A support vector machine for predicting spontaneous termination of paroxysmal atrial fibrillation episodes

The aim of this work is to predict the spontaneous termination of atrial fibrillation (AF) episodes. The database includes three record groups: non-terminating AF (N), AF that terminates one minute after recording end (S), and AF that terminates immediately after recording end (T). A first goal consisted on separating N from T group records (event 1), and a second, for separating S from T records (event 2). A Support Vector Machine was used for the classification problem. For event 1, four indexes were extracted: the atrial fibrillatory frequency (AFF) and the mean, standard deviation, and approximate entropy of RR intervals. For event 2, the AFF, the energy of the 3-7 Hz and 7-11 Hz bands, from the ten and five final seconds of the records, were used. The groups were divided in two sets: learning and test. For event 1, a 100% in learning, and 86.66% in test set were correctly classified. For the event 2, we classified 100% in the learning, and 80% in the test set.