Studies Of Pseudo-label Attribute Reduction Under Local Perspective

Since the beginning of the new century,the Internet of everything has become a major feature of the new era,which has changed the way of obtaining information in the past and generated a huge amount of data.Therefore,many scholars have put forward many theoretical models for processing massive data.Among these theoretical models,rough set theory has been widely applied and many extended models based on rough sets have been born.As an extended model of classical rough set,neighborhood rough set can well fill the gap that classical rough set can’t process continuous data.In spite of this,the traditional neighborhood rough set does not consider the influence of radius variation on neighborhood relationships,so different label samples may fall into the same neighborhood.In view of this,the pseudo-tag policy is introduced into the neighborhood rough set and the pseudo-tag neighborhood rough set model is constructed,which solves the problem well.As an important research content of neighborhood rough set,attribute reduction is to remove unnecessary data to improve classification performance.However,the traditional local attribute reduction based on the neighborhood rough set ignores the difficulties faced by the neighborhood rough set mentioned above.Therefore,this paper constructs an attribute reduction method from the local perspective and combines with the pseudo tag neighborhood rough set model.On this basis,combined with the idea of sample selection and three decision acceleration,and on the basis of ensuring the classification performance,a method of acceleration is proposed.(1)Pseudo-label voting reduction algorithm in local view.The specific algorithm includes the following steps: 1)Using K-means clustering to calculate the pseudo-label and add it to the decision system;2)Use heuristic algorithm to calculate the attribute importance degree of each decision class in the pseudo-label decision system,and select the attribute with the highest attribute importance degree;3)According to the obtained attributes,use the method of integrated voting selection to select the attributes that meet the constraints and put them into a reduction set.And so on until the constraints of the algorithm are met.From the point of view of time consumption of the algorithm.On the one hand,pseudo-label voting reduction algorithm increases the time consumption when calculating pseudo-label from local perspective;On the other hand,from the point of view of the process of attribute reduction,both local and global have redundant reduction,thus increasing the reduction time.In view of this,this paper will introduce the idea of this choice and three decisions into the pseudo-label voting reduction algorithm from local perspective,and improve the reduction speed by reducing the reduction data and reducing the redundant attributes.(2)Acceleration algorithm of pseudo-label voting attribute reduction based on sample selection and three-point decision from local perspective.Specific algorithm includes the following steps: 1)given a decision system,a neighborhood,calculated using the Euclidean distance and select the smallest distance from his own neighbor and concludes that the neighbor’s decision,if the neighbor’s decision classes with own decision the same class,select the sample,and then all of the sample to form a decision-making system;2)Calculate the attribute importance degree,and then divide the calculated attribute importance degree into three parts by using the three decision ideas.Those with a high attribute importance degree are divided into positive domain,those with a zero attribute importance degree are divided into negative domain,and the remaining attributes are divided into boundary domain.3)Repeat the second step until the constraint conditions are met.In order to verify the effectiveness of the pseudo-label voting reduction algorithm from local perspective and the pseudo-label voting attribute reduction acceleration algorithm based on sample selection and three-branch decision from local perspective,several sets of data sets were selected from the UCI data set and compared with the classification ability and time consumption of the traditional algorithm.Experiments show that the pseudo-label voting reduction algorithm can effectively improve the classification performance of the reduction in local view,and the pseudo-label voting attribute reduction acceleration algorithm based on sample selection and three-decision in local view can effectively reduce the reduction time consumption.