A COMPARATIVE EVALUATION OF FILTER-BASED FEATURE SELECTION TECHNIQUES FOR HEART DISEASE CLASSIFICATION

ABSTRACT

Cardiovascular diseases (CVDs) are the primary cause of mortality worldwide. As such, the need for effective early prediction techniques is paramount. Machine Learning (ML) algorithms can be used to predict heart disease; however, their effectiveness greatly depends on the quality of the input features. This study investigates the use of different filter-based feature selection techniques when applied alongside popular classifiers in predicting heart disease. Four different filter-based feature selection techniques, namely Chi-Square, ANOVA F-Test, Mutual Information (MI), and Information Gain (IG), are tested against five classifiers, Decision Tree (DT), Support Vector Machines (SVM), Logistic Regression (LR), Random Forest (RF), and eXtreme Gradient Boosting (XGB). The experiments were carried out using the Cleveland Heart Disease dataset, which consists of 303 instances and 14 attributes. Results showed that Chi-Square and ANOVA F-Test outperform MI and IG. Chi-Square with LR produced the best classification result (Accuracy=86.36%). Also, Chi-Square with SVM produced the best AUC value of 0.9362. Hence the results from this study are consistent with other benchmark studies.

Keywords: Heart disease prediction, Feature Filter methods, Chi-square test, ANOVA F-test, Machine learning classification.