基于属性加权的主成分分析算法  

An attribute-weighted principal-component analysis algorithm

作　　者：王永欣[1] 张化祥[1] 王爽[1] 

机构地区：[1]山东师范大学信息科学与工程学院,山东省分布式计算机软件新技术重点实验室,山东济南250014

出　　处：《济南大学学报（自然科学版）》2015年第6期438-443,共6页Journal of University of Jinan(Science and Technology)

基　　金：国家自然科学基金(61170145)

摘　　要：基于主成分分析提出一种通过属性加权的方式对特征进行预处理的改进算法,实现特征选择与特征提取的结合,从而降低计算复杂度并提高分类准确度。属性加权是通过量化样例与分类标记之间的相互依赖关系,即结合了线性判别分析的映射思想,线性拟合样例与标记,得到一组反映各属性对分类贡献大小的权值w。通过将改进算法与主成分分析法、线性判别算法、局部保持投影算法做了分类准确度、计算时间的综合实验比较,证明了改进算法的有效性。In view of PCA,we propose an improved algorithm for preprocessing the feature data by attribute weighting. This method combines feature extraction with feature selection so that it reduces computational complexity and increases accuracy of classification. In combination with the concept of mapping of linear discriminant analysis（ LDA）,we quantify the connection of sample data and labels to get a set of weighted values w which are computed by linear fit. These values represent the extent of each attribute contribute to classification. And finally we take an experiment with PCA,LDA,and locality preserving projection on classification rates and running time.The results show that the improved algorithm is superior to the PCA algorithm on classification.

关 键 词：数据降维 属性加权 主成分分析 特征提取 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]