基于特征集中度和差别对象对集的特征选择方法

本文首先简单分析了几种经典的特征选择方法，总结了它们的不足，然后提出了特征集中度的概念，紧接着把差别对象对集引入粗糙集并提出了一个基于差别对象对集的属性约简算法，最后把该属性约简算法同特征集中度结合起来，提出了一个综合性特征选择方法．该综合性方法首先利用特征集中度进行特征初选以过滤掉一些词条来降低特征空间的稀疏性，然后再使用所提属性约简算法消除冗余，从而获得较具代表性的特征子集．实验结果表明该综合性方法效果良好．

Feature Selection Based on Feature Concentration and Discernibility Object Pair Set

Firstly, several classic feature selection methods are simply analyzed and their deficiencies are summarized. And then, the concept of feature concentration is presented. Next, discernibility object pair set is introduced into rough sets and a new attributes reduction algorithm based on discernibility object pair set is proposed. Finally, combining the proposed attribute reduction algorithm with the presented feature concentration, a syntaxic feature selection method is provided. The syntaxic method firstly uses the presented feature concentration to select feature and filter out some terms to reduce the sparsity of feature spaces, and then employs the proposed attribute reduction algorithm to eliminate redundancy, so that the feature subset which is more representative is obtained. The experimental results show that the syntaxic method is promising.