机构地区: 北京科技大学计算机与通信工程学院
出 处: 《信息与控制》 2005年第4期403-406,共4页
摘 要: 提出了用核的偏最小二乘进行特征提取.首先把初始输入映射到高维特征空间,然后在高维特征空间中计算得分向量,降低样本的维数,再用最小二乘支持向量机进行回归.通过实验表明,这种方法得到的效果优于没有特征提取的回归.同时与PLS提取特征相比,KPLS分析效果更好. We apply kernel partial least square(KPLS) to least square support vector machines (LSSVM) for feature extraction. The original inputs are firstly mapped into a high dimensional feature space, then score vectors are calculated in high dimensional feature space so that dimensions of the sample are reduced. Experimental results show that LSSVM by feature extraction using KPLS performs much better than that without feature extraction. In comparison with PLS, there is also superior performance in KPLS.
关 键 词: 偏最小二乘 最小二乘支持向量机 核的偏最小二乘 回归
领 域: [自动化与计算机技术] [自动化与计算机技术]