基于约简核矩阵的稀疏最小二乘支持向量机

Sparse Least Squares Support Vector Machine Based on Simplified Kernel Matrix

为了解决最小二乘支持向量机模型稀疏性不足的问题,提出了一种约简核矩阵的LS-SVM稀疏化方法。按照空间两点的欧式距离寻找核矩阵中相近的行(列),并通过特定的规则进行合并,以减小核矩阵的规模,进而求得稀疏LS-SVM模型。以高斯径向基核函数为例,详细阐述了改进方法的实现步骤,并通过仿真表明了采用该方法求得的稀疏LS-SVM模型泛化能力良好。

To solve the problem of sparseness lacking in the Least Squares Support Vector Machine（LS-SVM） model,a sparse LS-SVM method based on simplifying kernel matrix was proposed in this paper.On the basis of the Euclidean distance between two points in space,the similar rows（columns） of kernel matrix can be obtained and merged according to a given rule,so as to reduce the scale of kernel matrix and obtain sparse LS-SVM model.Taking Gaussian radial basis kernel function for example,the implementation steps of this method were elaborated in detail, and the simulation experiment results show that the sparse LS-SVM model obtained by this method has an excellent generalization capacity.