A Comparative Overview of Machine Learning Models for the Diagnosis of Heart Disease

Abstract

Abstract: Heart Disease is the leading cause of death in the United States, with one out of every 4 deaths being linked to it. A recent study revealed that 40% of the participating middle-age adults had coronary heart disease without being aware of it, making early-stage diagnosis extremely important. 1To combat human error, Machine Learning has become increasingly involved in the process of medical diagnosis, with a standard ML Algorithm outperforming 72% of doctors2. Moreover, researchers have found that 5% of patients in the USA are annually misdiagnosed, which can be potentially fatal. Several Algorithms have been tested to find the optimal accuracy of heart disease diagnosis, with Support-Vector Machines (SVM) and Decision Trees being the most accurate. This paper further introduces SVM Ensembles and Random Forests as a result of Ensemble Learning often being more accurate than single models due to reduced bias and overfitting, as found by (Opitz and Maclin, 1999). We evaluate each of the four models to find the most accurate model for medical diagnosis, particularly detection of heart disease. We then use this to calculate the best way to lower a patient's probability of developing heart disease in the next five years, analysing how simple lifestyle changes can make a significant difference. The results portray that the SVM Ensemble is the most accurate, depicting its potential in medical diagnosis.