Research On Methods Of Classification Based On Granular Computing

KDD, Knowledge Discovery in Database, is a novel technology of intelligent informatign processing for knowledge discovery in large database. Classification is a very important task in Data Mining. It builds a model according to the data whose class labels are known, and then uses this model to predict the classes of the data whose class labels are unknown.The idea of granular computing was emerged in 1970's. Its main idea is to analogy the manner of people's thinking. That is people can observe and a.nalysis problem from quiet different granularity, and can easily transfer from one granularity world to another. In recent years, people have begun to apply granular computing to data mining, and achieved some good results. It has become a new and prospective research topic in data mining.In this thesis, we apply granular computing to classification and do some primitive research. Our work mainly includes:1. In this thesis, the author carries out a deep and comprehensive analysis towards classification problem: discussing its intension (Classifier Construction), extension (Feature Selection, Rule Extraction) and nature. And the thesis also concentrates on present hot research issues of classification, then study the granular computing in classifier construction.2. We discuss the granular principle of classification. First, independent of idiographic algorithms, data classification modeling based on granular computing is studied with Granular Computation after systematically reviewing related work; and run-of-mill laws are drown out.3. The classification of incomplete information systems is studied. According to the human cognitive rules, people can utilize limited knowledge to obtain rather satisfactory results at a very simple knowledge level in order to avoid the incomplete information at a deep knowledge level. Based upon this rule, an algorithm is developed which combines the uses of the granularity model in the quotient space theory and the Rough Set theory. After projecting attributes, the samples are processed with rough granular to provide a decision consistence systems. This algorithm enables classification of an incomplete information system by utilizing the existing samples at hierarchy levels. It therefore reduces the limitations of most current algorithms that can only be applied in the classification of complete information systems. As consequence, the proposed algorithm should expand the application scope of such classifiers which are used to classifier the complete information system.Main conclusions and significances of this thesis are as following:1. A data set is expressed in granular forms and then data classification modeling is educed based on Granular Computing.2. The classification of incomplete information systems is solved by combining the granularity model and the Rough Set theory.