Authors: Changzhong Wang, Yuhua Qian, Weiping Ding, Xiaodong Feng

Abstract:

Classical fuzzy rough set often uses fuzzy  rough dependency as an evaluation function of feature  selection. However, this function only retains the maximum  member- ship degree of a sample to one decision class, it  can not describe the classification error. Therefore, in this  work, a novel criterion function for feature selection is  proposed to overcome this weakness. To characterize the  classification error rate, we first introduce a class of  irreflexive and symmetric fuzzy binary relations to redefine  the concepts of fuzzy rough approximations. Then, we  propose a novel concept of dependency: inner product  dependency to describe the classification error, and  construct a criterion function to evaluate the importance of  candidate features. The proposed criterion function not  only can maintain a maximum dependency function, but  also guarantees the minimum classification error. The  experimental analysis shows that the proposed criterion  function is effective for data sets with a large overlap  between different categories.

Keywords： Fuzzy rough set; Dependency function; Fuzzy inner product; Feature selection

Feature selection with fuzzy-rough minimum classification error criterion.pdf

查看详情

Thu Nov 25 18:50:00 CST 2021