摘要
为了处理模式识别问题中具有加权信息的样本集,提出一种加权支持向量机(weightedsupportvectormachine,WSVM)算法,并对算法进行了理论分析.通过引入样本与超平面加权距离的概念,使得WSVM算法可以对样本的权值信息进行有效处理.针对未明确给出权值分布的样本集,提出一种基于类间中心距离确定权值的经验方法,对加权支持向量机算法采用交叉验证技术在人工及真实数据上进行了仿真,结果表明,加权支持向量机比标准支持向量机具有更小的误识率和更好的稳定性.
In order to deal with the different importances of the samples in a sample set in pattern recognition, a weighted support vector machine method is presented and analyzed in this paper. Samples' weights are properly solved through introducing the concept of weighted distance between weighted sample and hyperplane. Under the circumstances that weight distribution is not presented explicitly, an empirical method based on interclass central distance is presented to estimate the weights of samples set. Cross validation simulation on man-made and real data set shows the weighted support vector machine is a new applicable classification method.
出处
《北京理工大学学报》
EI
CAS
CSCD
北大核心
2005年第3期211-215,共5页
Transactions of Beijing Institute of Technology
基金
云南省省校省院合作项目
关键词
加权支持向量机
模式识别
最大间隔
统计学习
加权距离
weighted support vector machine
pattern recognition
maximal margin
statistical learning
weighted distance
