摘要
针对数据分类交叉空间易错分问题,提出一种基于加权特征子空间的支持向量机核函数方法。该方法利用加权特征子空间与稀疏表达等相关理论进行支持向量机核函数优化,首先利特征子空间重叠率与数据的信息熵对数据特征进行加权,再通过对L1范数正则项的加权处理调节异类数据间的稀疏性和同类数据间的稠密性;最后对处理好的数据进行分类测试。仿真实验表明该算法能够在一定程度上提升分类效果,以达到优化核函数的目的。
To avoid misclassification of data classification cross-space,a support vector machine kernel function was proposed in weighted feature subspace,in which the feature-weighting subspace,sparse expression,and kernel function,and other related theories are integrated to support kernel function optimization of support vector machine.Firstly,the feature weighting of subspace overlap rate was calculated for each feature-weighted subspace.Secondly,the weighting of the L 1 norm regular term was used to adjust the sparsity of different-class spaces and increase the density of the same class space.Finally,the processed data were classified by the kernel function.The simulation experiments show that the algorithm could improve the classification effect to some extents and optimize the kernel function.
作者
梁礼明
郭凯
盛校棋
LIANG Li-ming;GUO Kai;SHENG Xiao-qi(Electrical Engineering and Automation College,Jiangxi University of Science and Technology,Ganzhou 341000,China)
出处
《科学技术与工程》
北大核心
2020年第15期6101-6106,共6页
Science Technology and Engineering
基金
国家自然科学基金(51365017,61463018)
江西省自然科学基金面上项目(20192BAB205084)
江西省教育厅科学技术研究重点项目(GJJ170491)。
关键词
支持向量机
L1范数
核函数
加权特征子空间
信息熵
support vector machine
L 1 norm
kernel function
weighted subspace
information entropy