Abstract: Neighborhood rough set theory is mainly used for knowledge discovery,attribute selection,decision analysis and data mining,and other fields.It can choose appropriate discretization strategy based on the characteristics of data and perform well in dealing with fuzzy and uncertain knowledge,but the traditional rough set attribute reduction algorithm is difficult to obtain reduction,and the attribute recognition accuracy of reduced rough set is low.Therefore,this paper put forward a kind of attribute reduction algorithm based on attribute importance.Considering the shortcomings of conditional information entropy in many aspects,the threshold parameters are re-selected by using the theory of varia-ble precision neighborhood rough set.Based on the new conditional information entropy as the measurement benchmark,the preference decision rule set is deduced according to the preference attributes in the decision information system.This paper extracted rough rules from preference decision rule set and established a variable precision neighborhood rough set model by using neighborhood granulation method.When dealing with large-scale rough set attribute data,this model takes a long time to calculate and has too many redundant attributes.Aiming at this problem,an evaluation strategy of attribute importance was given.Based on this,a variable precision neighborhood rough set attribute reduction algorithm was theoretically designed by fusing multi-tree.The experimental results show that compared with the traditional method,the accuracy of attribute recognition of the proposed method is 92%,which is improved by 10%.This fully verifies that the proposed attribute reduction algorithm has strong effectiveness and higher application value.

Key words: Attribute importance, Attribute reduction, Multi-way tree, Rough set, Variable-precision neighborhood