摘要
分析了支持向量机理论中支持向量的特性,利用支持向量与样本空间划分的等价性,提出一种新的基于支持向量机的增量学习算法.该算法考虑新增样本集的分布可能改变对已有样本的分类结果,利用支持向量的分布特性,用对样本的划分差集构造新的支持向量集和分类平面,使差集中的样本点对分类贡献尽可能最大,有效提高了分类精度.同时差集操作简单易行,有效降低了问题的计算复杂度.实验结果表明,与常规增量算法相比,该算法在不改变时间复杂度量级的前提下对分类精度有显著提高.
By analyzing the support vectors＇ properties, an algorithm on incremental learning with a support vector machine is proposed that is based on the classification equivalence between the support vectors set and the training data set. Consideration of the possible impact of new samples to history data and the support vectors＇ location properties, the new support vectors set and the hyperplane were constructed using the partition difference set of training data. The classification precision was improved because the difference set includes the most contributing data. The operations of difference set were easy to perform, so the computation complexity of the algorithm was decreased greatly. Experiment results show that the algorithm improves the classification precision without more computation time compared with the regular incremental support vector machine algorithm.
出处
《哈尔滨工程大学学报》
EI
CAS
CSCD
北大核心
2005年第5期643-646,共4页
Journal of Harbin Engineering University
基金
黑龙江省自然科学基金资助项目(F0304)
关键词
支持向量机
支持向量
增量学习
分类
support vector machine
incremental learning
classification
training
