基于AdaBoost的类不平衡学习算法

发布时间：2018-07-29 20:50

【摘要】：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Ada Boost的集成算法ABTAdaBoost。ABTAda Boost算法主要包括三个阶段:对训练数据集采用AB-SMOTE算法,降低数据集的类不平衡度;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;使用Ada Boost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,ABTAda Boost算法的预测性能优于其他几种算法。
[Abstract]:When dealing with class imbalance, the boundary instances of a few classes are easily misclassified. In order to reduce the influence of class imbalance on the performance of classifier, an adaptive boundary sampling algorithm (AB-SMOTE). AB-SMOTE algorithm is proposed to self-adaptively sample a few kinds of boundary samples, which improves the balance and effectiveness of the data set. At the same time, the integration of AB-SMOTE algorithm and data cleaning technology is combined to form the ABTAdaBoost.ABTAda Boost algorithm based on Ada Boost, which includes three stages: to reduce the class imbalance of the data set by using AB-SMOTE algorithm to the training data set; The Tomek links data cleaning technique is used to remove the noise in the data set and the overlapping sample generated by the sampling method, which effectively improves the availability of the data, and uses the Ada Boost integration algorithm to generate an integrated classifier based on N weak classifiers. J48 decision tree and naive Bayes are used as basis classifiers respectively. The experimental results on 12 UCI datasets show that the prediction performance of UCI Boost algorithm is superior to that of other algorithms.
【作者单位】：南通大学电子信息学院;南通大学计算机科学与技术学院;南通理工学院计算机与信息工程学院;
【基金】：国家自然科学基金资助项目(NSF61202006/61272424) 计算机软件新技术国家重点实验室开放课题(KFKT2012B29) 江苏省自然科学基金资助项目(BK2010277) 江苏省科技创新基金资助项目(BC2013167)
【分类号】：TP181

【相似文献】