ABSTRACT
A new kernel-based modeling technique of Electrocardiogram (ECG) signal is presented in this work to discriminate normal and arrhythmia beats. ECG signals are characterized by time series modeling using Fast Regularized Kernel Function Approximation (FRKFA) technique. The characteristics parameters of the nonlinear regression models are considered as the features of ECG beats. The ability of these features to discriminate normal and arrhythmia beats are verified using Support Vector Machine (SVM) classifier in a global beat classification approach. The results are compared with the existing linear Autoregressive (AR) signal modeling technique. The test results on large data sets show that the performance of kernel-based modeling technique achieves an accuracy as high as 95.43% and this shows an improvement in performance by more than 11% to that of linear AR modeling technique to discriminate normal and arrhythmia ECG beats.
