摘要
训练样本选择是支持向量机的一个重要研究课题。但是,目前大部分样本选择方法的一个共同的不足就是,其训练样本的候选集是整个样本空间,因此可能会选择一些对分类效果影响不大的内部样本,或者选择一些可能会降低分类效果的"过边界"样本。提出了两种基于"有效"候选集的样本选择方法。该方法首先通过"挖心"和剔除"过边界"样本来确定训练样本的"有效"候选集,然后在此"有效"候选集上进行训练样本的选择。实验结果表明,该方法在保留"有效"候选样本的同时,也提高了支持向量机分类器的正确识别率。
Sample selection is an important issue for Support Vector Machines（SVMs）.But,at present most sample selection methods have a common disadvantage that the candidate set for training sample is the whole sample space,so,it may select the interior samples or ＂outliers＂ that have little or even bad effect on the classifying quality.So,two improved methods based on effective candidate set are proposed in the paper.By using these two methods,the effective candidate set is identified through ＂removing center＂and eliminating the＂outliners＂,and then training samples in this effective candidate set are selected.The experimental results show that the methods reserve effective candidate samples undoubtedly,and also improve the performance of the SVM classifiers.
出处
《计算机工程与应用》
CSCD
北大核心
2008年第23期214-216,共3页
Computer Engineering and Applications
关键词
“有效”候选集
挖心
过边界
支持向量机
effective candidate set
removing center
outliers
Support Vector Machines（SVMs）
