The extended nearest neighbor classification
Zeng Yong, Bing Wang, Liang Zhao, Yang Yupu · 2008
The k-nearest neighbor classification rule (k-NNR) is among the most popular and successful pattern classification techniques. However, it usually suffers from the existing outliers, and in the small training samples situation, it performed poor. In this paper, a variant of the k-NNR, the extended nearest neighbor classification based on the local mean vector and the class mean vector has been proposed. The proposed classification method overcomes the influence of the existing outliers and performs obviously well than the traditional k-NNR in terms of the classification error rate on the unknown patterns.