期刊
PATTERN RECOGNITION
卷 42, 期 11, 页码 2979-2988出版社
ELSEVIER SCI LTD
DOI: 10.1016/j.patcog.2009.02.008
关键词
ECG; Pan Tompkins algorithm; Linear prediction; PCA; GMM; Chernoff bound; Bhattacharya bound
资金
- VECC, Kolkata
An automatic classifier for electrocardiogram (ECG) based cardiac abnormality detection using Gaussian mixture model (GMM) is presented here. In first stage, pre-processing that includes Fe-sampling, QRS detection, linear prediction (LP) model estimation, residual error signal computation and principal component analysis (PCA) has been used for registration of linearly independent ECG features. GMM is here used for classification based on the registered features in a two-class pattern classification problem Using 730 ECG segments from MIT-BIH Arrhythmia and European ST-T Ischemia datasets. A set of 12 features explaining 99.7% of the data variability is obtained using PCA from residual error signals for GMM based classification. Sixty percent of the data is used for training the classifier and 40% for validating. It is observed that the overall accuracy of the proposed strategy is 94.29%. As an advantage, it is also verified that Chernoff bound and Bhattacharya bounds lead to minimum error for GMM based classifier. In addition, a comparative study is done with the standard classification techniques with respect to its overall accuracy. (C) 2009 Elsevier Ltd. All rights reserved.
作者
我是这篇论文的作者
点击您的名字以认领此论文并将其添加到您的个人资料中。
推荐
暂无数据