一种双线性分段二分网格搜索SVM最优参数方法

A Bilinear Segmented Dichotomy Grid Searching SVM Optimal Parameters Method

提出了一种双线性分段二分网格搜索方法来确定SVM模型的最优参数。首先,在初始搜索范围内,以等间隔固定取样值,对惩罚因子进行采样,分别计算这些取样点的SVM的交叉验证正确率,并寻找出满足交叉验证最高正确率所对应的惩罚因子,确定搜索SVM最佳参数的最佳搜索段;然后,在搜索段间分段采用二分法,迭代求解出每段SVM的最高正确率,已得到所对应的最佳参数;最后,找出所有最佳搜索段的SVM最高正确率的最大值,其对应的最佳参数即为SVM模型参数优化结果。与传统的双线性法、网格搜索法和双线性网格搜索法等方法相比,论文提出的参数优化方法训练量小,计算简单,学习精度高,优选的参数能够使得SVM具有更高的泛化性能。

A bilinear segmented dichotomy grid searching method is proposed to determine optimal parameters of SVM model.Firstly,the sample values are fixed at equal intervals in the initial search range,and the penalty factors are sampled.The cross-validation accuracy of SVM of these sample points is calculated,and the penalty factors that met the highest cross-validation accuracy are found,so as to determine the best search section for the best SVM parameters.Then,the highest accuracy of each segment of SVM is solved by using dichotomy in the intersegment of the search segment,and the optimal parameters are obtained.Finally,the maximum value of the highest accuracy of SVM in all optimal search segments is found.Compared with traditional methods such as bilinear method,grid search method and bilinear grid search method,the proposed method has small training amount,simple calculation,high learning precision,and the optimized parameters can make the SVM model have higher generalization performance.