Application of Transition Patterns in the Classification of Electrocardiograms

The aim of this research work was to test the ability of transitions between ordinal patterns to classify one dimensional signals from single-channel ECGs, obtained from the PhysioNet database of healthy individuals from those with a diagnosed pathology. For this purpose, the transition frequencies between ordinal patterns were used to build the feature space to train three classification algorithms (KNN, SVM and RF). In order to select the most appropriate classification algorithm, quality parameters were calculated (accuracy, area under the ROC curve, and F1 score), added to the time spent in the calculations. The result obtained is at the same level as those reported in the available literature, showing that with the database used, ordinal patterns transitions are a recommendable option to differentiate single-channel recordings of normal ECGs from those with a diagnosed pathology. Due to the conceptual simplicity of the present proposal, it is promissory for implementation in mobile and IoT devices.