标题:
新型二分类支持向量机P2M-SVMNew Binary Classifier P2M-SVM
作者:
梁锦锦, 吴德
关键字:
可能性二均值聚类, 半监督二分类支持向量机, 全局最优, 稳健性, 泛化能力P2M; Semi-Supervised; SVM; Robustness; Generalization Ability
期刊名称:
《Artificial Intelligence and Robotics Research》, Vol.2 No.1, 2013-02-21
摘要:
提出基于可能性二均值聚类(Possibilistic Two Means, P2M)的二分类支持向量机(Support Vector Machine, SVM)。该算法先用P2M对未知类别的二分类数据进行划分，然后利用支持向量机对划分后的数据进行训练。人造数据和UCI数据上的分类实验表明，该算法综合利用了P2M聚类的稳健性和SVM分类的强泛化能力，提高了传统聚类的分类精度并降低了SVM的类别采集代价。A semi-supervised binary support vector machine (SVM) is proposed based on possibilistic two-means (P2M) clustering. First, divide the unlabeled data using PCM; then, train the labeled data using SVM. Experiments on artificial and UCI data show the superiority over existing algorithm. P2M-SVM utilizes both the robustness of P2M for binary clustering and the strong generalization ability of SVM for classification thus increases the classification accuracy of traditional clustering and reduces the cost of sample collecting of the SVM.