Data reduction for instance-based learning using entropy-based partitioning

Abstract

Instance-based learning methods such as the nearest neighbor classifier have proven to perform well in pattern classification in several fields. Despite their high classification accuracy, they suffer from a high storage requirement, computational cost, and sensitivity to noise. In this paper, we present a data reduction method for instance-based learning, based on entropy-based partitioning and representative instances. Experimental results show that the new algorithm achieves a high data reduction rate as well as classification accuracy. © Springer-Verlag Berlin Heidelberg 2006.