Attribute Reduction And Rule Acquisition Research In Complex Order Information System

In practical problems, due to the error of the decision makers, information missing and other reasons, we face the order information system with its data being inconsistent or incomplete. How to make use of rough set theory based on dominance relation to mine knowledge which is hidden in a complex order information system is of practical significance. This paper mainly researches the attribute reduction and rule acquisition problems in complex order information systems. And we primarily do the following several aspects.(1) The lower approximation reduction of an inconsistent order decision information system is researched, a new indicator of attribute importance was defined, and a new lower approximation reduction algorithm was proposed with the previous attribute importance as heuristic information. At the end we use some UCI datasets to do experiments with the reduction algorithm and the experimental results show that the proposed algorithm is effective.(2) In order to guarantee the inconsistent order decision information system in harmony, by combining with the domination principle, a generalized assignment function is defined which is easier and can reflect the consistent of system better. In Addition, a concept of generalized assignment reduction is defined, and a general distribution reduction method of discernibility matrix is proposed.(3) Through analyzing the existing limitations of all kinds of dominance relations in the incomplete order information system and combining with the principle of the probability distribution of attribute values, a definition of β-dominance relation is given in this paper. The relationship is able to compare the dominance relation of any two objects in the universe. Then, a definition of approximation reduction based on β-dominance relation is defined and developing a calculation method of the approximate reduction with discernibility matrix.(4) In generalized incomplete order information systems, a definition of a the degree of dominance relation is given. Using the granularity entropy of the a the degree of dominance relation as the heuristic information, an attribute reduction algorithm is proposed. Finally we show by an example that the heuristic reduction algorithm is effective.(5) In incomplete order decision information systems, the object of "at least" and "at most" determined decision rules are obtained from the lower approximation set and the object of the "at least" and "at most" possible decision rules are obtained from the bounded approximation set. And it shows by examples that the method can obtain decision rules of an information system.