Applications of Approximate Reducts to the Feature Selection Problem

In this paper we overview two feature rankings methods that utilize basic notions from the rough set theory, such as the idea of the decision reducts. We also propose a new algorithm, called Rough Attribute Ranker. In our approach, the usefulness of features is measured by their impact on quality of the reducts that contain them. We experimentally compare the reduct-based methods with several classic attribute rankers using synthetic, as well as real-life high dimensional datasets.