Classification of arrhythmia using machine learning techniques

Abstract

The electrocardiography (ECG) is a non-invasive routine to measure electri- cal activity of the heartbeat. Regular heartbeat activities are controlled by a complex set of electrical impulses. If these electrical impulses are interrupted or misguided the arrhythmia is occurred and it can be interpreted as a heart disease. The main aim of the thesis is design and test novel ECG classifi- cation approach to improve accuracy recent results. The testing of the pro- posed method will be divided into three experiments. First experiment will be devoted to Intra-Patient paradigm and will show what exactly classifier has suitable assumptions to further testing with more realistic scenario. Last two experiments will follow the Patient-Adapted and Inter-Patient paradigm to measure robustness of proposed method and also the proposed approach will be compared with state-of-the art results. In order to measure results of each experiment it will be used three following ECG databases: MIT-BIH Arrhythmia Database, Physionet Challenge 2017 a St.-Petersburg Institute of Cardiological Technics 12-lead Arrhythmia Database.