基于类间距的径向基函数-支持向量机核参数评价方法分析

分析了径向基函数（RBF）核参数γ对空间映射结果的影响,得出3条结论。在此基础上,找到了1种新的核参数评价方法,该方法通过计算特征空间中两类之间的平均距离（ICMD）来评价γ的优劣。文章分别从理论和实验两方面证明了ICMD最大值的存在性。为验证该方法的有效性,文中对7个样本集进行了两组参数选择实验：第一组实验通过ICMD找到最优核参数γ,再由10-折交叉验证得到最优惩罚因子C,称为“两步法”;第二组实验采用基于10-折交叉验证的网格搜索法进行参数选择。结果显示两种方法均选择出了适当的参数,但前者花费的时间比后者大大缩短,验证了ICMD方法的有效性。

Analysis of the Inter-class Distance-based Kernel Parameter Evaluating Method for RBF-SVM

The effect of radial basis function(RBF) kernel parameter γ on the mapped space was analyzed.A novel kernel parameter evaluating method was proposed,which is based on the inter-class mean distance(ICMD).The theoretical and experimental analyses were made for the proposed method.Two sets of parameter selection experiments were made in order to prove the validity of the proposed method.In the first set,the optimal parameter γ was chosen by the ICMD method,and then the optimal cost parameter C was obtained by the 10-folds cross validation,which is called Two-Stage Method;In the second set,the 10-folds cross validation-based grid search method was adopted.The results show that the Two-Stage Method can select the optimal parameters with significantly decreased time cost,which proves the validity of ICMD method.