Heart Disease Prediction Using Machine Learning

Abstract

Heart disease is the leading cause of death worldwide today; therefore, it is important that early detection of heart disease occurs as this increases available treatments and survival rates of patients. This paper reports on an experiment utilizing machine learning to predict heart disease from patient data. The UCI Cleveland Heart Disease Data Set was used along with six different supervised machine learning algorithms (Logistic Regression, Decision Trees, Random Forest, Support Vector Machines, Gradient Boosting {XGBoost} and Artificial Neural Networks {ANNs}) in addition to the evaluation of overall accuracy and area under the receiver operating characteristic curve (ROC AUC). Results demonstrated that Random Forest had the highest level of accuracy (89%) in predicting heart disease and the highest ROC AUC score (0.92). The results of a feature importance analysis identified a number of features as the most predictive regarding a diagnosis of heart disease, they are: the type of chest pain experienced by the patient, the total cholesterol level, the age of the patient, the rate at which the patient's heart can achieve its maximum heart rate, and the presence of exercise-related angina. This research demonstrates the strong potential for machine learning techniques to help facilitate the early diagnosis of heart disease, especially in lowresource countries such as Bangladesh where medical technology resources are limited.