Research Of Knowledge Reduction Algorithm Based On The Relativity Of Attributes In Information Systems

Attribute reduction is one of the most important work in Rough Set Theory.Attribute reduction is NP-Hard problem. It is inefficient. As the latest research resultof the Rough Set Theory, information entropy theory has been proved to be one kindof effective attribute reduction methods. The research of its expanded definition andalgorithms of Information Entropy has become a hot topic in the study of Rough Set.Taking complete information system and incomplete information system as studysubjects, a new method of knowledge reduction is proposed using the relativitybetween two attribute subsets. Information quantity is introduced to measureuncertainty of knowledge. This paper analyzes connection between informationquantity, conditional information quantity, joint information quantity and mutualinformation quantity. Correlation coefficient is introduced to describe the relativitybetween two attribute subsets. The completeness of information system can bedescribed by using the correlation coefficient. Attribute correlation is introduced todepict how important the attributes are. The attribute correlation is smaller, theattribute is more important. A new algorithm based on attribute correlation is given incomplete information systems. Finally, the experimental result shows that thisalgorithm is effective.Taking random information systems as study subjects, new methods ofknowledge reduction are proposed using the relativity between two subsets.Information entropy about logarithmic form and information entropy aboutcomplementary set are introduced to measure uncertainty of knowledge. For eachentropy, this paper analyzes relativity connection between information entropy,conditional information entropy, joint information entropy and mutual informationentropy. Correlation coefficients are introduced to describe the relativity between twoattribute subsets. Attribute correlations are introduced to depict how important theattributes are. Two new algorithms based on attribute correlation are given in randominformation systems. Finally, the experimental result shows that these algorithms iseffective.