Clustering of hrv data obtained from normal sinus rhythm ecg records of paf and non-paf subjects using a new ensemble method

Atrial fibrillation (AF) is one of the most common diseases in which the atria cannot completely push the blood to ventricles and therefore clot formation could occur which may contribute to serious health problems. In most cases, AF starts with short episodes and progresses to longer and developed form over time. With respect to the acceleration of the number of patients, certain criteria have been used to determine the disease in order to start necessary treatment. Paroxysmal AF is a type of AF in which the episodes could last from minutes to days and ends by itself. If PAF condi-tion continues, it could be converted to Persistence AF. Therefore, it is very im-portant to determine PAF patients and treat them accordingly. The determination of PAF condition is very easy by recording the ECG of the subjects during an episode. However, since PAF episodes mostly last for a short time period, it is very difficult to record the ECG of the subjects during the arrhythmic event. Therefore, a system that would be able to detect PAF patients based on ECG records taken during non-arrhythmic time periods would be very beneficial. The aim of this study is to analyze the structure of the data obtained from arrhythmia-free ECG records of PAF and non-PAF subjects. In other words, the separability of the two types of data is to be investigated. In this study, the dataset which contains HRV features obtained from normal sinus rhythm (NSR) ECG records of non-PAF and PAF subjects is aimed to be clustered by using two most famous unsupervised classification methods; K-Means and Fuzzy C-Means algorithms. The obtained results show that there is a sig-nificant overlap between the two types and thus there is a need for a good classifier.