Detecção de artefatos de arritmia utilizando máquinas de vetores de suporte e coeficientes de energia wavelet

Abstract

Cardiovascular diseases are responsible for about one-third of annual deaths, being this number even higher in low-income populations. The development of computational methods to aid the diagnosis of these conditions can facilitate the discovery of it still in its initial phase, by reducing the specialized labor time required to analyze the data provided by exams. Algorithms capable of identifying patterns in large amounts of data are possible using techniques of Machine Learning, which gain more and more space by the growing amount of information generated and stored by devices worldwide. In this way, a set of Wavelet energy coefficients was used to conduct training for binary SVM classifiers, expecting to obtain a tool capable of detect arrhythmia artifacts on electrocardiogram signals. The project explores different non-binary SVM classification techniques to classify four distinct classes of electrocardiogram signals, which one refers to healthy cardiac sections, and the other 3 assigned to different arrhythmia categories. With an average recall of 84.6%, the Binary Tree of SVM presents a promising answer to these particular problem, although it may not be the most appropriate in a more practical setting.