Heart Disease Diagnosis: Performance Evaluation of Supervised Machine Learning and Feature Selection Techniques

Abstract

Heart diseases are the leading cause of deaths nowadays. Due to the high severity of the problem, it has attracted several researchers around the globe. Researchers have considered the heart diagnosis as a classification problem where meaningful patterns are detected using data mining techniques. This paper presents an evaluation of various supervised learning algorithms and feature selection techniques for heart disease prediction. The performance of six machine learning classifiers (Naïve Bayes, Decision Tree, Logistic Regression, Random Forest, Support Vector Machine, k-Nearest Neighbour) and five feature selection techniques (Chi-Square, Gain Ratio, Information Gain, One-R and RELIEF) have been investigated on the benchmark dataset obtained from UCI Machine Learning Repository, Cleveland. The experimental results show that machine learning classifiers can achieve prediction accuracy up to 82.81% for heart disease prediction. The feature selection techniques further improve the classification performance and achieve prediction accuracy up to 83.41%.