基于先验知识下支持向量机P-SVM的分类算法

Classifying Algorithm Based on P-SVM

支持向量机在分类算法原理中的顺次最小优化算法SMO一般比传统的块算法和固定工作样本集的算法具有更好的时间和空间复杂性,但是由于在实际应用领域中对样本的需求量很大,使得对样本的标记是应用中耗时耗力的工作.本文提出了基于先验知识下的支持向量机,通过引入先验信息量而减少所需样本的数量,同时给出了相应的P-SMO算法.分类应用背景利用中医证候数据,通过专家知识提供的证候知识规则,对训练样本集进行置信度的计算,然后使用P-SMO算法训练出P-SVM,实验结果表明分类效率有较大的提高.

SMO（Sequence Minimum Optimization） algorithm has less complexity of time and space compared with traditional block algorithm and fixative sample collection algorithm, however, as the sample needed in real application field is very large, marking on samples is a waste work of time and power. This paper propose a method that can reduce the demanding quantity of samples by introducing the information on prior knowledge to the samples. The paper also gives the related algorithm named P- SMO. In the application of heribalist disease data, putting the P-SVM idea and P-SMO algorithm into the experiment, the result show a great improvement in classifying efficiency.