在研究了支持向量机(SVM)多分类算法的基础之上,针对一对一多分类算法出现不可分区域问题,提出了基于SVM一对一多分类算法的二次细分方法,并将该方法应用于弹簧应力小样本数据的多分类仿真实验.通过与原始方法的仿真结果进行对比,改进方法在多花费了极短时间的前提下,显著提高了分类正确率.针对改进方法可能存在的问题,又通过10次仿真实验验证了该方法的可行性,同时也为SVM在小样本分类领域提供了新的思路.%On the basis of research on multi-classification algorithm of SVM, aiming at problem of inseparable area appeared in one-versus-one multi-classification algorithm, the secondary subdivision method based on the one-versus-one multi-classification algorithm based on SVM is proposed. And this method is applied to classify the small samples data of the spring stress in the simulation experiment. Compared with the simulation results of original method,the new method improves correct rate of classification which it costs some extra time. Aiming at problems the improved method possibly has, the feasibility is verified through ten simulation experiments. At the same time,it provides a new concept in classification field of the small samples.
展开▼
