Unsupervised attribute reduction: improving effectiveness and efficiency.

Attribute reduction has shown its effectiveness in improving the performance of classifiers. Different from widely studied supervised attribute reduction, unsupervised attribute reduction faces great challenges from two main aspects: performance requirement and computationally demanding. Therefore, both effectiveness of selected attributes and efficiency of searching qualified reduct are addressed in the problem solving of unsupervised attribute reduction. Firstly, an ensemble selector is introduced into forward greedy searching. The objective is to identify more suitable attribute for each iteration in the process of searching. Secondly, both sample and attribute based acceleration mechanisms are introduced into our ensemble selector. The first stage is used to derive reduct with better performance, and the second stage is used to speed up the procedure of searching. Finally, our approach is compared with several well-established attribute reductions over 16 UCI datasets. The comprehensive experiments clearly validate the superiorities of our study from the perspectives of both effectiveness and efficiency. [ABSTRACT FROM AUTHOR]