Classification models for heart disease prediction using feature selection and PCA

Abstract

The prediction of cardiac disease helps practitioners make more accurate decisions regarding patients' health. Therefore, the use of machine learning (ML) is a solution to reduce and understand the symptoms related to heart disease. The aim of this work is the proposal of a dimensionality reduction method and finding features of heart disease by applying a feature selection technique. The information used for this analysis was obtained from the UCI Machine Learning Repository called Heart Disease. The dataset contains 74 features and a label that we validated by six ML classifiers. Chi-square and principal component analysis (CHI-PCA) with random forests (RF) had the highest accuracy, with 98.7% for Cleveland, 99.0% for Hungarian, and 99.4% for Cleveland-Hungarian (CH) datasets. From the analysis, ChiSqSelector derived features of anatomical and physiological relevance, such as cholesterol, highest heart rate, chest pain, features related to ST depression, and heart vessels. The experimental results proved that the combination of chi-square with PCA obtains greater performance in most classifiers. The usage of PCA directly from the raw data computed lower results and would require greater dimensionality to improve the results.

Keywords

• Machine learning
• Heart disease
• Apache spark
• PCA
• Feature selection