An analysis of how training data complexity affects the nearest neighbor classifiers