Lempel–Ziv complexity and Shannon entropy-based support vector clustering of ECG signals

Abstract

This work presents a Support Vector Machine (SVM)-based clustering method to cluster normal and pathological ECG signals on a Lempel-Ziv (LZ) complexity and Shannon entropy (SE) space. One normal ECG and three ECG signals with arrhythmic processes were selected from the MIT-BIH Arrhythmia Database and, those were processed to remove muscle and breathing noise, electrode motion artifacts, power line interference and DC offset. Each ECG signal was divided into 35 four-second segments. Training Input data to the SVM-based clustering machine were obtained by applying the LZ complexity algorithm and SE to each 35 segments of ECG signals. SVC machine was capable to separate the ECG signals (each signal represents a group) in four clusters (with an accuracy of 95.7 %) according to the four different ECG records chosen for this study. Keywords: support vector clustering, ECG signal, arrhythmia, Lempel-Z