基于关联规则的分类规则约简方法

摘要

分析分类规则内属性之间的相关性,提出一种分类规则约简方法.针对原始训练集构造FP树,获取相应的关联规则集,对关联规则后件属性(集),采用置信度α描述该属性(集)相对于其所在分类规则的重要程度.在分类规则集中,约简α值小于阈值η的属性,从而约简分类规则长度.利用UCI机器学习及SDSS DR7数据进行实验,结果表明该方法具有较高的分类效率.%This paper proposes a classification rule reduction method by analyzing the correlation of attributes in classification rules. It obtains the association rule set by analyzing the correlation among the attributes of training set, describes the importance degree in the classification rule by using the degree of confidence a of the association rule. The later part of the association rule, whose a is larger than threshold value 77, is deleted in the classification rule. Experimental results validate that this method has higher classification effectiveness by using UCI and SDSS data as the decision system.