Cardiovascular diseases are one of the most common causes of mortality in the world. Therefore, early detection and treatment of arrhythmia remains a major challenge in cardiac care. Cardiac activity is one of the most significant tools to determine the status of the patient, which is primarily reflected by a physiological signal also known as Electrocardiogram (ECG). A number of algorithms have been proposed for the automatic recognition of cardiac arrhythmia. We propose in this paper a modified architecture of the fuzzy adaptive resonance theory - supervised predictive mapping neural network (fuzzy ARTMAP) algorithm for the diagnosis of cardiac rhythm; we introduced a statistical and a probabilistic approach in the neurons of the hidden layer, requiring only one epoch for learning. This showed a significant improvement in the obtained results, which were validated using ECG signals of different patients from the 'MIT-BIH arrhythmia databases'. Our results therefore demonstrate the effectiveness of this modified classifier compared to those already reported in the literature.