Rough Set Based Uncertain Information Processing And Knowledge Acquisition

With the explosive growth of the computer techniques and Internet, data and information have been generated very rapidly. This expansive growth of data leads to a requirement of the development of more powerful techniques to convert the huge and mess data into valuable information and knowledge. It is a challenge for the study of intelligent information processing. Thus, data mining becomes one of the key research fields in artificial intelligence.Among many methods of data mining, rough set theory is an effective method for handling complex systems because it's significant advantage of not requiring any prior knowledge except data sets compared with some other theories like probability theory, fuzzy set and evidence theory, etc. It has been applied successfully in many domains such as pattern recognition, intelligent control, medical data analysis, fault diagnosis and so on.However, there are still some important problems of rough set theory to be addressed as a method for data mining. For example, no mechanism to handle the uncertainty or imprecise of original data, high complexity for processing huge data sets and low performance on processing decision tables with continuous value attributes. It is becoming an important research topic in rough set fields to extend the existing theories and approaches of rough set to deal with imprecise or uncertainty original data. More and more researchers are interested in it.In this dissertation, we summarize the research status of theories and methods about rough set at first. Then, we have a clue which the existing research problems of rough set for handling the uncertainty or imprecise of original data and present the research contents and objectives of the dissertation.The major research results achieved in this dissertation are as follows:(1) Approaches for attribute reduction and knowledge acquisition from fuzzy decision information systems are proposed.A variable precision fuzzy rough data model (VPFRDM) is proposed by extending Ziarko's variable precision rough set theory (VPRS). A VPFRDM based heuristic algorithm for attribute reduction from fuzzy decision information systems is developed. Then, through calculating the classification quality of each fuzzy pattern classes to the decision categories, a method for knowledge acquisition from fuzzy decision information systems is developed. Simulation experiment results show that VPFRDM is effective and has better data generalization ability compared with Ziarko's VPRS.(2) Through integrating rough set theory and vague set theory, the models of vague rough set and rough vague set are proposed.As an extension of fuzzy set theory, vague set theory is one of the most vitality research aspects of fuzzy information processing, and is attracting more attention of researchers. In this dissertation, two generalized models of Pawlak approximation sets, rough vague set and vague rough set, are proposed through integrating rough set and vague set. The two models could be used to improve the performance of rough set theory based data mining on original fuzzy data. The algebra properties of these two models are also studied.(3) Based on rough vague set and vague rough set, attribute reduction and knowledge acquisition methods in vague objective information systems (VOIS) and vague decision information systems (VDIS) are developed respectively.Through defining the rough vague lower approximation distribution and the vague rough lower approximation distribution, the concepts of attribute reduction are proposed in VOIS and VDIS respectively. Then, algorithms based on discernibility matrix for computing attribute reduction are developed. At last, the viable approaches for extracting decision rules from VOIS and VDIS are proposed. These results extended the traditional methods of classical rough sets theory, and provided a new way for uncertain fuzzy knowledge acquisition.(4) Attribute reduction and knowledge acquisition from decision information systems containing continuous value attributes are proposed.Rough set based attribute reduction and knowledge acquisition methods are mainly applicable to information systems containing discrete values. For knowledge acquisition from decision information systems with continuous value attributes, a new definition of positive region of rough set is proposed. Using related statistical methods, a criterion for measuring the significance of continuous attributes is proposed. Then, an approach for knowledge acquisition from decision information systems containing continuous values (CDISKA) is developed. Simulation experiment results show that the CDISKA algorithm has better performance in classification accuracy compared with the classical rough set approaches and decision tree approach in processing decision information systems containing continuous values attributes.