成对耦合分类器的多球体预处理方法

为了解决基于支持向量机(SVM)的成对耦合(PWC)决策算法的训练实时性问题,提出一种简化最终决策候选集的预处理方法.通过基于多球体(MS)的简略分类器计算样本与类别间的模糊隶属度,挑选部分隶属度较高的类别用于最终PWC决策.在算法实现方面,通过预设条件简化参数需求,设计基于核函数计算次数的评分函数,并借助交叉验证构造最优参数搜索方法.仿真实验表明,预处理方法以极小的分类正确率损失为代价大大缩短了决策时间,使PWC适用于对决策实时性要求较高的复杂问题.

Preprocess method of pairwise coupling based on multi-spheres

A preprocess algorithm for pairwise coupling (PWC) decision was proposed to solve the time-consuming problem of training complex dataset of PWC algorithm based on support vector machine (SVM). Multi-sphere (MS) was used to calculate fuzzy memberships of the classes. A set of classes with larger value of fuzzy memberships were picked out as the candidate set for further PWC. Fewer parameters were required with a set of preconditions. The cardinality of the candidate set was evaluated and the parameter search method was given through cross-validation based on the times of kernel evaluations. Simulation results showed that the total decision complexity decreased a lot with a slight loss of accuracy, thus PWC can be applied to quick decisions on complex systems.