Research On Knowledge Discovery And Reasoning Methods Based On Rough Set And Database Technology

Knowledge discovery in database—KDD (also named data mining DM)is a new kind of intelligent information processing technology to the largequantity data in databases, and a new kind of basic component ofbuilding new intelligent information systems such as Intelligent BusinessSystem, new Decision Support System, etc. And it is presently a researchhotspot on intelligent information technology.Rough set is a mathematical tool for research on imprecision and vagueproblems, and is well admitted as a theoretical foundation of researchingon DM, granular computing and so on. Database technology is the mosteffective and advanced technology for managing and manipulating thelarge quantity data in the databases. It is a new effective way that takingthe advantage of database technology to improve or design new efficientalgorithms suitable for large data sets, which is being explored by manyscholars. This paper take Rough set theory as a main theoretical basis andmake full use of the advantage of database technology to research someproblems in data mining and propose some effective and feasiblesolutions. The main research contents are : 1. Based on the analysis of the time cost and space cost as well asthe necessity of using discernibility matrix to obtain core attributes inattribute reduction algorithm, propose an improved attribute reductionalgorithm based on Rough set and database technology. Experimentsshow that its efficiency on large data sets is much faster than some otherattribute reduction algorithms based on memory and is easy to realize anduse. 2. For the problem that the main mechanisms for measuring theclassification ability of an attribute in Rough set, such as the positive area,can not reflect the synthetic contribution ability of the attribute forclassification, proposes a synthetic measure —attribute classificationrough degree (ACRD) for measuring attribute contribution toclassification in rough set. Theoretical analysis and experiments showthat as a measure of choosing attribute in classification algorithm, ACRDis better than information gain and at the same extent with informationgain ratio, and is simpler in calculating. Based on this measure and otherrelative research results worked out in this paper, a new algorithm ofclassification is proposed which has good scalability and adaptability, andcan generates decision tree or rule sets directly.3. Based on the analysis of the problems on processing noise data andinconsistent data in data mining algorithms, propose a new method ofprocessing noise data—predictive pruning method in decision treealgorithm based on variable precision Rough set, and propose a simpleand effective method of detecting and processing inconsistent data basedon Rough set theory, which is highly syncretic with the decision treealgorithms.4. Propose a way to organize a knowledge base which has treestructure in logic in relation database, and propose a reasoning methodusing database query. on this kind of knowledge base.