基于排序的关联分类算法

Classification Mining Using Association Rules Based on Rule Ranking

提出了一种基于排序的关联分类算法。利用基于规则的分类方法中择优方法偏爱高精度规则的思想和考虑尽可能多的规则,改进了CBA(Classification Based on Associations)只根据少数几条覆盖训练集的规则构造分类器的片面性。首先采用关联规则挖掘算法产生后件为类标号的关联规则,然后根据长度、置信度、支持度和提升度等对规则进行排序,并在排序时删除对分类结果没有影响的规则。排序后的规则加上一个默认分类便构成最终的分类器。选用20个UCI公共数据集的实验结果表明,提出的算法比CBA具有更高的平均分类精度。

A new associative classification algorithm based on rule ranking was proposed. The proposed method takes advantage of the optimal rule method preferring high quality rules. At the same time, it takes into consideration as many rules as possible,which can improve the bias of CBA that builds a classifier according to only several rules covering the training dataset. In the proposed algorithm, after the generation of association rules whose consequences are class labels, rules are ranked according to their length, confidence, support, lift and so on. Rules having no influence on the classification result are deleted during ranking. The set of the ranked rules with a default class constructs the final classifier. Finally,20 datasets selected from UCI ML Repository was used to evaluate the performance of the method. The experi- mental results show that our method has higher average classification accuracy in comparison with CBA.