基于SVC算法的SVM工作集优选法

工作集的规模很大时,支持向量机的学习过程需要占用大量的内存,寻优速度很慢.文中提出一种基于支持向量聚类的工作集优选方法,分别最优化每一类样本集获取支持向量,利用支持向量几何分布特性,筛选后构造工作集.针对样本集不平衡情况,根据每一类支持向量个数对惩罚系数加权的加权优选法解决最优分离超平面偏移问题.该算法所选工作集具有代表性,能大幅度降低学习代价,同时具有较高的分类效率.

Training Set Selection of SVM Based on SVC

When used in large-scale data classification,large memory will be occupied by training support vector machine,and the optimization is very slow.In this paper,a novel training set selection method is presented after describing the principle of SVM.The corresponding weight scheme is also presented,which is to choose the penalty term according to the number of support vector of each class,thus to optimize the separating hyper-plane and to improve the classification accurate ratio.The simulation results show that the training set obtained from this method is very representative with a high classification efficiency and little time consumption.