摘要
提出了一种新的多输出支持向量回归算法(M-SVR),给出了定义在超球上的损失函数,并将训练SVM的问题转化为迭代解线性方程组的问题.在求解过程中采用边计算边使矩阵降阶的办法,使得在求解的同时找到了支持向量.实验结果表明:M-SVR算法与SVR算法相比,支持向量明显减少,并且具有更好的整体预测精度和抗噪性能.
A new approach is proposed for SVM multi-output regression in which a hyper-spherical ,insensitive function is defined and iterative procedure is used to obtain the desired solution. During the operation process support vectors can be found directly by lowering the rank of matrix. The results of the experimentations illustrate that M-SVR can get higher whole precision and enhanced capability of anti-noise while the time expense is little.
出处
《华东交通大学学报》
2007年第1期129-132,共4页
Journal of East China Jiaotong University
关键词
支持向量机
多输出
回归
SVR
support vector machine
multi - output
regression
SVR
