集成学习算法在不平衡分类中的应用研究

Using Ensemble Learning Strategy to Handle Class Imbalance Problems

提出一种基于训练集分解的不平衡分类算法,该算法使用能输出后验概率的支持向量机作为分类器,使用基于测度层次信息源合并规则实现分类器的集成。在4个不同领域的不平衡数据集上的仿真实验表明:该算法有效提高分类器对正类样本的正确率,同时尽量减少对负类样本的误判。实验结果验证集成学习算法处理不平衡分类问题的有效性。

Based on the strategy of partitioning training set, an algorithm was proposed to handle class imbalance problems. Support vector machines which can output posterior probability were used as base classifiers and then combined by a rule of information fusion at measure - level. Experimental results on four problems in different fields show that the proposed algorithm cart get higher classification accuracy to positive class while does its best to decrease the classification error to negative class.