摘要： 为解决1-v-r和1-v-1支持向量机中存在的拒识区域问题，提出一种加权k近邻法。该方法计算落入拒识区域中的样本，即拒识样本到所有训练样本的距离，选择最近的k个样本为拒识样本的类别投票，并根据距离大小进行加权，得票多的类即拒识样本的所属类。实验结果表明，加权k近邻法实现了零拒识，提高了传统多分类支持向量机的分类性能。

关键词: 加权k近邻法, 拒识区域, 多分类, 支持向量机

Abstract: To solve the unclassifiable region problem in 1-v-r and 1-v-1 Support Vector Machine(SVM), this paper proposes a weighted k-nearest neighbors method(Wk-NN). Wk-NN computes distances between the sample falling into the unclassifiable region called unclassifiable sample and each training sample. It selects k samples with least distances and votes for the classes by them. The votes are weighted according to the distances and the class with the largest votes is labelled for the unclassifiable sample. Experimental results show that Wk-NN eliminates the unclassifiable region in conventional multi-classification SVM and improves its classification performance.

Key words: weighted k-nearest neighbors method, unclassifiable region, multi-classification, Support Vector Machine(SVM)

中图分类号: 

• TP181