摘要
径向基核函数在支撑向量机中应用最广,其参数取值直接影响着SVM分类器的性能.目前核函数参数选择广泛采用k-遍交叉验证法,该方法计算精度较高,但计算量很大.本文从径向基核函数构造的Gram矩阵的意义出发,定义了理想核函数所对应的Gram矩阵,并采用基于距离测度的方法确定其参数.通过对标准数据集进行实验表明,该方法计算精度与k-遍交叉验证法相当,而且计算量可以明显减少.为了进一步验证该方法的有效性,本文将其应用于中医舌色、苔色样本进行实验,实验结果也是令人满意的.
Radius basis function(RBF) is a kernel used in SVM widely. Tuning kernel parameters has a significant impact on the performance of SVM generalization, K-fold Cross-Validation is usually adopted in the selection of kernel parameters at present, but it has high demand on computation. Analyzed the meaning of Gram matrix constructed by kernel function, an ideal Gram matrix for demonstrating the similarity between samples is defined;To select the parameter of RBF, a novel method which measures the similarity between the ideal Gram matrix and the Gram matrix obtained through RBF kernel by distance measurement is also presented. Theory analysis and experiment results show that the presented approach not only has a good precision, but simplifies calculation evidently.
出处
《电子学报》
EI
CAS
CSCD
北大核心
2005年第B12期2459-2463,共5页
Acta Electronica Sinica
基金
国家自然科学基金(No.60301003,No.60227101,No.60431020)
北京市教委项目(KM200410005030)
北京市基金(No.3052005)
关键词
支撑向量机
核函数
k-遍交叉验证法
矩阵相似性度量
support vector machines ( SVM )
kernel function
k-fold cross-validation
measurement of matrix similarity