Recognition of Atrial Fibrilation Episodes in Heart Rate Variability Signals Using a Machine Learning Approach

Atrial fibrillation (AF) is the most common heart arrhythmia. Asymptomatic (silent) AF may be recognized during long term monitoring of the heart rate (HR) variability. The HR variability features are widely used for detection of AF. Automated classification of heart beats into AF and non-AF presented in this paper was carried out with a help of the Lagrangian Support Vector Machine. The classifier input vector included five beat-to-beat interval measures, seven adult’s HR variability parameters, and four features taken from the analysis of the fetal heart rate as being characterized by high sensitivity to changes in subsequent intervals. The performance of the improved AF detection method was examined using the MIT-BIH Atrial Fibrillation Database, which includes 25 ten-hour ECG recordings. Results obtained during the classifier testing phase showed the sensitivity 95.91%, specificity 92.59%, positive predictive value 90.56%, negative predictive value 96.83%, and classification accuracy 94.00%.