Attribute Reduction Algorithm Based On Neighborhood Rough Sets

In recent years,the large number of data explosive growth,with the rapid development of information technology,which contains a large number of redundant data.And these data will have an impact on us to make the right decision,so it will be more urgent to preprocess the data.Attribute reduction as the core of rough sets theory has achieved great development:the main work of it is under the condition of the ability for classification system unchanged,deletes the redundant features or not important,so as to reduce the search space and improve the reduction efficiency.It is widely used in data mining,pattern recognition and so on.In the process of data analysis,the classical neighborhood rough set theory only uses the positive region as the effective classification of the sample,in this case,the separability of the boundary samples would be ignored,so positive region can not truly reflect the real situation of the classification.For this reason,in this paper,we reconstruct a rough set attribute reduction model based on the maximum decision function,and discuss its related properties.The main work of this paper is as follows:1.In the neighborhood rough set,the positive region is usually used to reflect the classification ability of the feature subset.However,the positive region is not an effective estimation of the classification accuracy,because it only considers the sample of consistent decision,which ignores the boundary sample.According to Bias classification rules,boundary samples also contain certain classification information.Based on this,in this chapter,by introducing the concept of maximum decision function,this paper proposes a new rough set model which not only considers the samples in the positive region,but also can distinguish the information of the boundary samples.The model guarantees the minimum classification error of samples.Under the model,we construct a feature evaluation function reflected the ability of classification--the dependency function,and design a attribute reduction algorithm.Through numerical experiments,we compared the proposed algorithm with some existing algorithms,and the experimental results show that the method is effective.2.In the rough set model of maximum decision function,feature evaluation is carried out by dependency function.The monotonicity of the dependency function is very important for the design of attribute reduction algorithms,that is why the monotonicity of the dependency function can achieve a better algorithm to stop the search steps.However,the dependency function in rough set model based on maximum decision function is not monotonic.To solvethis problem,based on the maximum decision function,we reconstruct the upper and lower approximation and boundary,then the dependence function based on the maximum decision function is defined,which can solve the problem of the monotonicity of the dependency function under the condition of the same classification accuracy.Finally,we analyze by using the UCI data set.The numerical experiments show that the algorithm can find smaller and more accurate feature subsets than the classical reduction algorithm,so the algorithm is feasible.