面向分类特征的无监督特征选择方法研究  

Unsupervised Feature Selection Method for Categorical Features

作　　者：王连喜[1] 蒋盛益[2] 

机构地区：[1]广东科贸职业学院,广东广州510430 [2]广东外语外贸大学信息学院,广东广州510006

出　　处：《小型微型计算机系统》2011年第1期47-50,共4页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(60673191)资助;广东省高等学校自然科学研究重点项目(06Z012)资助;广东省自然科学基金项目(9151026005000002)资助

摘　　要：针对分类特征数据给出一种新的特征重要性程度度量方法.以一趟聚类算法为基础,提出一种无监督特征选择方法.理论分析表明该方法时间复杂度与数据集的大小和特征个数成近似线性关系,适合于大规模数据集中的特征选择.在UC I数据集上的实验结果表明,与文献中的经典方法相比,本文方法具有较好的性能,说明提出的特征选择方法是有效可行的.In this paper,a new definition of measuring the importance of features is proposed for categorical data.Furthermore an unsupervised feature selection method based on one-pass clustering algorithm is presented.Theory analysis indicates that the time complexity of the feature selection method is nearly linear with the size and the number of features of dataset.It can be applied in feature selection for high dimensional data.Experimental results on UCI datasets show that the performance obtained by the proposed method is effective and practicable in features selection through comparing with other traditional feature selection approaches.

关 键 词：特征选择 分类特征 一趟聚类 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]