一种基于相似性的规则集一致性度量的新方法

规则学习算法通过学习样本产生规则集,如何判断规则集的好坏?目前规则集的评估标准有很多,如一致性、可测量性和易理解性评估,但它们有各自的缺点。提出一种新的评估规则集方法:相似性度量。这种度量方法可以计算出两个规则集之间的正相似性与负相似性。实验说明这种新的度量方法可以被用来评估规则集间的一致性,并且可以决定使用哪种算法解决某类问题或选择组合分类模型中的基模型。

A New Measuring Rule Set Consistency Method Based on Similarity

The rule extraction algorithm produces the rule set by learning examples.How to evaluate the rule set? There are several evaluation criteria for rule set at the present time,such as consistency,measurability and comprehensibility,but they have various disadvantages.Proposes a new evaluation measurement: similarity.The evaluation measurement can measure positive similarity and negative similarity between two rule sets.The experiment shows this new measurement can be used to measure consistency between rule sets and decide to choose algorithm or select the elemental model of combination classification model.