摘要
用支持向量机SVM进行分类时,针对在某些机器学习中,存在训练样本获取代价过大,且训练样本中类的数量不对称的问题,提出了基于主动学习策略的加权支持向量机。其在机器学习的进程中,每次从候选样本集中,主动选择最有利于改善分类器性能的n个新样本添加到训练样本中进行学习,引入类权重因子和样本权重因子,将惩罚参数与类权重因子和样本权重因子联系。实验结果表明,该方法能够有效减少训练样本数量,解决类的数量不对称的样本产生的最优分界面偏移的问题,使分类器获得较好的分类性能。
In classification with support vector machine, aiming at problems of the excessive expense caused by obtaining the training samples set and imbalanced samples of the class in some machine learning. A weighted support vector machine based on active learning algorithm is proposed. In machine learning process, n new samples which most conduce to improving the performance of classifier are forwardly chose from the candidate samples set every time and added to the training samples set to study. The class weighting factor and the sample weighting factor are introduced and whose punished parameters are related to these weighting factor. Experiment shows that it can reduce the number of training examples effectively and resolve the bias of the optimal hyperplane of SVMs when samples are unbalanced. The weighted support vector machine based on active learning algorithm can obtain a better classification performance in the lower number of training samples and unbalanced samples.
出处
《计算机工程与设计》
CSCD
北大核心
2009年第4期966-970,共5页
Computer Engineering and Design
关键词
主动学习
支持向量机
分类
样本不平衡
权重因子
分类间隔
active learning
support vector machine
classification
unbalanced samples
weighting factor
classification margin
