Research On Dynamic Attribute Reduction Of Weakly Labeled Incomplete Rough Sets

Rough set theory analyzes and deals with uncertain,inconsistent and incomplete information without borrowing any prior information.It’s an effective mathematical tool.His main research interests are model extension and attribute reduction.Most of the existing attribute reduction algorithms are proposed in decision-making systems or information systems,and can only solve the reduction problem of labeled data or unlabeled data.And traditional attribute reduction algorithms consume a lot of time when dealing with dynamically changing datasets.Therefore,there are data with missing markers and missing attribute values in the dynamically changing dataset.In this paper,based on rough set theory,we will study the dynamic attribute reduction problem of numerical and symbolic weakly labeled incomplete data.Attribute reduction problem for numeric data with missing attribute values and tags.Extend the neighborhood relationship and equivalence relationship into neighborhood tolerance relationship and limit tolerance relationship.And the definition of conditional entropy of neighborhood tolerance is given.Use the marker missing rate as a weighting factor for the intersection of the decision attribute and the conditional attribute tolerance class.A form of weighted neighborhood tolerance conditional entropy is proposed to measure the uncertainty of numerical weakly labeled incomplete data.And constructed a heuristic attribute reduction algorithm.Further consider the dynamic variability of data.Based on the proofs of the incremental learning theorem of dynamic sample change and the incremental learning theorem of dynamic attribute change,an improved dynamic attribute reduction algorithm for numerical weakly labeled incomplete data is proposed.Attribute reduction problem for symbolic data with missing attribute values and tags.The concepts of consistent tolerance class and inconsistent tolerance class in the positive domain are introduced into the traditional discriminant matrix,and an improved definition of the discriminant relation is obtained.Using the discriminative relationship to construct the relative discriminative degree of attributes to describe the uncertainty of weakly labeled data,an attribute reduction algorithm based on relative discriminative degree is proposed.Further considering the dynamic variability of the data,the incremental update theorems for distinguishing the relationship between attributes and samples are given respectively.Based on this framework,all newly added sample pairs are used to establish an attribute reduction and increment mechanism for the hub,in which samples and attributes are added at the same time.Based on this incremental mechanism,a dynamic attribute reduction algorithm for symbolic weakly labeled incomplete decision-making systems is constructed.Six groups of data were selected from the UCI data set,and the incremental algorithm proposed in this paper was compared with the existing algorithms of the same type,and the effectiveness of the proposed incremental algorithm was proved.