Arrhythmia Recognition by a Deep Spectral Autoencoder

In this study, a scheme for the detection of cardiac arrhythmias as well as normal heartbeats is presented. The proposal is based on a deep autoencoder trained with different spectral representations, specifically: Autoregressive Burg, Co-variance, Yule-Walker, Fast Fourier Transform, Periodogram and Welch. The experiment was developed taking into account the
$K$
-fold cross-validation technique in order to have an objective frame of reference considering metrics from the literature. The quantitative results obtained suggest better recognition stability when the classifier was trained with the Autoregressive Burg spectral feature.