DEC算法是一种基于支持向量机的改进算法,但是研究表明,对于某些不平衡数据集,DEC分类器不能达到较高的少数类分类精度。针对这种情况,提出了一种基于欠抽样方法和DEC方法结合的新算法US-DEC来提高少数类分类精度。先用一个新的欠抽样方法来预处理数据集,再用DEC分类器训练处理过的数据集。通过在七个不平衡数据集上的仿真实验,比较了US-DEC算法和支持向量机,欠抽样方法以及DEC方法的分类性能。实验结果表明,在提高每类数据集的少数类分类精度上,US-DEC算法要优于其他算法。%Different Error Costs (DEC) was a modified algorithm based on Support Vector Machine (SVM),but for the some imbal-anced datasets,it may fail to reach high classification accuracy. In view of this condition,a new algorithm US-DEC based on the under-sampling combined with DEC was presented. In the algorithm,first a new undersampling method was used to preprocess the imbalanced datasets,then DEC classifier was trained on the processed imbalanced dataset. The experiments on seven UCI imbalanced datasets show that the generalization performance of US-DEC is compared with SVM,undersampling,and DEC,the results indicate that this algorithm outperforms all of them.
展开▼
