Feature Extraction on Multi-Channel ECG Signals using Daubechies Wavelet Algorithm

Abstract

Listen

Translate article

An electrocardiogram (ECG) is a recording of the heart’s rhythm and electrical activity over a period of time that can be obtained through an electrocardiography process. Through this process, the produced ECG can represent the condition of the heart whether it is healthy or not. Irregularity in heartbeat’s ECG signal causes several problems such as stroke and heart failure. Atrial fibrillation (AF) is a type of irregular heartbeat that is frequently found in a stroke patient. There have been studies exploring features of atrial fibrillation based on the ECG signal. However, most of them produce low accuracy due to the use of inappropriate features in identifying AF. To overcome the problem, this research proposes a study of feature extraction using Discrete Wavelet Transform (DWT) to increase accuracy in detecting AF. This research focused on analyzing both Daubechies Wavelet basis function and decomposition level of ECG signal as AF features. To classify the features either within AF or normal class, K-Nearest Neighbor (KNN) has been set up. Rigorous experiments have been conducted and show that Daubechies 2 (Db2) with decomposition level 6 produces the best performance of AF detection, i. e. accuracy 93.83%, sensitivity 96.77%, specificity 88.64%, 95.24% f1-score, and precision 93.75%.