一种改进的支持向量机模型优化算法

An Improved Algorithm for Model Optimization of Support Vector Machine

核函数与参数选择即模型优化是影响支持向量机泛化能力的主要因素。为提高支持向量机的泛化能力,文中在最优保存遗传算法的基础上引入学习算子和主成分分析方法,提出一种新的支持向量机模型优化新算法(简称PCA-SLGA)解决支持向量机分类器模型自动优化问题。仿真实验结果表明,与用于支持向量机模型优化的隐马尔可夫、贪心算法、遗传编程等算法相比,PCA-SLGA算法具有快速收敛性和较强的全局搜索能力。实验进一步表明采用混合算法寻找最优核模型是一种可行途径。

M odel optimization,the choice of kernel functions and its parameters,has a profound impact on the generalization ability of the support vector machine. In this paper,to improve the generalization ability for SVM,a newkernel optimization algorithm（ for short PCA- SLGA）,based on the elitist of genetic algorithm,which adopts self- learning operator and principle component analysis method,is put forward in order to solve the problem of automatic optimization of VSM classifier model. Compared with the SVM s optimized by hidden M arkov,greedy algorithm,genetic programming,the experimental results showthat PCA- SLGA converge faster and has stronger global search ability than the algorithms mentioned above. The experiments further indicates that using the hybrid algorithm to optimize the kernel is a promising way to find the optimal kernel model.