Abstract:
Attribute reduction is a critical issue to find a minimal subset of features from the initial dataset by eliminating redundant and unnecessary features. The Rough set has a powerful technique for identifying the superfluous features that can be removed without losing any valuable information. However, it can not find minimal reduct sets in an available time when the dataset has many attributes. Therefore, to overcome this difficulty, some natural inspired meta-heuristics algorithms combined with Rough set have been developed. This paper develops a novel attribute reduction strategy based on Rough Set (RS) and Tug of War Optimization (TWO) algorithm. The original TWO is appropriate for a problem with a continuous search space. However, attribute reduction is a binary problem. Therefore, we have proposed a binary version of TWO combined with RS theory called BTWORSR to find the best attribute reduct sets. For performance evaluation of the proposed binary-coded TWO, seven standard benchmark datasets from UCI are selected and employed. The experimental results show that the developed binary of the TWO significantly gave better results in terms of classification accuracy rate compared to other Rough Set based algorithms. Besides, it also yielded the most informative attributes for classification tasks.
