基于Help-Training 的半监督支持向量回归.pdfVIP

第27 卷 第2 期 控 制 与 决 策 2012 年 2 月 Vol. 27 No. 2 Control and Decision Feb. 2012 文章编号: 1001-0920 (2012) 02-0205-06 基于Help-Training 的半监督支持向量回归 程玉虎, 冀 杰, 王雪松 (中国矿业大学信息与电气工程学院，江苏徐州221116) 摘 要: 提出一种基于Help-Training 的半监督支持向量回归算法, 包含最小二乘支持向量回归(LS-SVR) 和 近 邻(NN) 两种类型学习器. 主学习器LS-SVR 通过选择高置信度的未标记样本加以标记, 并将其添加到已标记样本 集, 使训练样本的规模不断扩大, 以提高LS-SVR 的函数逼近性能. 辅学习器NN 用以协助LS-SVR 从训练样本比较 密集的区域选取未标记样本加以置信度评估, 可以减弱噪声对学习效果的负面影响. 实验结果表明所提算法具有良 好的回归估计性能, 学习精度较高. 关键词: 半监督学习；助训练；支持向量回归； 近邻；置信度 中图分类号: TP18 文献标识码: A Semi-supervised support vector regression based on Help-Training CHENG Yu-hu, JI Jie, WANG Xue-song (School of Information and Electrical Engineering，China University of Mining and Technology，Xuzhou 221116 ，China. Correspondent ：CHENG Yu-hu，E-mail ：chengyuhu@163.com) Abstract: A semi-supervised support vector regression based on Help-Training is proposed, which includes two kinds of learners: a least squares support vector regression(LS-SVR) and a -nearest neighbor(NN). As a main learner, the LS- SVR chooses unlabeled samples with the highest confidence to label and adds these samples to the labeled sample set, which is repeated for given iterations to enlarge the scale of the training samples so as to improve the property of function approximation of the LS-SVR. As an auxiliary learner, the NN is used to help the LS-SVR choose unlabeled samples to evaluate confidence from a high-density region of training samples, which can weaken the negative influence of noise on the learning performance of the LS-SVR. Experimental results show that the Help-Training LS-SVR has advantages of good regression performance