Selection of higher order subband features for ECG beat classification

Five levels of discrete wavelet transform are applied to decompose the ECG beat signal into six subband components. Higher order statistics proceeds to calculate valuable features from the three midband components. These features together with three RR interval-related features construct the primary feature set. A feature selection algorithm based on correlation coefficient and Fisher discriminality is then exploited to eliminate redundant features from the primary feature set. The feedforward backpropagation neural network (FFBNN) is employed as the classifier to justify the capacity of the method. The proposed method achieved an imposing ECG beat discrimination rate of more than 97.5%. By using the feature reduction method, the feature dimension can be readily reduced from 30 to 18 with negligible decrease in accuracy. Compared with other methods in the literature, the proposed method improves the sensitivities of most beat types, resulting in an elevated average accuracy. The results demonstrate the effectiveness and efficiency of the proposed method in ECG beat classification.