摘要
针对支持向量回归机(support vector regression,SVR)的参数选择问题,提出了基于遗传算法的SVR参数自动确定方法。分析了SVR各参数对其性能的影响,根据已有的样本集确定遗传算法的搜索区间,然后在该区间内对搜索的参数进行最优选取。为了减少所选参数对训练样本的依赖性,借鉴交叉验证的方法,把训练集分为估计子集,用来选择模型;确认子集选择参数,以推广能力最好的一组参数作为最终参数。将所提出的方法应用于受噪声影响的标准函数,实验结果表明,由该方法所得参数确定的SVR具有较优的预测性能。
An automatic method for selecting the parameters of support vector regression (SVR) based on genetic algorithms is presented. This method defines the search area by analyzing the behavior of SVR with different parameters that also has different influence on the performance of SVR model and then choose the optimal parameters in the given region. In particular, the generalization ability is guaranteed by invoking an internal cross-validation procedure during the optimization process of the SVR model parameters. Experimental results assess the feasibility of the proposed approach for data sets with several noises and demonstrate an improvement of generalization performance.
出处
《系统工程与电子技术》
EI
CSCD
北大核心
2006年第9期1430-1433,共4页
Systems Engineering and Electronics
基金
陕西省自然科学基金资助课题(2004JC12)
关键词
遗传算法
支持向量回归机
参数选择
交叉验证
genetic algorithms; support vector regression (SVR); parameter selection; cross-validation