A new method of optimizing prototypes for nearest neighbor classifiers using a multi-layer network

Abstract This paper proposes a new method of optimizing the prototypes for a nearest neighbor classifier which uses a network with a hidden layer. After training, the neural network is mapped back to a nearest neighbor classifier with optimized prototypes. The main characteristic of the present method is that both the trained neural network and the mapped nearest classifier have the same recognition performance. Experimental results show that this method outperforms the method recently proposed by Yan (1994).