基于数据分布特性的支持向量机核参数选择方法

SVM parameters selection method based on data distribution

针对支持向量机(SVM)核参数选择困难的问题,通过对SVM分类的原理分析,提出一种新的基于数据分布特性衡量SVM核参数优劣的标准,即使包含正负样本的两个超球体应该尽可能地远且不相交;利用粒子群优化算法(PSO)求出在该衡量标准下的最优核参数。通过对UCI标准数据集的实验,验证了该算法所得到的核参数能在一定程度上提高SVM的泛化能力。

Aiming at the kernel parameters selection problem of Support Vector Machine（ SVM）, a new criterion, to make the two hyperspheres which contain the positive and negative samples as far as possible and not intersecting, for measuring the performance of the SVM kernel parameters was proposed by analyzing the classification principle of SVM. The criterion was optimized by the Particle Swarm Optimation（ PSO） algorithm to get the best parameters. The experiment results on datasets from UCI show that the proposed method is more suitable for parameter selection and can improve the generalization performance of SVM.