一种基于投票的不平衡数据分类集成算法

不平衡数据分类是机器学习的研究热点之一。传统的机器学习分类算法通常假定用于训练的数据集是平衡的,不能直接应用于不平衡数据分类。利用朴素贝叶斯和决策树对数据不平衡的敏感度不同,提出一种基于投票的不平衡数据分类集成算法。基分类器选择NB和C4.5,通过投票平均方法进行分类决策;并选择公开的不平衡数据集进行实验验证。实验结果表明,该算法能有效提高不平衡数据的分类性能,特别是对正类(少数类)的误报率较低,具有良好的鲁棒性。

An Ensemble Algorithms based Voting for Imbalanced Data Classification

Imbalanced data classification has become the hot research in machine learning. Traditional machine learning classification algorithm is usually assumed for the training data set is balanced, the imbalanced data cannot be directly learned. Naive Bayes and decision trees for different imbalanced sensitivity of classification, we proposed an ensemble algorithms based voting for imbalanced data classification. NB and C4.5 were selected for base classifier, classification decisions by voting average method, and select public imbalanced datasets for experimental verification. Experimental results show that the algorithm can effectively improve the classification performance of imbalanced data, especially the low FPR and highSrobustness for the positives minority class.