基于抽样和规则的不平衡数据关联分类方法

不平衡数据的出现给传统关联分类算法带来了巨大的挑战.为了提高关联分类方法对不平衡数据集的分类精度,本文分别从数据和规则层次着手,提出了关键值抽样法(key value sampling,KVS)和规则验证法(rule validation,RV).关键值抽样法通过增加与少数类相关性强的数据,减少与多数类相关性弱的数据来达到数据类分布平衡.避免了大量有效信息的流失,并且增强了与少数类相关性强的数据信息.规则验证法对初步生成的分类器进行了规则验证,并对分类性能不好的规则进行调整,从而保证了分类器中规则的质量.实验表明,本文中的研究方法能够有效提高关联分类方法处理不平衡数据的精度.

Sample and rule centric approach for associative classification on imbalanced data

The emergency of imbalanced data has brought a great challenge for AC method. To improve AC's performance on imbalanced data, this paper presents key value sampling (KVS) method and rule validation (RV) method respectively from data and rule processing. KVS samples the original imbalanced data and achieves class balance by removing the instances weakly correlated with majority class and increasing those strongly correlated with minority class, which can prevent a lot of useful information from losing and highlight the useful information related with minority class. RV method is to validate the initially generated classifier and improve the rules with bad performances, which can enhance the whole classifier's performance. Through experiment analysis, the methods in this paper can improve the performance of AC on imbalanced data classification.