Feature measurements of ECG beats based on statistical classifiers

In this paper, we have studied two statistical classifiers: Mahalanobis and Minimum distance based, with new features of ECG (electrocardiogram) beats. We have used third-order cumulant, wavelet entropy, and AR (auto-regressive) coefficients as features. On testing with the MIT/BIH Arrhytmia database, we observed a better performance for the Mahalanobis distance classifier. To compare the obtained figures with the results in the literature by using different techniques of beat recognition, we provide some figures. The comparison denotes the moderate rate of the proposed method, but it is really difficult to compare the results respect to the same type and numbers. The proposed method also has an advantage that the training computation time is lower than that of artificial neural network (ANN) based classifiers. Because the mentioned statistical classifiers use only one iteration for the training step to obtain the center of classes, whereas many iterations are used by an ANN for training.