摘要
为了提高对存在噪声的大规模数据集的分类效果,本文提出了一种边界自适应三角模糊非线性优化支持向量分类器BAT-FNOSVC。该分类器在支持向量分类器SVC的基础上引入边界自适应三角模糊隶属函数以更好地解决噪声带来的干扰问题,同时在模型中构造模糊列核矩阵及稀疏化函数,提高了算法的可解释性。对含噪数据集的实验结果表明,与采用三角形模糊隶属函数的稀疏非线性优化分类器TFNOSVC、SVC、1-范数支持向量分类器L1SVC及最小二乘支持向量分类器LSSVC相比,BAT-FNOSVC的准确率有明显提高,说明BAT-FNOSVC算法对有噪声的数据集具有较好的分类效果。
In order to improve the classification effect of large-scale noisy datasets,a boundary adaptive triangular fuzzy nonlinear optimization support vector classifier BAT-FNOSVC was proposed.Based on the support vector classifier SVC,the boundary adaptive triangular fuzzy membership function was introduced to better solve the interference problem caused by noise,and at the same time a fuzzy column kernel matrix and a sparse function were constructed in the model,which improves the interpretability of the algorithm.The experimental results on noisy data sets show that the accuracy of BAT-FNOSVC is significantly improved compared with triangular fuzzy nonlinear optimization support vector classifier TFNOSVC,SVC,1-norm support vector classifier L1SVC and least squares support vector classifier LSSVC,indicating that the BAT-FNOSVC has the better classification effect on noisy datasets.
作者
王琰
李秀芳
张志旺
周莉
WANG Yan;LI Xiufang;ZHANG Zhiwang;ZHOU Li(School of Information and Electrical Engineering,Ludong University,Yantai 264039,China;Division of Science and Technology,Ludong University,Yantai 264039,China;College Information Engineering,Nanjing University of Finance&Economics,Nanjing 210023,China)
出处
《鲁东大学学报（自然科学版）》
2021年第3期220-227,共8页
Journal of Ludong University:Natural Science Edition
基金
国家自然科学基金重大研究计划项目(91538201)
国家自然科学基金青年科学基金项目(61304052)。
关键词
特征选择
核方法
非线性规划支持向量分类器
fuzzy set
feature selection
kernel method
non-linear programming support vector classifier
作者简介
第一作者:王琰(1996-),女,山东德州人,硕士硕士生,研究方向为知识发现与智能决策支持系统。E-mail:1787274000@qq.com;通信作者:周莉(1966-),女,山东烟台人,教授,硕士研究生导师,博士,研究方向为信息融合。E-mail:zxm2zl@126.com。
