Classification of ECG complexes using self-organizing CMAC

In this study, we apply a self-organizing cerebellar model articulation controller (SOCMAC) network to design an ECG classifier by observing the QRS complex of each heartbeat. The SOCMAC network is an unsupervised learning method, which combines Kohonen’s self-organizing map into CMAC. In order to achieve the goal of real-time classification, the data collected are divided into two datasets: the training set for the unsupervised learning of the ECG classifier and the testing set for the real-time classification. In the learning stage, with the help of the proposed performance index, we search for optimal parameters of the network that achieve the best performance. Then the well-trained classifier is used, with the optimal parameters, to classify the testing set. Tested with all the 48 recordings from the MIT/BIH arrhythmia database, the proposed method achieves a classification accuracy of 98.21%, which is comparable to the existing results.