摘要
提出了一种基于支持向量回归的增量学习算法,该算法在增量学习中除了考虑原训练集中的支持向量(SVs)外,还考虑了非SVs与ε-带(-iεnsensitive zone)的边界距离较近的样本,并将这些样本与新的训练集一起训练.试验结果表明,与传统的支持向量机增量学习算法相比,此算法提高了训练精度;与经典的SVR相比,此算法大大节约了训练时间,是一种行之有效的增量学习算法.
An incremental leaning algorithm based on support vector regression was proposed. First, we generated support vectors （SVs）and non-SVs for origin training data set,then we kept the data that exist near the boundary of ε-insensitive zone as candidates for support vectors and deleted others among non-SVs . The rest of origin training data set and incremental training data set were trained together again, the testing results demonstrated that the proposed method was feasible and effective, because it improved the accuracy comparing with the traditional support vector machine incremental learning algorithm, and shortened the training time greatly comparing with the classical SVR.
出处
《山东理工大学学报（自然科学版）》
CAS
2010年第3期56-59,63,共5页
Journal of Shandong University of Technology:Natural Science Edition
关键词
支持向量
支持向量回归
增量学习
support vectors
support vector regression
incremental learning
