论文部分内容阅读
支持向量机(support vector machine,SVM)分类性能主要受到SVM模型选择(包括核函数的选择和参数的选取)的影响,目前SVM模型参数选择的方法并不能较好地确定模型参数。为此基于Fisher准则提出了SVM参数选择算法。该算法利用样本在特征空间中的类别间的线性可分离性,结合梯度下降算法进行参数寻优,并基于Matlab实现选择算法。实验结果表明参数选择算法既提高了SVM训练性能,又大大减少了训练时间。
The classification performance of support vector machine (SVM) is mainly affected by the choice of SVM model (including the choice of kernel function and the selection of parameters). At present, the method of SVM model parameter selection can not determine the model parameters well. To this end, SVM parameter selection algorithm is proposed based on Fisher criterion. The algorithm exploits the linear separability of the samples in the feature space, combines with the gradient descent algorithm to optimize the parameters, and implements the selection algorithm based on Matlab. The experimental results show that the parameter selection algorithm not only improves the SVM training performance, but also greatly reduces the training time.