Performance Evaluation of SVM Classifiers for Atrial Fibrillation Detection

Abstract

Atrial fibrillation (AtrF) is described as uncoordinated atrial activity and inefficient atrial contraction, a supraventricular tachyarrhythmia. 1-2% of the general population suffers from AtrF, which is more common with older people but may go undiagnosed for a long time. Effective methods of identifying AtrF are required due to the increasing occurrence and rising hospitalization expenses and treatment associated with AtrF. In this study, 6000 ECG signals were used to evaluate AtrF classification using different types of Support Vector Machine (SVM) classifiers (Linear, Quadratic, Cubic, Fine Gaussian, Medium Gaussian, and Coarse Gaussian). Parameters used for feature extraction related to RR interval are (RR Interval) the interval between two consecutive R-waves of the ECG's QRS signal, Standard Deviation (SDRR) of normal RR intervals throughout the 24-hr of all normal RR intervals, Standard Deviation of the Average (SDANN) of all 5-min RR interval segments during the 24-hour recording period, (pNN50)  the percentage of the discrepancy between adjacent normal RR intervals, Root Mean Successive Square Difference (R-MSSD) between adjacent normal RR intervals. In this work, the quadric SVM classifier was the most proficient with 89.9% accuracy, 0.93 AUC, 97% sensitivity, and 61% specificity. While the least proficient classifier was the cubic SVM, with 61% accuracy, 0.84 AUC, 96% sensitivity, and 54% specificity).