Support Vector Machine Based Sub-Classification of Arrhythmia Using ECG Signal

Abstract

A method, based on support vector machine (SVM) and discrete wavelet transform (DWT), for automatic classification of arrhythmia using Electrocardiogram (ECG) is presented in this work. The detection of ailment like arrhythmia is troublesome due to inherited complexities and non-stationary nature of the ECG signal. The method uses SVM for classification and DWT for feature extraction which is crucial for accurate diagnoses. The subtle changes in the ECG are not accurately identified leading to incorrect feature extraction. Incorrect feature extraction in turn provides distorted feature space for SVM. To improve the same, DWT is used in this method, which provides sufficient resolution in both time and frequency domain, to detect the changes of the ECG signal. The different heart ailments which are detected by this method are Atrial Fibrillation, Malignant Ventricular, Normal Sinus Rhythm, Supraventricular arrhythmia. The algorithm gives an accuracy of 99.32% on MIT-BIH arrhythmia database, which is an improvement as compared to other algorithms. The proposed method provides a fairly less complex and more robust algorithm, which is very effective.