Research Of Knowledge Discovery Application Based On Rough Sets Theory

With the rapid development of modern information technology, people are no longer be satisfied with making simple statistical analysis and scientific calculation by computer, but putting forward an intellectualized demand to it, such as Computer- Aided Design, Computer Assisted Instruction, Computer-Aided Decision Making and so on. In this situation, a new technology, Knowledge Discovery in Database (KDD) appears. Additionally, with the increasingly expand of the realistic database, Data Preprocessing is also given more and more attention in recent years, as one of the important front-work of Knowledge Discovery in Database. The results of Data Preprocessing have a directly impact on the efficiency and effect of KDD subsequents. Rough Sets theory has great superiority in Data Preprocessing because of its particular expression of knowledge, as well as it makes an effective means of Knowledge Discovery in Database.In this thesis, the concept of Knowledge Discovery in Database is introduced first, as well as its contents and applicating fields. The task and general procedure of KDD has also been expounded. Then we discussed the theory frame, the basic concept and the core of RS. As one of the core of this thesis, we make a systemic study in Data Preprocessing of KDD. Above all, we construct a Procedure Model of Data Preprocessing and state its main idea, combining to primary applicating flat of KDD, namely data warehouse. Furthermore, our study emphasis on the dimension reduction in Data Preprocessing. After summarizing the current dimension reduction algorithms, we propose Bi-directional Selection Dimension Reduction Algorithm (BSDRA), a new dimension reduction algorithm. Another main topic of this thesis is rule extraction algorithms based on Rough Sets in which we set forth the concepts of compatible rule and incompatible rule. Considering that the Coverage Factor might mis-delete the valuble rules because of noise, we amended and improved the Coverage Factor and introduced the concept of Belonging Degree. We also defined it in mathematical methods and added to the algorithm regarding it as the rule of filtrating acoustic noise factor. We propose the rule extraction algorithms with Belonging Degree. In the end of this thesis, a KDD application is carried out in order to verify the validity of the above research, combined with the author's job, which studying the Data Preprocessing and Knowledge Discovery in a database of Teacher's...