ECG Multiclass Classification Using Temporal Features, Information Measures, and Non-linear Parameters

The classification of electrocardiograms signals into multiple classes according to health or the type of disease of interest is a matter widely explored by the scientific community. This work proposes a mixed feature space formed by non-linear parameters, informational measures, and temporal characteristics of the ECG to train a classification model based on support vector machines (SVM) with optimized hyperparameters. The good performance of SVM applied to classify ECG signals has been extensively proved in the literature. In the present study, previous to the classification task, a principal component analysis was applied to determine the importance to explain the model of the variables in the proposed feature space. The quality of the classifier was analyzed using a 10-fold cross-validation to compute the balanced accuracy. The results showed that the proposal of using a feature space with different types of variables encourages this research field.