基于凸包估计的核参数选择方法

核及相关参数的选择是支撑向量机（support vector machine,SVM）研究中的核心问题之一.基于统计学习理论,提出一种通过角度切割样本集求解训练样本的近似凸包来进行最优核参数选择的方法.该方法可以克服传统的基于求解优化问题的方法所具有的计算复杂度高的缺点,且无论数据是否稠密,分布是否均匀都可适用.数值实验说明了提出的方法可行性与有效性.

Kernel parameter selection method based on estimation of convex

The selection of the kernel and the corresponding parameter is one of the key problems for support vector machine(SVM).On the basis of the statistical learning theory(SLT),a new way to select the optimal kernel parameter by the estimation of approximate convex of the sample set is presented,in which the division of the sample set by the degree is adopted.The presented method overcome some disadvantages such as high computation cost,which exist in the traditional optimization-based methods.Moreover,it is used no matter whether the data set is dense or whether the distribution is uniform.The simulation experiments demonstrate the feasibility and the effectiveness of the presented method.