关键词: K最近邻居,原型选择,自然邻居,最小生成树,分类

Abstract: K-nearest neighbor (KNN) is one of the most popular algorithms for supervised classification.However,the traditional KNN classification has two limitations that the option of parameter K and prohibitive computational and storage demand for large datasets.To overcome these limitations,a new prototype selection algorithm was proposed,which retains some key prototypes that make a large contribution to classification and removes the most of other prototypes with little contribution for classification.Differing from other prototype selection algorithms,the proposal uses a novel neighbor concept natural neighbor to preprocess the dataset and builds minimum spanning tree based on the speci-fic terminal conditions.According to the MST,the prototypes close to boundaries and some internal prototypes are preserved.Experimental results show that the proposed algorithm effectively reduces the number of prototypes while maintaining the same level of classification accuracy as the traditional KNN classification algorithm.Moreover,it is a little bit superior to other prototype selection algorithms in classification accuracy and retention ratio.

Key words: K-nearest neighbor,Prototype selection,Natural neighbor,Minimum spanning tree,Classification