Recognition and classification system of arrhythmia using ensemble of neural networks

The paper presents different methods of combining many neural classifiers into one ensemble system for recognition and classification of arrhythmia. Majority and weighted voting, Kullback–Leibler divergence and modified Bayes methods will be presented and compared. The numerical experiments will be performed for the problems concerning the recognition of different types of arrhythmia on the basis of ECG waveforms of MIT BIH AD. The results have shown that combining many classifiers into one classification system brings important benefits and result in significant improvement of accuracy.