Research On Acceleration Methods Based On Attribute Reduction Under Supervised Neighborhood Rough Set

Rough set theory as a tool for data processing and analysis,can extract useful information from countless data and is widely used in many fields.The neighborhood rough set model as an extension of rough set theory that can handle continuous and mixed-type data,uses a single radius to granulate samples in the neighborhood,without considering the label information of samples.Therefore,it cannot effectively reduce the inconsistency and uncertainty of information.To address this issue,the supervised neighborhood rough set model introduces intra-class and inter-class radii in the process of neighborhood granulation,so that the number of samples with the same label in the neighborhood particles remains unchanged while the number of samples with different labels is reduced.This supervised granulation method can filter out some inconsistent or uncertain information in the process of neighborhood granulation,greatly improving the discriminability of neighborhood information particles.Attribute reduction as an effective data dimensionality method,can eliminate redundant attributes in conditional attributes,thus obtaining the minimum attribute subset that satisfies the constraint conditions.However,the forward greedy search method in solving the reduction problem needs to repeatedly calculate the significance of all candidate attributes,which obviously results in significant time consumption.In view of this,this paper considers the selection of conditional attributes and systematically studies how to accelerate attribute reduction under supervised neighborhood rough set model,and constructs a corresponding algorithm framework based on this.Specifically,the main research contents of this paper include the following points.1.An attribute reduction method based on sample standard deviation attribute significance is proposed.In analyzing existing attribute reduction methods,it was found that using an importance function to solve the reduction problem often requires repeated calculation of the significance of all candidate attributes,resulting in high computational complexity.To address this issue,the method first calculates the dispersion degree of each sample under all conditional attributes and uses this to determine the significance of each attribute,thereby reducing computational complexity.Secondly,a clustering algorithm is used to group similar conditional attributes into the same group,which compresses the search space for attributes.Finally,an attribute reduction method based on sample standard deviation attribute significance is proposed.Comparative experiments show that the proposed method not only reduces computational time but also improves the classification performance of the resulting attribute reduction.2.Proposed a weighted attribute based attribute reduction method.In the attribute reduction method based on sample standard deviation attribute significance,the significance of all candidate attributes is calculated only considering the degree of aggregation among samples with the same label,while ignoring the degree of aggregation among samples with different labels.To address this issue,we consider both aspects and calculate the weight values between samples with different and same labels for all candidate attributes,and use these weight values to determine the significance of all candidate attributes.Finally,a weighted attribute based attribute reduction method is proposed.Comparative experiments have demonstrated the superior time efficiency and classification performance of our proposed method for attribute reduction.