基于均匀设计的最小二乘支持向量机改进算法

Improved Least Squares Support Vector Machine Algorithm Based on Uniform Design

针对最小二乘支持向量机模型的参数选取耗时长,容易陷入局部最优而导致过拟合的问题,提出了一种基于均匀设计的将大样本搜索转化为小样本搜索技术的参数寻优方法。把支持向量机算法的每一次训练过程作为一个试验考虑,试验影响因子为算法的参数,运用均匀设计的手段进行方案设计,采用统计的方法对结果进行分析和选择。最后把该方法应用于3个大样本数据集建模中的参数优化,仿真结果表明方法大幅度减小了时间复杂度,较好地解决了最小二乘支持向量机的参数优化问题,同时获得精确的建模效果。

The parameters selection in Least Squares Support Vector Machines（LSSVM） takes long time and is easy to fall into local optimum and leads to the ovefitting problem.A new method is proposed based on uniform design,in which the large-scale search is changed into the small-scale search.Each training process the the support vector machine algorithm is considered as a pilot to test the impact factors of the algorithm parameters,then uniform design is used to design parameters,and statistical methods are used to analyze the results and select parameters.Three large data sets are predicted using this method for selecting the parameters of least squares support vector machines.The simulation result shows that it can both solve this problem of least squares support vector machines and reduce training time markedly.