Univariate and Multivariate Filter Feature Selection for Heart Disease Classification

Feature selection (FS) is a data preprocessing task that can be applied before the classification phase, and aims at improving the performance and interpretability of classifiers by finding only a few highly informative features. The present study aims at evaluating and comparing the performances of six univariate and two multivariate filter FS techniques for heart disease classification. The FS techniques were evaluated with two white-box and two black-box classification techniques using five heart disease datasets. Furthermore, this study deals with the setting of the hyperparameters’ values of the four classifiers. This study evaluates 600 variants of classifiers. Results show that white-box classification techniques such as K-Nearest Neighbors and Decision Trees can be very competitive with black-box ones when hyperparameters’ optimization and feature selection were applied. © 2022 Institute of Information Science. All rights reserved.