the machine learning techniques, is used in the diagnosis of heart disease and successful results are obtained. In this study, data from the 2020 survey presented by the Centers for Disease Control and Prevention (CDC) on the Behavioral Risk Factor Surveillance System (BRFSS) were analyzed using 8 different machine learning classification methods. These methods are Logistic Regression (LR), Support Vector Machines (SVM), Naive Bayes (NB), k-Nearest Neighbor (k-NN), Decision Tree (DT), Adaboost, Multilayer Perceptron (MLP), and XGBoost (XGB). However, the nominal dependent variable, which is the heart disease, is unbalancedly distributed in the presented data. To overcome this problem, the dependent variable was stabilized using the Synthetic Minority Oversampling Technique Tomek Links (SMOTE -Tomek Link) method before applying the classification methods. The imbalance in the data have been eliminated with the production of synthetic data. In order to make powerful and unbiased estimates, an outlier analysis was performed using classical statistical methods and the data were divided into outliers and non-outliers. During the classification process, 10-fold cross-validation was performed to obtain more stable results and to better compare the performance of the methods. Thus, the XGB algorithm achieved an accuracy rate of 89% for the non-outlier data and 84.6% for the outlier data in both the early diagnosis of disease and the detection of patterns in the diagnosis of heart disease. However, the k-NN algorithm achieved an accuracy rate of 85.6% in the non-outlier data and 81% in the outlier data in a very short time. Therefore, XGB and k-NN algorithms will be performed for early disease diagnosis and pattern discovery in heart disease diagnosis.

Topics & Concepts