引用本文:顾晓清,张聪,倪彤光.基于随机投影的快速凸包分类器[J].控制与决策,2020,35(5):1151-1158
【打印本页】   【HTML】   【下载PDF全文】   查看/发表评论  【EndNote】   【RefMan】   【BibTex】
←前一篇|后一篇→ 过刊浏览    高级检索
本文已被:浏览次   下载 本文二维码信息
码上扫一扫!
分享到: 微信 更多
基于随机投影的快速凸包分类器
顾晓清,张聪,倪彤光
(常州大学信息科学与工程学院,江苏常州213164)
摘要:
传统的基于核函数的分类方法中核矩阵运算复杂度较高,无法满足大规模数据分类的要求.针对这一问题,提出基于随机投影的快速凸包分类器(FCHC-RP).首先,使用随机投影的方法将样本投影到多个二维子空间,并将子空间数据映射到特征空间;其次,根据数据分布的几何特征得到凸包候选集;再次,基于凸包的定义计算出特征空间中的凸包向量;最后,使用与凸包向量对应的原始样本及其权值训练支持向量机.此外,FCHC-RP还适用于不平衡数据的分类问题,根据两类样本的不平衡程度选择不同的参数,可以得到规模相当的两类样本的凸包集,实现训练数据的类别平衡.理论分析和实验结果验证了FCHC-RP在分类性能和训练时间上的优势.
关键词:  大规模数据  凸包  随机投影  核方法  分类  快速
DOI:10.13195/j.kzyjc.2018.1266
分类号:TP273
基金项目:国家自然科学基金项目(61806026,61572085);江苏省自然科学基金项目(BK20160187,BK20180956);江苏省教育科学“十三五”规划2018年度课题项目(B-a/2018/01/41).
Fast convex hull classifier based on random projection
GU Xiao-qing,ZHANG Cong,NI Tong-guang
(School of Information Science and Technology,Changzhou University,Changzhou213164,China)
Abstract:
Due to the high computational complexity of kernel matrixes, traditional kernel-based methods can not satisfy the requirement of large-scale data classification. To solve this problem, a fast convex hull classifier based on random projection(FCHC-RP) is proposed. In the FCHC-RP, the samples in the original space are projected into several two-dimensional subspaces, and then the data in subspaces are mapped into the kernel space. Then, the convex hull candidate set is computed according to the geometric characteristics of data distribution in the kernel space. Based on the definition of the convex hull, the convex hull vectors in the kernel space are computed. Finally, the support vector machine is trained by the convex hull vectors and their weights. In addition, the FCHC-RP is also suitable for imbalanced classification problems. The FCHC-RP adopts classifier parameters according to the degree of class imbalance between two classes, so that the size of convex hull sets belonging to two class samples is similar. Thus, the training data in two classes are comparative. Theoretical analysis and experimental results verify the advantages of the FCHC-RP in classification performance and training time.
Key words:  large-scale data  convex hull  random projection  kernel method  classification  fast

用微信扫一扫

用微信扫一扫