CLASSIFICATION OF HEARTBEAT SOUNDS TO DIAGNOSE CARDIAC DISORDER

Abstract

Cardiac disorder is one of the leading causes of death worldwide. There are around 18 million people died of cardiac disorder every year. Clinically, cardiac disorder detection can be diagnosed by a doctor listening to the heartbeat sound using a stethoscope. However, this method is time consuming and requires the patient to be present with the doctor physically. In this study, three machine learning models are proposed to classify cardiac disorder using heartbeat sound collected from three different devices: i-stethoscope iPhone application, digital stethoscope, and stemoscope. The three objectives in the study are: 1)remove anomalies from the heartbeat sound (audio format). 2) convert the heartbeat sound into audio features and 3) classify the converted data. The techniques used in this study include: wavelet decomposition method for data cleaning, ANOVA in feature selection and semi-supervised machine learning algorithm used to fill up null class value. All in all, eight features were extracted from the audio files. Finally, five types of machine learning algorithms were used to train the models. The best results obtained are 74% accuracy for I-stethoscope iPhone application and 73% for digital stethoscope and 72.13% for Stemoscope.