Multistage Classification of Arrhythmia and Atrial Fibrillation on Long-term Heart Rate Variability

Abstract

This article proposes a Multi-Stage Heart Rate Variability Classification (MSHVC) system to diagnose Normal, Arrhythmia (AR) and Atrial Fibrillation (AF) for Long-Term ECG analysis. The MSHVC methodology comprises of ECG pre-processing, QRS detection, HRV feature extraction, statistical analysis and classification. The frequency-domain, time-domain, and geometrical-domain HRV features were extracted and accuracy was improved using Analysis of Variance (ANOVA) test. Artificial Neural Network (ANN), Support Vector Machine (SVM), and k-Nearest Neighbor (kNN) classifiers are utilized at various levels to classify in two-stage and three-stage classification. The MSHVC classification system demonstrates a higher accuracy compared to that of other state of the art methods when applied to MIT/ BIH Normal Sinus Rhythm (NSR), MIT/ BIH Arrhythmia (AR) and MIT/ BIH Atrial Fibrillation (AF) databases. To classify normal ECG from abnormal, proposed system attained maximum overall accuracy of 98.36% by ANN at 2-stage classification. Multi-stage classification of abnormal ECG further divided into AR and AF attains 99% of overall accuracy by ANN after statistical analysis