Critic特征加权的多核最小二乘孪生支持向量机  

Multi-kernel Least-squares Twin Support Vector Machine Based on Critic Feature Weighting

作　　者：贺智鹏 吕莉[1,2] 陈娟 康平 HE Zhipeng;LYU Li;CHEN Juan;KANG Ping(School of Information Engineering,Nanchang Institute of Technology,Nanchang 330099,China;Nanchang Key Laboratory of IoT Perception and Collaborative Computing for Smart City,Nanchang 330099,China)

机构地区：[1]南昌工程学院信息工程学院,江西南昌330099 [2]南昌市智慧城市物联感知与协同计算重点实验室,江西南昌330099

出　　处：《信息与控制》2025年第1期123-136,共14页Information and Control

基　　金：国家自然科学基金项目(62066030)。

摘　　要：针对最小二乘孪生支持向量机受误差值影响大,对噪声样本敏感及核函数、核参数选择困难等问题,提出一种Critic特征加权的多核最小二乘孪生支持向量机(Multi-Kernel Least-Squares Twin Support Vector Machine based on Critic weighted,CMKLSTSVM)分类方法。首先,CMKLSTSVM使用Critic法赋予特征权重,反映不同特征间重要性差异,降低冗余特征及噪声样本影响。其次,根据混合多核学习策略构造了一种新的多核权重系数确定方法。该方法通过基核与理想核间的混合核对齐值判断核函数相似程度,确定权重系数,可以合理地组合多个核函数,最大程度地发挥不同核函数的映射能力。最后,采用加权求和的方式将特征权重与核权重进行统一并构造多核结构,使数据表达更全面,提高模型灵活性。在UCI数据集上的对比实验表明,CMKLSTSVM的分类准确率优于单核结构的SVM(support vector machine)算法,同时在高光谱图像上的对比实验反映了CMKLSTSVM对于包含噪声的真实分类问题的有效性。Aiming at the problems of least-squares twin support vector machines,which are significantly affected by error values,sensitive to noise samples,and difficult to select kernel functions and kernel parameters,we propose a Critic feature-weighted multi-kernel least-squares twin support vector machine classification method(CMKLSTSVM).Firstly,CMKLSTSVM uses the Critic method to assign feature weights,which reacts to the importance differences between different features and reduces the influence of redundant features and noise samples.Secondly,we construct a new method for determining the weight coefficients of multiple kernels according to the hybrid multi-kernel learning strategy.The method determines the weight coefficients by judging the degree of similarity of the kernel functions through the hybrid kernel alignment values between the base kernel and the center kernel,which allows for the reasonable combination of multiple kernel functions to maximize the mapping ability of different kernel functions.Finally,we use the weighted summation to unify the feature weights with the kernel weights and construct the multi-kernel structure,which makes the data expression more comprehensive and improves the model flexibility.Comparison experiments on UCI dataset show that the classification accuracy of CMKLSTSVM is better than the support vector machine algorithm with single kernel structure,and meanwhile the experiments on hyperspectral images prove the effectiveness of CMKLSTSVM for the real classification problem containing noise.

关 键 词：Critic权值法 混合多核学习方法 加权多核模型 孪生支持向量机 最小二乘损失函数 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]