组合核函数优化的稀疏最小二乘支持向量机

针对传统最小二乘支持向量机(LSSVM)稀疏性较差的问题，在传统支持向量机的基础上提出了新的LSSVM模型，并对其进行优化。利用选主元Cholesky分解，进行迭代操作，简化求解过程；利用径向基-卡方组合核函数，提高核函数的稀疏性；最后利用遗传算法，对组合核函数与支持向量机的参数寻优，解决了传统LSSVM在大样本情况下稀疏性较差，求解时间过长的问题，提高了LSSVM的泛性与精确度。仿真实验证明了所提出的模型是有效的。

Combination kernel function optimization for sparse Least Squares Support Vector Machine

A new Least Square Support Vector Machine(LSSVM) model is proposed and optimized based on the traditional SVM for the problem that the solution of the traditional LSSVM lacks sparseness. Firstly, the Cholesky factorization is adopted to simplify the solution process. Secondly, the sparseness of the kernel function is improved by using the radial basis-chi-square combination kernel function. Finally, the genetic algorithm is utilized in the parameters optimization of combination kernel function and the SVM. The improved SVM algorithm solves the problem that the traditional LSSVM solution lacks sparseness and the solution time is too long for training large-scale problems. It also improves the generalization and precision of LSSVM. The simulation results verify the effectiveness of the proposed model.