一种基于成对约束的特征选择改进算法  

An Improved Algorithm for Feature Selection Based on Pairwise Constraint

作　　者：杨杨[1] 刘会东[2] 

机构地区：[1]南京师范大学强化培养学院,江苏南京210046 [2]南京师范大学计算机科学与技术学院,江苏南京210046

出　　处：《南京师范大学学报（工程技术版）》2011年第1期56-61,共6页Journal of Nanjing Normal University(Engineering and Technology Edition)

基　　金：南京师范大学2010年学生科学基金

摘　　要：基于成对约束的特征选择算法通过度量单个特征的重要性得到一个特征序列,但由单个重要特征构成的特征子集未必是最有效的.为此,提出了一种基于成对约束的特征选择改进算法,该算法采用对特征子集进行度量的策略,逐步选择使新的特征子集最有效的特征,从而得到一个有效的特征序列.实验表明新提出的算法是有效可行的.Feature selection is key issue in machine learning field. As compared with unsupervised feature selection methods, supervised feature selection approaches have more better performances. However, most of the existing supervised feature selection algorithms mainly aim at the cases using the labels as supervised information, here these methods are not applied to the cases with pairwise constraints. In the real application, it is more easier to get the pairwise con- straints as comparing with getting labels. So the researchers proposed a feature selection based on pairwise constraint, the ＇algorithm obtains a feature sequence by measuring the significance of each single feature, but in fact the feature sub- set combining by those more important features may be not an effective feature subset. Therefore, in this paper, we introduce an improved feature selection algorithm based on pairwise constraint, the newly developed algorithm focuses on evaluating the importance of a feature subset hut not a single feature, that is, it uses the empty feature subset as starting point, and then gradually extends this feature subset by adding a most effective feature in every round, in this way an effective ranking feature list is obtained. Experimental results show that the newly proposed algorithm is flexible.

关 键 词：机器学习 特征选择 成对约束 分类 

分 类 号：TP311[自动化与计算机技术—计算机软件与理论]