Performance Analysis and Comparison of Heart Disease Prediction Models

Since cardiovascular diseases (CVDs) are the leading cause of death, it is significant for people to detect them early and take certain precautions. As a result, the paper performs the study of heart disease prediction with statistical models. In this study, the researcher analyzed a dataset from four different regions, including Cleveland, Hungarian, Switzerland, and Long Beach. Each of the regions is used as one testing dataset and the rest of the three regions are used as one training dataset, with a total of four sets of training and testing datasets. The goal is to predict heart disease with various variables using three main statistical techniques, lasso regression, random forest, and logistic regression (with feature selection). The paper will compare aspects such as accuracy, confusion matrix, and roc-auc for the models created. The results show that the models tested on the Hungarian dataset performed best overall, while the random forest algorithm achieved the highest prediction accuracy.