基于序贯均匀设计的支持向量机超参数的优化方法

Optimization of Hyper-parameters in Support Vector Machine by SNTO Algorithm

求解支持向量机超参数的优化问题可通过多因素多水平的试验设计来实现。运用序贯均匀设计的方法对高斯径向基核函数参数σ和惩罚因子C进行优化,可以快速且有效地找到最优参数。仿真结果表明,运用序贯均匀设计比直接运用均匀设计效果更佳,且比网格搜索法更具稳健性。

The optimization of hyper-parameters can be solved by a multi-factor and multi-level experimental design.This paper uses the sequential number theoretical optimization(SNTO)algorithm to optimize the Gaussian radial basis kernel function parameterσand the penalty factor C.The SNTO algorithm can find the optimal parameters quickly and efficiently.The simulation results show that the SNTO algorithm has a better performance than the direct uniform design and the grid searching method.