利用广义粗近似挖掘缺省规则

MINING DEFAULT RULES BASED ON GENERALIZED ROUGH APPROXIMATIONS

由于对称性和传递性在一些情况下是不必要的,因此以仅有自反性的广义相似关系为基础研究缺省规则的发现是十分有意义的工作。本文首先给出新关系的形式和特点,再利用它得到挖掘缺省规则的广义粗近似框架,指出属性上的关系是反映客观世界之间联系的本质因素。

A great deal of work done in data mining has focused on the generation of rules from the training data with entirely consistency. In such cases, definite rules that map all objects into the same decision class may be generated. There is also a clear need for reasoning in the presence of inconsistencies in many cases. Also, if objects are classified inconsistently, we still want to be able to generate rules that reflect the normal situation. Such normalcy rules typically sanction a particular conclusion under given information, and then with additional knowledge the previous conclusion may be invalidated. The rought set approach was designed as a tool to deal with uncertain or vague knowledge. Classical definitions of lower and upper approximations were originally introduced with reference to an indiscernibility relation, which was assumed to be an equivalence relation. Extending indiscerni-bility to generalized similarity imposes on weakening some of the properties of the binary relation in terms of reflexivity, symmetry and transitivity. Generalized similarity relation retains the reflexivity property only. In this paper the form and characteristic of new similarity relation are given, then generalized approximating frame for mining default rules based on generalized similarity class is obtained. A point is made that the binary relation on attributions is a natural factor to reflect relationship among objects. Which default rules will be generated in the process depend upon the setting of threshold value μt. It is also a result of the desired confidence that the user wishes to have in the generated rules.