基于Relief的特征加权壳近邻分类算法  

Shell nearest neighbor classification of feature weighted based on Relief

作　　者：郑洁[1] 秦永彬[1] 许道云[1] 

机构地区：[1]贵州大学计算机科学与信息学院,贵州贵阳550025

出　　处：《计算机工程与设计》2013年第3期951-954,1017,共5页Computer Engineering and Design

基　　金：国家自然科学基金项目(60863005;61262006);贵州省科学技术基金项目(黔科合J字[2012]2125号);贵州大学引进人才科研基金项目(贵大人基合字(2011)14号)

摘　　要：壳近邻分类算法克服了k近邻分类在近邻选择上可能存在偏好的问题,使得在大数据集上的分类效果优于k近邻分类,为了进一步提高壳近邻算法的分类性能,提出了基于Relief特征加权的壳近邻分类算法。该算法在Relief算法的基础上求解训练集的特征权值,并利用特征权值来改进算法的距离度量方法和投票机制。实验结果表明,该算法在小数据和大数据上的分类性能都优于k近邻和壳近邻分类算法。Relative to k nearest neighbor classification, shell nearest neighbor classification is designed which is without bias at selecting nearest neighbors, and it has the better accuracy on lager data sets. In order to improve performance of shell nearest neighbor classification ulteriorly, the shell nearest neighbor classification of feature weighted based on relief is proposed, which solve the feature weights of training set based on relief algorithm. The feature weights are used to modify method of distance metric and voting mechanism. The experimental results show that the shell nearest neighbor classification of feature weighted based on relief has the better performance on classification than k nearest neighbor and shell nearest neighbor classification both on small and lager data sets.

关 键 词：近邻分类 RELIEF算法 特征权值 距离度量 投票机制 

分 类 号：TP301.6[自动化与计算机技术—计算机系统结构]