Feature selection using self-information and entropy-based uncertainty measure for fuzzy neighborhood rough set

Abstract Feature selection based on the fuzzy neighborhood rough set model (FNRS) is highly popular in data mining. However, dependent function of FNRS only considers information present lower approximation decision while ignoring upper decision. This construction method may lead to loss some information. To solve this problem, paper proposes a joint entropy self-information measure (FNSIJE) and applies it feature selection. First, construct four uncertain measures variables, concept introduced into approximations from algebra view. The relationships between these their properties are discussed detail. It found that fourth measure, named tolerance self-information, has better classification performance. Second, an uncertainty been proposed Inspired by both views, FNSIJE proposed. Third, K–S test used delete features with weak distinguishing performance, which reduces dimensionality high-dimensional gene datasets, thereby reducing complexity then, forward algorithm provided. Experimental results show compared related methods, presented can select less important have higher accuracy.