多任务学习的不平衡SVM+算法

Multi-task learning of SVM+ for imbalanced classification

处理不平衡数据分类时,传统支持向量机技术(SVM)对少数类样本识别率较低。鉴于SVM+技术能利用样本间隐藏信息的启发,提出了多任务学习的不平衡SVM+算法(MTL-IC-SVM+)。MTL-IC-SVM+基于SVM+将不平衡数据的分类表示为一个多任务的学习问题,并从纠正分类面的偏移出发,分别赋予多数类和少数类样本不同的错分惩罚因子,且设置少数类样本到分类面的距离大于多数类样本到分类面的距离。UCI数据集上的实验结果表明,MTL-IC-SVM+在不平衡数据分类问题上具有较高的分类精度。

When dealing with imbalanced datasets,the traditional support vector machine(SVM)has a low rate of identification on the minority class.Inspired by that SVM+can utilize the additional information hidden in the training data,this paper proposed a new support vector machine called multi-task learning SVM+for imbalanced classification(MTL-IC-SVM+).MTL-IC-SVM+represented the classification of imbalanced datasets as a multi-task learning problem based on SVM+.From the perspective of correcting the deviation of the classification surface,different misclassification penalty factors were assigned to the majority and minority class samples respectively,and the distance from the minority class samples to the classification surface was set to be larger than that from the majority class samples.Experimental results on the UCI datasets show that the proposed methods lead to very encouraging results on imbalanced datasets.