Research On Knowledge Reduction And Rules Fusion Method In Information System

Rough Set Theory (RST) proposed by a Poland Scholar Z.Pawlak in 1982 is a new mathematical tool deal with uncertain knowledge. After 20 years of study and development, RST has achieved great improvement both in theory and practical use. Not depending on any prior information except needed data set, RST can analyze and process the imprecise, uncertain or incomplete data and knowledge. Due to its successful application in such fields as knowledge discovery in late 1980s and early 1990s, RST has drawn wide attention from many scholars in the world. Currently, RST has been applied in many fields as artificial intelligence, machine learning, data mining, decision support and analysis, process control, pattern recognition, fault detection.Centered round knowledge reduction and knowledge discovery in information systems, the thesis has made the following studies.1. Since the core of the set of attributes is the intersection of all the reductions, many current reduction algorithm of the attributes originate from it, in which one has to make use of the heuristic information (the quantity of relative information, the relative significance) to add the more important attributes to the core of the attributes set step by step until works out a reduction of the original information system. However, this thesis puts forward a new method in working out the core of the attributes set. This method successfully avoids the weakness of conventional algorithm what frequently makes operation of intersection between equivalence matrices and shortens the time complexity of algorithm by the conventional one.2. Knowledge reduction is deletion of irrelevant or unimportant attributes with keeping some properties of knowledge base. At present, many scholars have made deep studies of knowledge reduction and achieved a lot. However, these studies were conducted under complete information systems and incomplete information systems with characteristic value. For many reasons, there exists in reality a kind of incomplete information system with fuzzy objectives, which means that conditions attributes are incomplete and objective attributes are fuzzy. Up to now few scholars have conducted further study of these information systems. Based on the tolerance relation of the incomplete information system, this thesis proposes approximation reduction,εreduction in incomplete information systems with single fuzzy objective, andεreduction is an extension form of approximation reduction. Also the thesis proposes approximation reduction and distribution reduction in the incomplete information systems with multi fuzzy objectives, and corresponding discernibility matrices of various reductions, so provides effective method for calculation of various reductions.3. The main subject of the RST is information systems. Knowledge discovery mainly refers to concept discovery in information systems and rule decision in objective information systems. To various objective information systems, the study on objective information systems is to get the decision rules. The number of decision rules is very limited. The thesis introduces a weighted inclusion degree on set vector space, proposed a fusion method of decision rules in consistent objective information systems. This method is the access to all the decision rules in consistent objective in formation systems. Namely, we can draw the corresponding decision values through this method in all the combinations of condition attributes values.