An improved kNN algorithm based on essential vector

Abstract

There are some limitations in traditional k-nearest neighbor (kNN) algorithm, one of which is the low efficiency in classification applications with high dimension and large training data. In this paper, an improved kNN algorithm EV-kNN is proposed to reduce the computation complexity by cutting off the number of training samples. It firstly gets k classes by the kNN calculation with the essential vector, then assigns corresponding category using the kNN again. Experimental results show that the improved algorithm can perform better than several other improved algorithms.