Combinações de redes neurais e discriminantes lineares para classificação de arritmias cardíacas
Natália Nagata, Renato Candido, Magno T. M. Silva

DOI: 10.14209/sbrt.2021.1570724172
Evento: XXXIX Simpósio Brasileiro de Telecomunicações e Processamento de Sinais (SBrT2021)
Keywords: Machine learning Neural networks Linear discriminant analysis Electrocardiogram
In this work, we use multilayer neural networks (MLP), recurrent neural networks, linear discriminant analysis (LDA) and combinations of these methods for automatic classification of cardiac arrhythmias. In order to obtain clinically realistic results for the diagnosis, patients records used during the training phase were not used in the test set. The results indicate that the combination of MLP with LDA presents a better performance compared to those of individual models. Furthermore, this combined scheme achieves classification metrics superior to those reported in the literature for the supraventricular ectopic beats (S) and the ventricular ectopic beats (V) arrhythmia classes.