Classificação de arritmias cardíacas em eletrocardiograma utilizando floresta de caminhos ótimos

Abstract

Currently in the world, millions of people die, victims of heart diseases, which in large part can be detected by analyzing signals of the electrocardiogram. This analysis involves the study of the signal corresponding to the arrhythmia studied and can be automated through machine learning. This work compares the Optimum Path Forest (OPF) classifier using 6 distance metrics, the Support Vector Machines classifier with radial basis function kernel (RBF-SVM) and the Bayesian classifier, applied to the problem of ECG arrhythmias classification. This is done using 6 feature extraction techniques and a methodology for separating sets, to avoid the interference of patient information in classification. The performance is evaluated in terms of accuracy, generalization, through specificity and sensitivity, and computational cost. Classification was done using 5 and 3 classes of arrhythmias. The OPF showed the best performance in terms of generalization, while the SVM-RBF had the highest accuracy rates. The training times of OPF were the lowest among the classifiers. In the test, the RBF-SVM classifier presented best computational cost.