Local weight-based k-nearest centroid neighbor algorithm

The k-nearest centroid neighbor rule( KNCN),as an effective extension of the k-Nearest Neighbor rule( KNN),is one of the effective algorithms in pattern classification. The KNCN is prone to be seriously influenced by the existing outliers. At the same time,all the k-nearest centroid neighbor samples have the same weight and the same contribution to classification results,which is unreasonable. To solve this problem,this paper proposes a nearest centroid neighbor algorithm based on the local weight,taking account of the proximity and spatial distribution characteristics of the neighbors for a query pattern. The experimental results show that the classification accuracy of LWKNCN is better than that of the traditional KNN algorithm and KNCN algorithm.