Research On Attribute Reduction Method Based On Neighborhood Rough Set

Rough set theory is often used to mine data internal information and data extraction and attribute reduction is one of the key topics in rough set theory.To be more specific,attribution reduction is a process that exclude redundant attributes in data,compress data sample and improve classification performance.Moreover,the quality of the reduction results is mainly measured from the two perspectives: classification performance and time consumption.Therefore,a new algorithm based on majority voting is proposed to improve the classification accuracy.And a new acceleration method for three-way decisions is put forward to reduce the time consumption.Furthermore,a great number of researchers pay attention to local perspectives in order to improve classification performance of reduced results.Based on this,an integrated attribute reduction has been developed.It divides the universe into multiple decision systems and calculates these systems respectively,according to the decision category.However,the traditional reducts ignores the different decision categories of the universe,and makes it difficult to show good balance.(1)To solve this problem,a new algorithm based on majority voting is proposed.The algorithm is mainly composed of three stages: 1)select and combine samples of the same decision class into multiple new decision systems;2)compute the local significances in multiple decision systems,and find the attribute with the local maximum value of significance in each decision systems;3)apply the majority voting mechanism among the attributes obtained from the last step to select the suitable attribute.From the perspective of time,there is redundant calculation in the reduction resulting at lots of time consumption.To fill such a gap,three-way decisions is introduced into the process of attribute reduction.It reduces calculation and time.(2)An acceleration method for attribute reduction based on sequential three-way decisions are proposed in this study.The specific algorithm includes three steps: 1)calculate the importance of the attribute in the decision system;2)the attributes will be divided into three domains in terms of the significance degree of corresponding attribute.And the attribute with maximal significance degree will be classified into the positive domain.The attributes whose significance value equal zero will be classified into the negative domain and other attributes will be classified into the boundary domain;3)calculate the significance degrees of the attributes in the boundary domain,and constantly divide results into three parts.Add the attributes to the positive domain,until the condition is satisfied.In order to verify the effectiveness of the two methods mentioned above,8 UCI data sets are selected to conduct experiments in two sets of experiments.The experimental results show that the local voting methods can effectively improve the classification performance and the acceleration methods can further reduce time consumption for solving reductions.