基于邻域混合抽样和动态集成的不平衡数据分类方法-计算机科学.PDF

Vo1. 44 No.8 第 44 卷第8 期 计算机科学 2017 年 8 月 COMPUTER SCIENCE Aug.2017 基于邻域混合抽样和动态集成的不平衡数据分类方法 高锋黄海燕 (华东理工大学信息科学与工程学院 上海 200237) 摘 要 不平衡数据严重影响了传统分类算法的性能，导致少数类的识别率降低。提出一种基于邻域特征的混合抽 样技术，该技术根据样本邻域中的类别分布特征来确定采样权重，进而采用混合抽样的方法来获得平衡的数据集;然 后采用一种基于局部置信度的动态集成方法，通过分类学习生成基分类器，对于每个检验的样本，根据局部分类精度 动态地选择最优的基分类器进行组合。通过 UCI 标准数据集上的实验表明，该方法能够同时提高不平衡数据中少数 类和多数类的分类精度。 关键词 数据挖掘，不平衡数据，K 近邻，混合抽样，集成学习 中固法分类号 TP391 文献标识码 A DOI 10. 11896/j. issn. 1002-137X. 2017. 08. 038 Imbalanced Data Classification Method ßased on Neighborhood Hybrid S创npling and Dynamic Ensemble GAO Feng HUANG Hai-yan CSchool of Information Science and Engineering ,East China University of Science and Technology ,Shanghai 200237.China) Abstract The class imbalance problems severely affect the performance of the traditional classification algorithm ,lea ding to decrease the recognition rate of the minority. In order to solve this problem ,a hybrid sampling technology based on neighborhood characteristic was proposed to enhance the classification accuracy of minority class. This technology changes the sampling weight according to the class distribution in the samples neighborhood ,and uses the hybrid sar口p ling to obtain the balanced data subset. Then the base classifiers are generated ,for each test sample ,a dynamic ensemble method based on local confidence is proposed to select the optimal base classifier sets. The experiments on UCI datasets show that the method has high classification accuracy rate of both minority and m句ority class for imbalance datasets. Keywords Data mining , Imbalanced data ,K-nearest neighbor , Hybrid sampli晤，Ensemble learning