BUILDING A ROBUST NEAREST NEIGHBOR CLASSIFIER CONTAINING ONLY A SMALL NUMBER OF PROTOTYPES

Publisher Website
Google Scholar
Abstract
A robust nearest neighbor classifier with only a small number of prototypes can be designed if the prototypes contain distinctive features that have a high classification power. We present in this paper an interset and intraset distance based method for reducing the number of prototypes and a neural network based technique for optimizing the prototypes of a nearest neighbor classifier. It is shown that the prototypes can be modified with a learning algorithm in such a way that useful features are created or enhanced to improve the matching capability of the classifier.

This publication has 0 references indexed in Scilit: