基于AdaBoost的类不平衡学习算法

处理类不平衡数据时,少数类的边界实例非常容易被错分。为了降低类不平衡对分类器性能的影响,提出了自适应边界采样算法(AB-SMOTE)。AB-SMOTE算法对少数类的边界样本进行自适应采样,提高了数据集的平衡度和有效性。同时将AB-SMOTE算法与数据清理技术融合,形成基于AdaBoost的集成算法ABTAdaBoost。ABTAdaBoost算法主要包括三个阶段：第一个阶段对训练数据集采用AB-SMOTE算法,降低数据集的类不平衡度；第二个阶段使用Tomek links数据清理技术,清除数据集中的噪声和抽样方法产生的重叠样例,有效提高数据的可用性；第三个阶段使用AdaBoost集成算法生成一个基于N个弱分类器的集成分类器。实验分别以J48决策树和朴素贝叶斯作为基分类器,在12个UCI数据集上的实验结果表明：ABTAdaBoost算法的预测性能优于其它几种算法。

AdaBoost-based class imbalance learning algorithm

When dealing with unbalanced data sets, the borderline examples of the minority class are more easily misclassified. To reduce the impact of class imbalanced about the performance of classifier, this paper presents a Adaptive Borderline-SMOTE(AB-SMOTE). AB-SMOTE algorithm samples the boundary samples of the minority adaptively, which improves the degree of balance and efficiency of the data sets. At the same time, the AB-SMOTE algorithm is combined with the data cleaning technology to form a new ensemble algorithm ABTAdaBoost based on AdaBoost. ABTAdaBoost algorithm consists of three stages: in the first stage, the training data sets adopt AB-SMOTE algorithm to reduce the degree of imbalance of data sets; in the second stage, using Tomek links data cleaning techniques to remove the noise and overlapping instances which are introduced from sampling methods in the data sets, the availability of data is improved at the same time; in the third stage, using the AdaBoost algorithm to generate an ensemble classifier based on N weak classifier. Experiment uses J48 decision tree and naive bayes as the base classifier respectively, the results show that ABTAdaBoost algorithm has the best overall performance compared with other algorithms in 12 UCI data sets.