Research Of Attribute Generalization Towards Classification Based On Domain Knowledge

Data mining become more and more important due to the sharply increase in data scale. Nowadays, data mining techniques always use original level of the data to mine, however, the value of the data can be level-exchanged according to the customer's demands in real applications. Because of the abundance and complexity in the real fields, many attributes have different methods and values to generalize under different conditions (the Multi-relational and Multi-level, MRML for short), and there is tremendous difference between data values and the relativity of problems. Therefore, the problem of generalization towards multi-relational and multi-level attributes is researched by this dissertation. The organization of this dissertation is as follows:(1)Different knowledge representation models in the field of domain knowledge are discussed in detail. The significant role of domain knowledge represented in different models in the process of data mining is discussed. The future applications and challenges of knowledge discovering, based on the domain knowledge, are exhibited.(2)Basic concept and representation of the concept hierarchy is described. And a MRML generalization model is constructed to express the relation of attributes in specific multi-generalization ways. A generalization model for classification is constructed to control level-exchanging of attributes.(3)A generalization method for obtaining classification rules (CG_DK), based on MRML, is proposed. This method chooses the most compact generalization level and way to generalize the attributes controlled by misclassification ratio. This method can obtain the best classification rules according to the individual demands.(4) Based on the work stated above, a prototype system which is Multi-relational and Multi-level Attributes Generalization towards classification is implemented.