基于ADASYN与AdaBoostSVM相结合的不平衡分类算法

Joint ADASYN and AdaBoostSVM for Imbalanced Learining

对于平衡数据集支持向量机(support vector machine,SVM)通常具有很好的分类性能和泛化能力,然而对于不平衡数据集,SVM只能得到次优结果,针对该问题提出了一种基于SVM的AS-Ada Boost SVM分类算法.首先,通过使用ADASYN采样,提高少类样本在边界区域的密度;然后,使用基于径向基核支持向量机(radial basis function kernel mapping support vector machine,RBFSVM)模型弱分类器的Ada Boost SVM算法训练得到决策分类器.通过将该算法在各种不平衡数据集上的测试结果与单纯运用ADASYN技术、Ada Boost SVM、SMOTEBoost等其他分类器进行比较,验证了该算法的有效性和鲁棒性.

For a balanced data set support vector machine(SVM) generally has good performance and generalization,but SVMs can only produce suboptimal results with imbalanced data sets.In this paper,a AS-Ada Boost SVM algorithm was proposed based on SVM.First,by using ADASYN sampling,the density of small class sample in the border area was improved.Then,the decision classifiers was achieved by using RBFSVM as the weak classifiers in Ada Boost algorithm.By comparing the test results on a variety of unbalanced data sets with ADASYN, Ada Boost SVM, SMOTEBoost, it shows that the proposed algorithm is effective and robust.