Edited Nearest Neighbor Rule for Improving Neural Networks Classifications

Abstract

The quality and size of the training data sets is a critical stage on the ability of the artificial neural networks to generalize the characteristics of the training examples. Several approaches are focused to form training data sets by identification of border examples or core examples with the aim to improve the accuracy of network classification and generalization. However, a refinement of data sets by the elimination of outliers examples may increase the accuracy too. In this paper, we analyze the use of different editing schemes based on nearest neighbor rule on the most popular neural networks architectures.