Adaptive Classification Boundary And Double Thresholds Supervised Neighborhood Rough Set And Its Attribute Reduction

Neighborhood rough set model is an extension of classical rough set model,which is suitable for mixed data sets including discrete and continuous data types.Its application fields include data mining,target detection,artificial intelligence and so on.It has broad application prospects and important research value.Attribute reduction is one of the important applications of neighborhood rough set,and it has the preprocessing function of information system.The basic idea is to delete redundant attributes and get a more concise description of data set without reducing the classification ability of given data set.It optimized the existing neighborhood rough set model and studied the attribute reduction based on the model.The specific work is as follows:(1)It analyzed the advantages and disadvantages of several existing neighborhood radius threshold value methods,and proposed a neighborhood radius thresholds value method based on classification boundary and double thresholds supervision.With this method,boundary function is introduced,and data objects are divided into two cases: the case of clear boundary and the case of indistinct boundary.In order to match the threshold values of different neighborhood radius adaptively,the granulation effect of the domain is optimized.(2)The adaptive classification boundary and double thresholds supervised neighborhood rough set model are constructed.Because the neighborhood particles under the thresholds value method of neighborhood radius supervised by classification boundary and double thresholds are not monotonic,the corresponding neighborhood upper and lower approximations are not monotonic.In this paper,we define the upper and lower approximations and three branches of the adaptive classification boundary and the double thresholds supervised neighborhood with monotonicity by iterative method.We propose the adaptive classification boundary and the double thresholds supervised neighborhood rough set model,and study the related properties of the model.(3)Based on adaptive classification boundary and double thresholds supervised neighborhood rough set model,attribute reduction is studied.Based on the model,the classification attribute reduction and the specific classification attribute reduction are constructed,and the heuristic algorithm of classification attribute reduction is designed.The algorithm is verified by using UCI dataset to delete redundant attributes and maintain high classification accuracy.The transformation relationship,derivation relationship and class cohesion relationship between classification attribute reduction and specific class attribute reduction under this model are discussed.It provides a new perspective for the study of attribute reduction.