摘要
支持向量机在解决小样本、非线性及高维模式识别问题中具有许多特有的优势,但支持向量的选择过程复杂。该文利用聚类技术的特殊性能,提出基于搜索机制的密度聚类算法,该算法通过一种简单的搜索策略可将密度高于一定限度的对象聚为一类。将该算法用于支持向量的预选取,可减少训练样本数目,提高支持向量机的训练速度。从仿真实验可以看出,通过基于搜索机制密度聚类的支持向量预选取,训练样本数目可减少2/3以上,线性可分的数据训练速度可加快12倍左右,非线性可分的数据训练速度可加快5倍左右。
Support Vector Machine（SVM） presents excellent performance to solve the problems with small sample, nonlinear and the problems of high-dimension pattern recognition, but the process of selecting support vector is quite complicated. Therefore a density clustering algorithm based on search is put forward. Through a sample search strategy the algorithm can cluster the object that its density is over certain threshold to one class, and the application of it to pre-extracting support vector can reduce the number of training samples and improve the training speed of SVM. From the simulation experiments, it can be found that through pre-extracting support vector based on search density clustering algorithm, the number of training sample can reduce 2/3, and the training speed can quicken 12 times for linear separable data and quicken 5 times for nonlinear separable data.
出处
《计算机工程》
CAS
CSCD
北大核心
2008年第19期206-208,共3页
Computer Engineering
关键词
搜索机制
支持向量机
预选取
search mechanism
Support Vector Machine（SVM）
pre-extracting
