K-Nearest Neighbor Clustering Algorithm Based on Kernel Methods
Sheng Sun, Yuanzhen Wang · 2010
KNN algorithm is the most usable classification algorithm, it is simple, straight and effective. But KNN can not identify the effect of attributes in dataset. For non-Gaussian distribution or non-Elliptic distribution, KNN can not solve these two kinds of problem effectively. A major approach to tackle this problem is to give each of the rest of attributes a weight value according to the relationship between these attributes. The bigger the attribute weight is, it has more importance extent in figuring out the distance of samples in kernel space. In this paper, we proposed a kernel-based KNN clustering algorithm which improved accuracy of KNN clustering algorithm. We tested the accuracy rate of the suggested algorithm KKNNC using the six UCI data sets, and compared it with KNNC algorithm in the experiments. The experimental results show that KKNNC algorithm outperform KNNC algorithm in accuracy significantly.