Dynamic ECG features for atrial fibrillation recognition

This study found that 4 s is the most appropriate windowing length, using two features (ω and u) for AF detection with an accuracy of 95.3%. Moreover, the pattern recognition learning machine uses an ANN with 10-fold cross validation based on DS.