Attribute reduction using distance-based fuzzy rough sets

Attribute reduction is one of the most important methods for feature selection in machine learning researches. In this work, a new fuzzy rough set model based on distance measures is proposed and the fuzzy dependency function is constructed. Then, the significance measure of a candidate attribute is defined, by which a greedy forward algorithm for attribute reduction is designed. The proposed algorithm is compared with several existing algorithms using UCI data sets. Experimental results show that the proposed reduction algorithm is feasible and effective.