Deciding Optimal Number of Exemplars for Designing an ECG Pattern Classifier Using MLP

ECG pattern recognition using artificial neural networks is now an established paradigm. Diagnostic systems derive robustness, reliability and speed because of the automatic pattern classifiers. However, a common problem associated with these types of classifiers is to decide the optimal number of exemplars. This paper attempts to find an optimal number of exemplars required for training a multilayer perceptron with acceptable accuracy. Extensive experimentation suggests a figure of 200. Although this figure is specific for multilayer perceptron based classifier, experimentation on similar lines can be performed for other ANN topologies.