面向不平衡数据集的一种基于SMOTE的集成学习算法

An Ensemble Learning Method Based on SMOTE for Imbalanced Data Set

合成少数类过采样技术(Synthetic Minority Over-sampling Technique,SMOTE)是一种用于处理不平衡数据的过采样方法,B-SMOTE在SMOTE的基础上把少数类的边界样本与少数类样本插值合成新训练样本。RB-SMOTE在B-SMOTE的基础上对少数类的边界样本进行区分,精细化分配新增样本的数量,从而提高了不平衡数据的分类精度。提出了面向不平衡数据集的一种基于SMOTE的集成学习算法,该方法通过RB-SMOTE合成不平衡率不一的多个新训练样本,组成相应的多个基分类器,再采用投票的方式对测试样本进行分类。仿真实验表明基于SMOTE的集成学习算法能有效改善不平衡数据集的不平衡性。

One of the important over-sampling approaches to solve the classification problem of imbalanced data is SMOTE(Synthetic Minority Over-sampling Technique). Based on SMOTE method, B-SMOTE generates new synthetic samples over-sampling the minority samples near the borderline. Based on B-SMOTE method, RB-SMOTE generates different number of new synthetic samples according to different borderline minority samples. In this paper, we proposed an ensemble learning method based on SMOTE for imbalanced data set which generates several new samples with different imbalanced rates to form corresponding base classifiers. Then the test samples are classified by voting. The experimental results show that the algorithms could improve the classification performance of imbalance data sets effectively.