基于信息赢取的适应度函数

Fitness Function Based on Information Gain

用遗传算法挖掘一阶规则的关键在于如何准确地评价一阶规则,即规则的适应度能有效地区分规则的优劣,从而指导算法逼近目标规则。该文在绑定概念的基础上,依据信息理论,提出了新的基于信息赢取的适应度函数。相比通常采用的基于规则覆盖的正、负例数目的评判标准,新的适应度函数能充分利用隐藏在例子和背景知识中的信息,更准确地量化一阶规则的优劣,从而提高算法的搜索性能和规则的可读性。

The key of using genetic algorithm to mine first-order rules is how to precisely evaluate quality of first-order rules, that is,the fitness of rules can score their quality rightly and effectively guide genetic algorithm to be close to the target rule.By adopting the concept of bingding and information theory, a new fitness fuction based on information gain is proposed. Comparing with the common evalution criteria based on amount of examples covered by rules, the new fitness function,by throughly utilizing the information hided in the examples and the backgroud knowledge, measures quality of first-order rules more precisely, and enhances the search performance of algorithm and the readability of rules.