Learning Classifier Methods in Prediction of Heart Disease

Abstract

Cardiovascular Disease (CVD) refers to any condition involving narrow or blocked blood vessels, which can lead to a heart attack, chest pain or stroke. CVD is a chronic illness and the leading cause of death for both men and women. CVD can attack a person instantly resulting in high healthcare costs and, in some cases, can result in death. A serious and important challenge facing medical practitioners is the ability to accurately diagnose patients with CVD early on. In recent years, medical practitioners have sought the help of computer scientists in order to apply advanced data mining techniques, which can facilitate decision support and help accurately diagnose CVD soon. In this thesis, three data mining techniques are evaluated for their accuracy in predicting CVD. The techniques implemented and analyzed are Logistical Regression, Naive Bayes and Artificial Neural Networks using Multi-Layer Perceptron (MLP). Results show that Logistic regression predicted the presence of CVD with an accuracy of 88.6%, while Naive Bayes predicted CVD with an accuracy rate of 83% and Neural Networks with an accuracy of 80%.