基于条件库覆盖的分类算法  

Classification Approach Based on Conditional Database Covering

作　　者：曹兰 李雪静[1] CAO Lan;LI Xuejing(Electronic Information Department,Zhangzhou Institute of Technology,Zhangzhou,Fujian 363000,China)

机构地区：[1]漳州职业技术学院电子信息学院,福建漳州363000

出　　处：《九江学院学报（自然科学版）》2022年第2期65-69,共5页Journal of Jiujiang University：Natural Science Edition

基　　金：福建省中青年教师教育科研项目(科技类JAT191419,科技类JAT210844);漳州职业技术学院校级课题(ZZY2021B042)的成果之一。

摘　　要：支持度-置信度的关联分类算法主要有两个不足:①通常会产生大量置信度不高的规则;②支持度和置信度的阈值的设定,常常使得规则不能覆盖属性值个数少的实例和少类实例。针对这些问题,文章提出了一种新分类算法。首先,根据支持度值较低的属性值,提取能覆盖所有的实例的属性值。其次,把提取的每个属性值所覆盖的实例当成一个独立可取出规则的条件库,即条件库作为独立的小训练集。最后,在每个独立的条件库里,规则中的每个属性值都是选取最好的属性值,直到所产生规则覆盖整个小训练集。实验结果表明,在25个UCI数据集上,该算法所产生的每条规则具有置信度高,且整个实例被产生的规则覆盖,规则的质量明显提高,其准确率比其他算法在大多数数据集上得到了显著提高。Traditional classification algorithms adopted cover technique to build classifier.So,an example was covered once by a rule,and some quality rules were missing.There was a great influence of support and confidence of associative classification method on the extracted rules of each class,setting high confidence often made some rules couldnot be extracted and declined classification accruacy.This paper proposed a new classification approach,CCCA(Conditional Cover Classification Approach).This new approach had three main steps:firstly,CCCA generates attribute-values that had lower support,and constructed the conditional database contained the instances with the attribute-value;secondly,every conditional database with the attribute-value would be a small database,where rules come from.At last,in every condition database,every rule was consisted of the best attribute-value,each conditional database was covered by the rules.The experiment showed that extracting classifier of our algorithm was consisted of a lot of the rules with higher support,and every instance of the database could be covered by some rules with higher quality,that had achieved high accuracy than the others in the 25 databases.

关 键 词：数据挖掘 关联分类 条件库 支持度 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论]