A Comparison between Different Classifiers for Diagnoses of Atrial Fibrillation

• Published in: 2019 International Conference on Computer, Control, Electrical, and Electronics Engineering (ICCCEEE) pp. 1 - 6
• Main Authors: Mahmood, Inas Salah, Abdelrahman, Iman Abuel Maaly
• Format: Conference Proceeding
• Language: English
• Published: IEEE 01.09.2019
• Subjects: and AdaBoost ensemble classifier; Classification algorithms; Decision trees; Discrete Wavelet Transform (DWT); Electrocardiogram (ECG); Electrocardiography; Feature extraction; Heart beat; K-nearest neighbor Algorithm (KNN) classifier; support vector machine (SVM) classifier; Training
• DOI: 10.1109/ICCCEEE46830.2019.9071190

this study is a comparison study with the purpose to propose an approach for selecting the best classifier for diagnoses of atrial fibrillation (AF) in coronary heartbeats. The Physionet Computing in Cardiology Challenge 2017 turned into used as the data source for this study. Automatic ECG processing consists essentially of the detection and location of the signal characteristic points and is an important tool in the management of cardiac diseases. The most relevant task is the detection of the QRS complex after which a complete analysis and delineation of each beat can be obtained. In the preprocessing stage, the Discrete Wavelet Transform (DWT) is used for removing noise and tuning to the morphological characteristics of the waveform features. For feature extraction, a set of features that consists of both morphological and temporal features is extracted using DWT. A comparison study was conducted between five classifiers (Decision trees, Random forest, AdaBoost ensemble classifier, support vector machine (SVM) and K-nearest neighbor Algorithm (KNN)) to know which give the best diagnoses for each type of Arrhythmia. In this study we used four classes of coronary heart beats atrial fibrillation, normal, other rhythms or noise. Results show that the AdaBoost classifier gives 100 % Accuracy scores for all types of Arrhythmia in the training set. The AdaBoost algorithm obtained a mean improvement report for all classes in testing set (97.3% in Area under curve accuracy (AUC), 94.7% in classifier accuracy (CA), 96.7% in sensitivity (Recall), and of 98 % in positive predictive value (Precision)). Keywords: Electrocardiogram