Research Of Knowledge Discovery Method Based On Rough Set Theory

With the rapid development of information technology, database storage is expanding larger and larger, thus developing the phenomenon of"data explosion but lack of knowledge". Therefore research on the technology of knowledge discovery in database is very important. Now knowledge discovery in database faces dealing with incomplete and uncertain data unefficiently and the poor explanation of knowledge. Rough set theory is a mathematical tool which deals with vague and uncertain knowledge. It does not deed prior knowledge and external parameters. So applying rough set theory to knowledge discovery in database has very important significance.Attribute reduction and rule extraction are two important stages in rough set theory. This paper based on the thought of Hu Keyun's attribute frequency, proposes attribute weight frequency function (AWFF) as heuristic information; against the problem of the low efficiency of attribute reduction, this paper proposes attribute weight frequency function_attribute reduction algorithm (AWFF_AR) which applies AWFF to attribute reduction with approximate accuracy and strong equivalent set. The algorithm can get an attribute reduction set, improve the algorithm's efficiency and solves the problem that can not distinguish which attribute is more important when the values of AWFF are equal; against the problem that the efficiency of the rule extraction is not high and the generated rules are redundant, the paper proposes attribute weight frequency function_value reduction algorithm (AWFF_VR). The algorithm starts at value core, makes AWFF as heuristic information and uses support and confidence to remove the redundant rules. The algorithm not only improves the efficiency but gets the rules which have the short length and the high coverage.Theoretical analysis and experimental results on UCI data set prove that AWFF_AR and AWFF_VR algorithms can achieve the expected results, reduce the algorithm's operating time, and get effective knowledge rules. The algorithms have important significance for future studies and practical applications.