Research On Continuous Attribute Attribute Reduction Based On Fuzzy Clustering And Rough Set

Attribute reduction is an important application study of rough set, which is a process to remove the useless attribute with classification capability unchanged. It will reduce the heavy burdens caused by large-scale data and countless attributes during data mining. However, the attribute reduction algorithms based on rough set are only appropriate for discrete data, which is one kind of data type existing in real data sets. Therefore, a discretization process is necessary for continuous data mining. While this process lose information to some extent, as it’s unable to reserve the value differences within objects. Hence, in this paper an algorithm based on fuzzy clustering and rough set to handle continuous data is raised.The algorithm based on fuzzy clustering and rough set can be divided into two parts. In the first part, fuzzy clustering is applied to transform the fuzzy quality existing in attribute value into the object relation. Based on the fuzzy relation within objects, the object partition will be done. This step is equal to partition operation of equivalent classes. In the second part, fuzzy clustering is used to make the closer attributes gather into one kind. From each outcome category, a representative attribute will be chosen as member of target attribute subset. Further, the concept of dependency degree in rough set is used to estimate the target attribute subset. The algorithm proposed in this paper takes both diversity and relativity within attributes into account, in order to insure the coverage width of information and the low damage during reducing attributes meanwhile retain the classification capability.Discriminative from the attribute reduction algorithm based on rough set, computation of core attributes and importance degree for single attribute in different level is unnecessary in algorithm proposed in this paper, which is a promotion. Finally, seven different kinds of dataset in UCI and three groups of dataset in economic area are used to test effectiveness of the algorithm. The algorithm in this paper turns out effective and powerful in decision-tree classification as well.