FILTER-BASED FEATURE SELECTION TECHNIQUES ON HEART DISEASE PREDICTION

Abstract

One of the predominant pre-processing steps used for development of robust diagnostic model in data mining is feature selection. The technique removes redundant features thereby enhancing the accuracy of computational intelligence. Building a model with irrelevant features may affect the accuracy or result into more memory space usage during the training and classification stage. This paper investigates the influence of filter-based techniques of feature selection on the heart disease diagnosis model. Three filter techniques: Relief-F, Information Gain and Principal Component Analysis were applied on the heart disease dataset. Some selected machine learning algorithms: Artificial Neural Network (ANN), C4.5 Decision Tree, Support Vector Machine (SVM), Naïve Bayes (NB) and K-Nearest Neighhood (KNN) were employed to classify the reduced features. Results from experimental analysis based on accuracy, precision and recall showed the effect of filter techniques on heart disease diagnosis.