Comparison of Atrial Fibrillation Detection Performance Using Decision Trees, SVM and Artificial Neural Network

Atrial fibrillation (AFib) is a supraventricular tachyarrhythmia characterized by uncoordinated atrial activation and ineffective atrial contraction. AFib affects 1–2% of the general population, its prevalence increases with age and may remain long undiagnosed. Due to costs of hospitalization and treatment related to AFib and increasing prevalence, effective methods of detecting atrial fibrillation are needed. In this study we compared AFib classification using support vector machine (SVM), artificial neural network (ANN) and binary decision trees on 10 ECG signals. We considered 8 parameters associated with RR intervals: mean RR, SDNN, RMSSD, PLF, PHF, LF/HF, SD1 and SD2. In this comparison the best performing AFib classifier was binary decision tree with maximum number of splits equal to 100 and the worst case was SVM classifier with medium Gaussian kernel and using only one feature. Achieved result should encourage further studies using decision trees.