摘要
可视化技术已经成为大数据分析的重要研究方向。非线性支持向量机(SVM)可视化表达有利于理解分类模型内在分析机制,增强分类可信度,对支持向量机应用推广具有重要意义。将超过两维空间的非线性核SVM模型分为三维特征模型与多维特征模型两类。针对不同模型研究实现了基于移动最小二乘法拟合的三维特征模型超平面可视化策略与基于t-SNE点重构的多维特征模型超平面可视化策略。在UCI公开数据集上验证所提出的策略,实验结果表明,该可视化策略能够剖析SVM模型的分类机制,在一定程度上解决了多维空间非线性核超平面难以刻画的问题。
Visualization technology has become an important research direction of big data analysis.The visualization expression of nonlinear support vector machine(SVM)is helpful to understand the internal analysis mechanism of classification model and can enhance the reliability of classification,which is of great significance to promote the application of SVM.In this paper,the nonlinear kernel SVM model which is more than two dimensional space was divided into three-dimensional feature model and multi-dimensional feature model.We developed the hyperplane visualization strategy for three-dimensional feature space based on moving least square fitting and for the multi-dimensional feature space based on t-SNE point reconstruction.The proposed strategies were validated on UCI open datasets.The experimental results show that the proposed visualization methods can demonstrate the SVM classification mechanism intuitively and depict the nonlinear kernel hyperplane in multi-dimensional space to some extent.
作者
郭明
朱焱
Guo Ming;Zhu Yan(College of Information Science and Technology,Southwest Jiaotong University,Chengdu 611756,Sichuan,China)
出处
《计算机应用与软件》
北大核心
2022年第2期32-37,127,共7页
Computer Applications and Software
基金
四川省科技计划项目(2019YFSY0032)。
