非一致性引导的无监督特征选择

由于无监督环境下特征选择缺少类别信息的依赖,所以利用模糊粗糙集理论提出一种非一致性度量方法DAM(disagreement measure),用于度量任意两个特征集合或特征间引起的模糊等价类含义的差异程度.在此基础上实现DAMUFS无监督特征选择算法,其在无监督条件下可以选择出包含更多信息量的特征子集,同时还保证特征子集中属性冗余度尽可能小.实验将DAMUFS算法与一些无监督以及有监督特征选择算法在多个数据集上进行分类性能比较,结果证明了DAMUFS的有效性.

Unsupervised feature selection guided by disagreement

Because feature selection in an unsupervised environment lacks dependence on category information, this paper proposed a disagreement measure(DAM) with the use of the fuzzy rough set theory. DAM measured the degree of difference in meaning of fuzzy equivalence of any two feature sets or features. On this basis, this paper proposed the DAMUFS unsupervised feature selection algorithm. The DAMUFS algorithm could select feature subsets that contained more information under unsupervised conditions, while also ensuring that the attribute redundancy in the feature subset was as small as possible. The experiment compared the classification performance of DAMUFS algorithm with some unsupervised and supervised feature selection algorithms on multiple data sets. And the results prove the effectiveness of DAMUFS algorithm.