自调用支持向量回归优化支持向量机参数  

Parameters Optimization of SVM Based on Self-calling SVR

作　　者：王志明[1,2,3] 谭显胜[1,2] 袁哲明[1,2] 伍朝华[3] 

机构地区：[1]湖南农业大学生物安全科学技术学院,长沙410128 [2]湖南农业大学湖南省作物种质创新与资源利用重点实验室长沙410128 [3]湖南农业大学理学院,长沙410128

出　　处：《系统仿真学报》2010年第2期376-378,共3页Journal of System Simulation

基　　金：国家自然科学基金(30570351);教育部新世纪优秀人才计划(NCET-06-0710);湖南省教育厅科研项目(09C514)

摘　　要：参数选择是支持向量回归分析的关键问题之一,在大训练样本条件下,大范围遍历搜索极为耗时。基于均匀设计和自调用支持向量回归,提出了一种将大样本搜索转化为小样本搜索的策略:在3因素9水平搜索范围,经混合均匀设计产生27个参数组合,每组合对训练集经交叉测试得其均方误差MSE;以MSE为目标函数,对该27个参数组合形成的小样本自调用支持向量回归以留一法进行大范围搜索建模,预测729个完全参数组合;取预测MSE最小的对应参数组合完成大样本的训练、预测。对5个基准数据集的独立预测表明,新方法在保证预测精度的同时,大幅度缩短了训练建模时间,为大样本支持向量机参数选择提供了新的有效解决方案。Parameters selection is a key problem in Support Vector Regression （SVR）. Exhaustive search needs lots of time especially＇ for training the large-scale samples. A new method that the large-scale search was changed into the small-scale search was proposed based on uniform design and self-calling Support Vector Regression. Firstly, the domain with 27 parameter combinations was abstracted from the large-scale samples via a 3 factors and 9 levels mixed uniform design table. Then the 27 MSEs were gotten by training these parameter combinations with SVR. Secondly, a new training dataset including the 27 MSEs and parameter combinations was trained and used to predict the all 729 parameter combinations search domains by SVR using leave-one-out method. The best parameters combinations were found based on the least MSE. Lastly, the large-scale samples was trained and predicted by the best parameter combination. Experiments on 5 benchmark datasets illustrate that the new method not only can assure the prediction precision but also can reduce training time markedly. The new method is an efficient solution to large-scale samples model selection for Support Vector Machine （SVM）.

关 键 词：均匀设计 支持向量回归 大样本 参数选择 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]