基于数据挖掘的非单调问题的缺省规则框架

Default Rules Frame of Non-monotonous Problems Based on Data Mining

提出一个求缺省规则的框架,通过合并条件属性所决定的类,生成组合类,可以构造覆盖更多对象的规则,生成从这些组合类映射到占优决策的规则。结果规则比确定规则至少具有两个重要的优点:(1)结构上简单;(2)即使规则相对训练集可能不完全,但是当处理未见的新事例时将表现得更好。系统对未来对象的分类质量,将在很大程度上依据系统一般化知识的能力。

It proposes a frame in finding the default rules. Through the category decided by combining conditions attributes, the new combining category can create rules to cover more objects. These combining categories reflect the advantageous rules. In contrast to the fixed rules, the result rules have at least two main advantages.. 1 ） simplicity in structure; 2） better performance in dealing with new cases despite its incompletion comparing with training set. System ＇s classification quantity to future object will according as ability of system generalized knowledge to a large extent.