Classiication with Learning K-nearest Neighbors

The nearest neighbor (NN) classiiers, especially the k-NN algorithm, are among the simplest and yet most eecient classiication rules and are widely used in practice. We introduce three adaptation rules that can be used in iterative training of a k-NN classiier. This is a novel approach both from the statistical pattern recognition and the supervised neural network learning points of view. The suggested learning rules resemble those of the well-known Learning Vector Quantization (LVQ) method, but at the same time the classiier utilizes the fact that increasing the number of samples that the classiication is based on leads to improved classiication accuracy. The performances of the suggested learning rules are compared with the usual k-NN rules and the LVQ1 algorithm.