【摘要】 针对支持向量机的参数选择问题,提出了一种最优化选择方法。通过分析支持向量个数与留一法的关系以及支持向量机参数的几何意义和对推广能力的影响,该算法利用支持向量机比例来衡量参数选择时推广能力的变化,使用不同的规则更新核参数和惩罚因子,简化了参数选择的难度。理论分析证明这种最小最大化参数选择方法可以选择支持向量机参数,仿真试验验证了该方法的有效性。更多还原

【Abstract】 A minmax method for selecting the parameters of SVM is presented. According to the geometric meanings of the parameters and its influences to generalization performance, and the relations between the number of support vectors and leave-one-out method, the method uses the ration of support vectors to training samples is applied to estimate the generalization performance, and updates the kennel parameter (s) and penalty factor respectively. Its feasibility and efficiency are proved theoretically and experimentally.更多还原