Publishing house Radiotekhnika

"Publishing house Radiotekhnika":
scientific and technical literature.
Books and journals of publishing houses: IPRZHR, RS-PRESS, SCIENCE-PRESS

Тел.: +7 (495) 625-9241


An atrial arrhythmia classification using wavelet transform of the ECG and machine learning approach


M.V. Voitikova, A.P. Voitovich

Classification of atrial arrhythmias problem in cardiology can be solved by the machine learning approach. In case of atrial arrhythmia classification (atrial fibrillation - AF and atrial flutter - AFL) we offer the support vector machine (SVM) based classifier. The classification problem consists of: i) the extraction of the ECG fragments which correspond to electric activity of the atrium and definition of statistical signs of arrhythmia type, ii) the dimensionality reduction of the feature parameters space. iii) the training of the binary classifier by training set of ECG samples with known diagnosis and iv) the classification of the test samples of ECG. Initially 10 different features are extracted from the input ECG and HRV signals and a subset of these features is selected to train the SVM-based classifier. Components of the vector represent the energy distribution of the wavelet coefficients of the ECG (ECG fragments - intervals concluded between 2 consecutive R-waves) and the statistical parameters of heart rate variability (the average RR-interval and standard deviation). SVM constructs the set of hyperplanes in multi-dimensional features space with the best separation and largest distance to the nearest training data points of two classes – AF and AFL. We considered the RBF kernel as a kernel function for SVM and selected the best kernel parameters empirically. SVM classifier training for AF and AFL arrhythmias is considered finished at for maximum value of quality indicator of signals classification (in our case this value is equal 0.98).
Sept. 2, 2020
Aug. 27, 2020
June 24, 2020

© Издательство «РАДИОТЕХНИКА», 2004-2017            Тел.: (495) 625-9241                   Designed by [SWAP]Studio