The Study Of Knowledge Discovery And Attributes Reduction In Set-valued Information Systems

Rough set theory is a mathematical tool in data analysis and can be effectively used to handle incomplete, imprecise and uncertain information. The theory has been successfully applied to various fields such as data mining, machine learning and pattern reeognition. One important application of rough set is the knowledge discovery and knowledge reduction in information system.Set-valued information system is a generalized model of single-valuedinformation system;it can be used to characterize uncertain information and insufficient information. With rough set theory as the tool, and the related outcomes of researches as the basis, the current dissertation systematically studies knowledge acquisition and attributes reduction for set-valued information and set-valued decision table. The primary contributions can be summarized as follows:1. The knowledge acquisition and attributes reduction of set-valued information system based on the variable precision tolerance relation are studied. By introducing a variable precision tolerance relation to set-valued information system, the attribute reduction under a certain similar level are proposed; the influence of similar level on the attribute reduction is analyzed; the extraction of generalized decision rules and optimal generalized decision rules from a set-valued decision table are discussed. Furthermore, in order to simplify the knowledge representation, the dissertation also studies the relative reduction and the assignment reduction of set-valued decision table based on the variable precision tolerance relation, and gives some equivalent conditions and the important properties of these reductions.2. Based on the maximum variable precision tolerance classes, the dissertation studies the knowledge acquisition and attributes reduction of set-valued information system. With the maximum variable precision tolerance classes as basic sets, the attribute reduction and relative reduction are discussed, the inherent relationship between the attribute reduction based on the variable precision tolerance relation are given. Furthermore, the dissertation discusses the generalized decision rules and optimal generalized decision rules of the set-valued decision table based on the maximum variable precision tolerance classes.3. The knowledge acquisition and attributes reduction of set-valued information system based on neighborhood relations are studied. Through the introduction of neighborhood relations, the neighborhood reduction of set-valued information system is proposed, the definition and extraction method of neighborhood decision rules in set-valued decision table are given. In order to derive optimal neighborhood decision rules, the dissertation also discusses the positive region reduction and approximate distribution reduction, and obtains approaches to these reductions.4. Based on the dominance relation, the dissertation studies the knowledge acquisition and attributes reduction of set-valued ordered information system. By introducing a dominance relation to set-valued ordered information system, the criteria reduction and the ranking approach for objects are given and the dominance rough set models are established based on dominance classes. Furthermore, the extraction method of dominance rules in set-valued ordered decision table is proposed, the approximate distribution reduction and boundary region reduction are discussed.5. The feature reduction and rule extraction of C-wavelengths of radio signals are studied. By a new discretization method with the data, the database of C-wavelengths of radio signals are converted into a set-valued information system, and the feature reduction and rule extraction of C-wavelengths of radio signals are discussed based on proposed variable precision rough set approach.