Research And Application Of Fast Attribute Reduction Algorithm For Neighborhood Rough Set

Rough Set Theory which is an excellent data analysis tool to handle imprecise, inconsistent and incomplete data, has been widely applied to many areas successfully such as Data Mining. But the classical Rough Set Theory is just applicable to the discrete feature data. In the space of continuous data which is widely used, the Rough Set Theory can’t be used directly. Neighborhood Rough Set model which is an extension of Rough Set Theory has increasingly attracted researchers’ attention because it has the ability of handling continuous data.As one of the most important problems of the rough set theory, the attribute reduction is used to delete the redundant or unrelated attributes based on the principle of keeping on the invariable classifying ability. Recently, the study of the attribute reduction based on the neighborhood rough set has been the new focus.This thesis focused on the neighborhood rough set theory and the works I did in this thesis are as follows:(1) A particular analysis about the complex operation of calculating the neighborhood elements of each record in neighborhood rough set model is given. Then according to the distribution of sample records in the space, the concept of Block Sets is proposed, which proves the Block-Set-neighborhood theory. The Block-Set-neighborhood theory shows that neighborhood elements of a record are contained only by its own block set and the adjacent block sets. Based on the concepts, a quick attribute reduction algorithm is proposed. Then the algorithm is proved to be an efficient way to reduce the time complexity to calculate the neighborhood elements of each record by the experiment.(2) After an analysis of setting the size of neighborhood in the process of calculating record’s neighborhood elements, the disadvantage of setting a single threshold is proposed. Then the concept of multiple thresholds to displace the traditional uniform threshold approach is proposed. Based on the concept, the block set is improved to the multi-threshold block set. Then the attribute reduction algorithm based on the multi-threshold block set is proposed. In the same way, this algorithm is proved to be an efficient way after the experiment.(3) The attribute reduction algorithm based on the neighborhood rough set theory is applied to feature subset selection in mechanism fault diagnosis systems. By one fault diagnosis example, the feasibility and the effectiveness of the algorithm are analyzed and verified. And the results show that this method can efficiently extract the main fault features.