Abstract
This paper proposes a computerised method to distinguish between normal and arrhythmia heartbeats. This is considered a significant problem as accurate and timely detection of cardiac arrhythmia can assist doctors to provide suitable medical attention to treat the ailment. The proposed scheme utilises cross-wavelet transform and library support vector machines (LIBSVMs) tools for investigation and classification of ECG signals. Feature extraction has been carried out from cross-wavelet spectrum (XWT) and cross-wavelet coherence spectrum (WTC). Support vector classifier with radial basis kernel is used to classify the heartbeats. This classification scheme is developed utilising a small training data-set and tested with a massive testing data-set to show the generalisation capability of the method. The performance of the LIBSVM classifier is also compared with three classifiers, probabilistic neural network (PNN), back propagation neural network (BPNN) and Elman’s recurrent neural network (ERNN). The proposed algorithm, when employed for 42 files corresponding to 97,461 beats of MIT/BIH arrhythmia database produces classification accuracy as high as 96.66%.