Study On Feature Selection From The Perspective Of Fusion Granulation Mechanism

Human beings are gradually entering an intelligent society driven by big data.How to efficiently and quickly dig out valuable information from high-dimensional data has become a top priority;among the many feasible methods and theories for processing high-dimensional data,rough set theory Its mathematical foundation is mature,and it does not require any prior knowledge and is highly valued.In rough set theory,attribute reduction is the most common feature selection technique,and its core essence is to reduce the dimensionality of data by deleting redundant and irrelevant attributes.The neighborhood rough set model is an important model to solve the problem of feature selection,but because the model adopts the "true or false" binary logic when constructing the neighborhood,it has performance when describing sample data under fuzzy background Fuzzy Rough Set Model.In view of this,this paper constructs a new rough set model by fusing the advantages of neighborhood and fuzzy rough set,and compares the performance difference between this model and other traditional rough set models.Specifically,the research content and innovative results of this article are mainly:(1)Introducing Gaussian kernel function to construct neighborhood fuzzy rough set model.In the process of solving the traditional neighborhood relationship,the relationship between the distance between the samples and the given radius is used to determine whether the samples are similar.Obviously,this method has problems that it is difficult to quickly give a suitable radius and cannot reflect the degree of sample similarity.In order to solve this problem,this paper combines the advantages of neighborhood rough set and fuzzy rough set,introduces Gaussian kernel function into the process of constructing neighborhood to calculate fuzzy similarity,and establishes neighborhood fuzzy according to the relationship between fuzzy similarity and similarity threshold.Rough set model,and defines basic concepts such as fuzzy decision,fuzzy upper and lower approximation,fuzzy neighborhood conditional entropy and fuzzy neighborhood conditional discrimination index.(2)Research on Feature Selection Based on Neighborhood Fuzzy Rough Set.In this chapter,in order to verify the advantages of the neighborhood fuzzy rough set model in solving feature subsets,two metrics based on conditional entropy and conditional discrimination index are designed for neighborhood rough set,k-nearest neighborhood rough set,and neighborhood fuzzy rough set.The experiment selects 6 sets of data in the UCI data set and uses the forward greedy heuristic algorithm to solve the feature subset.The experimental results show that the neighborhood fuzzy rough set model can obtain higher classification accuracy on the two measurement criteria and the number of features at the same time Also significantly reduced.(3)Research on Feature Selection Based on Feature Relationship and Fuzzy Decision.First,a single-label feature selection neighborhood fuzzy rough set model is established.Two different membership calculation methods are used to obtain the fuzzy membership of the equivalence class of the sample,and the minimum membership value in each equivalence class is regarded as the membership Then use the relationship between the membership degree of the neighborhood sample and the threshold to redefine the fuzzy rough upper and lower approximation,and then use the decision attribute to measure the dependence of the feature subset for feature selection;perform feature selection on 7 public UCI data sets The experiment results show that,compared with several existing feature selection methods,the classification accuracy is improved while the number of features is significantly reduced.