A Novel Condensing Tree Based Genetic Algorithm for Attribute Reduction

To efficiently reduce space complexity of a discernibility matrix, a compact structure, the so-called "condensing tree" (denoted by C-Tree for short ), was introduced, and two efficiently heuristic algorithms based on C-Tree for attribute reduction were presented, but the previously proposed algorithms only obtain one attribute subset. Therefore, in this paper, a novel condensing tree based genetic algorithm for attribute reduction is proposed. The new algorithm not only obtain multiple effective attribute sets, but also can sufficiently use the compactness of C-Tree, hence has high efficiency. Theoretical analysis and experimental results show that the algorithm of this paper has better or comparable performance on the six UCI benchmark datasets than that directly based on discernibility matrix.