Proposal of asymmetric multi-classifier of arrhythmias

This paper presents a new methodology for the development of multi-classifiers SVM with One-Againts-One (OAO), which allows each node to use different features or attributes to differentiate each pair of classes, called asymmetric OAO-SVM. We evaluated this method by developing a classification system to identify four types of arrhythmias (Atrial Fibrillation, Atrial Flutter, Supraventricular Tachyarrhythmia and Ventricular Tachycardia) and Normal ECG, using nonlinear characteristics such as Shannon entropy and Lempel-Ziv complexity.This method presents a positive prediction of 90.72% which represent an improve with respect a typical multi-classifier OAO-SVM.