Diagnosis of Heart Disease Using a Mixed Classifier

Abstract

Nowadays, a large amount of medical information can be used to help for medical diagnosis. Especially in computer science, many researchers attempt to use data mining techniques for analyzing large amount of data to discover pattern or knowledge from those data. In this study, a classification technique with supervised learning is considered. From the preliminary study, each classification technique yields different performance. With this reason, a new classification model that combines the advantage of individual classifiers is proposed to achieve higher performance. The proposed model consists of two classification stages. Three techniques composed of Support Vector Machine (SVM), K-Nearest Neighbor (KNN) method and Naïve Bayes approach are applied to an instance separately for the first stage. Then, the results from the techniques will be fed into multi-layer perceptron with back-propagation learning for the second classification stage. Moreover, heart disease, one of the leading causes of death in Thailand, has been selected for diagnosis in this study. A dataset of the disease consists of two classes: risk and non-risk. From the experimental results, the proposed model mixed from three techniques outperforms the existing classifiers including the individual techniques.