Effect of feature fusion for discrimination of cardiac pathology

Automatic diagnosis of electrocardiogram (ECG) signal is significant for timely and accurate diagnosis of heart diseases like arrhythmia. Several researchers have proposed different methods in last two decades. In this work we have employed a global ECG beat classification approach based on transformed features like discrete cosine transform (DCT) and discrete wavelet transform (DWT) rather than conventional time interval or morphology features to classify six different types of ECG beats. It is observed that a few features from the ranking of combined DCT and DWT features perform better than the individual feature sets on this problem. The experimental results are validated on large data sets taken from MIT/BIH arrhythmia database by employing two kernel classifiers, namely support vector machine (SVM) and vector valued regularized kernel function approximation (WRKFA), and a single layer feedforward neural network (SLFN) classifier known as extreme learning machine (ELM). Experimental results indicate the that six different types of beats can be classified with an accuracy of 96.83% which is probably the best figure compared to the results reported in literature so far on classifying ECG beats by global classification approach.