摘要
核函数选择是支持向量机研究的热点和难点。目前大多数核函数选择方法主要应用验证方法选择,很少考虑数据的分布特征,没有充分利用隐含在数据中的信息。为此提出了一种应用样本分布特征的核函数选择方法,即先行分析样本分布特征,然后结合核函数蕴含的几何度量选择合适的核函数,使非线性样本映射得到的特征空间线性可分性得到提高,增强可分性和预测能力。仿真结果证明,提出的方法对支持向量机核函数选择能提供有效的指导,且对泛化能力也得到提高,方案具有可行性和有效性。
In Support Vector Machine study, kernel function selection is hot and difficult. Most current methods of selection kernel function mainly use verification choice method, take little account of the characteristic of the sample distribution, and make no use of the implicit in the information in the data. The paper introduced a method of selection kernel function based on the characteristic of the sample distribution. First of all, the paper analyzed the characteristic of the sample distribution base on mathematics description method. Secondly, according to different distribution characteristics, the kernel function was selected. The experiment proves that this method has obviously improved the generalization ability of SVM, and the scheme is practical and feasible.
出处
《计算机仿真》
CSCD
北大核心
2013年第1期323-328,共6页
Computer Simulation
基金
国家自然科学基金资助项目(51164014)
关键词
支持向量机
样本分布特征
核函数
SVM
Characteristic of the sample distribution
Kernel function