基于微分进化算法的SVM参数选择

支持向量机(support vector machines, SVM)的性能在很大程度上取决于参数的设置, 所以参数选择问题一直是SVM理论和应用研究中的重点问题.SVM的参数选择本质上是一个优化搜索过程, 并且这个优化问题往往是多峰的.微分进化算法(differential evolution, DE)是一种实数编码的基于种群进化的优化算法, 具有强劲的全局搜索能力, 在多峰函数的寻优问题上已表现出优异的性能.为此, 将DE算法用于SVM参数选择, 提出了基于DE算法的SVM参数选择方法(DE-SVM).在标准数据集上的几个仿真实验证明了该方法的有效性.与基于微粒群算法(partical swarm algorithm, PSO)的参数选择方法相比, DE-SVM在复杂问题或多参数的寻优问题上具有更快的寻优速度.

Parameter selection for an SVM based on a differential evolution algorithm

The performances of support vector machines(SVM) usually depend on the parameters,so parameter selection is a very important research problem in SVM theory and application.In nature,parameter selection is a search optimization process;moreover this optimization problem almost always has many peak values.A differential evolution(DE) algorithm is a real-coding optimal algorithm based on swarm evolution.It has powerful global searching ability and shows high performance on optimal problems with multi-peak func...