Segment Labeling Method for ML-based AFIB Detection

Abstract

Atrial Fibrillation is one of the most common and mortal types of heart rhythm problems - arrhythmias. Therefore, early and accurate detection is important in detecting heart diseases and prescribing appropriate treatment therapy. Developing a technology of this kind is of pivotal importance and a challenging problem for noninvasive tools for patient monitoring and analysis. Electrocardiography provides comprehensive information that can be efficiently used in the management of the patients heart health. Detecting and classifying episodes of the different types of heart diseases is a subject of continuous research and immediately with new technological advances. Machine learning methods emerged as frequently used technology recently and become acknowledged for their relevance and results in this field. Developing an effective model for detecting and classifying Atrial Fibrillation in ECG recordings requires the right data and adequate feature engineering. For this purpose we propose two methods, majority and pure segment labeling method used in the performed segmentation for feature engineering using the most popular ECG database and by integrating them in three machine learning algorithms, Support Vector Machines, Decision Trees and Random Fores. The research concluded that the majority method trained on the Random Forest algorithm gives the highest results in the defined research space.