Fuzzy-rough set based attribute reduction with a simple fuzzification method

The fuzzy-rough set based attribute reduction, which can get better reducts than the crisp rough set approach, has been paid more attention recently. Fuzzification is a step of data preprocess which was studied less in the application of fuzzy-rough set. In this paper, a simple fuzzification method deriving fuzzy discretization from K most important cuts in the application of feature selection is proposed. A comparative experiment between the proposed fuzzification method and a general fuzzy c-means based method is constructed on the UCI machine learning data repository. The experimental results show the obtained reducts using the proposed method can get higher classification accuracies and less number of selected attributes.