Performance Evaluation of SVM and FFNN Classifier for Cardiac Arrhythmias Classification using Wavelet Features

The electrocardiographic signal represents the electrical activity of the heart and one of its applications is diagnosis of cardiac arrhythmias. However there are complications against analysis of a long record. Therefore, certain automatic diagnosis methods are needed. In this paper, an algorithm using a combination of wavelet features and SVM classifier is proposed. For this purpose, the signal noises are removed initially by digital filtering and then by wavelet transform. Then, the R waves are extracted through Pan-Tompkins algorithm. Afterwards, the features of each heartbeat are determined by discrete wavelet transform (DWT) and feature space dimension are reduced by PCA transform. Classification is done through SVM with different kernels. MIT-BIH arrhythmia database and MATLAB Software are used for evaluation of the proposed method in comparison with well-known Feed-Forward Neural Network (FFNN). Five groups including normal heartbeats (N), left bundle branch block (LBBB), right bundle branch block (RBBB), atrial premature beats (APB), and paced beat (PB) are classified through SVM with accuracy of 95.68% in the best case while the obtained classification accuracy for FFNN method is 90.30%. Results show that the proposed algorithm can detect cardiac arrhythmias more effectively.