基于改进局部极化准则的多核SVM模型

Multiple kernel SVM model based on improved local polarization

为提升SVM在行业领域内应用的普适性以及预测效果,提出了基于改进局部极化准则的多核SVM模型.构造了广义p-范数柯西核,将不同类型的核函数进行核组合,建立了多核SVM模型.重新定义关联系数,建立非线性规划模型求解最优的核权重与核参数.检验了不同p-范数距离对多核SVM性能的显著性影响.通过在5个真实医学数据集上的实验分析,结果表明与传统的单核SVM相比,本文提出的方法在多数情况下具有更好的分类预测性能.

In order to improve the universality and prediction effect of SVM application,a multiple kernel SVM model based on improved local polarization is proposed.The generalized p-norm Cauchy kernel is constructed.The multiple kernel SVM model is established by combining different types of kernel functions.The affinity coefficient is redefined and the nonlinear programming model is built to solve the optimal kernel weights and parameters.The effect of p-norm distance on the performance of multiple kernel SVM is tested.Experimental analysis on five real medical datasets shows that compared with traditional single kernel SVM,the proposed method has better classification prediction performance in most cases.