采用二重扰动机制的支持向量机的集成训练算法  

Ensemble algorithms for training support vector machine based on the double disturbance mechanism

作　　者：贾华丁[1] 游志胜[1] 王磊[2] 

机构地区：[1]四川大学计算机科学与工程学院,成都610064 [2]西南财经大学经济信息工程学院,成都610074

出　　处：《控制与决策》2008年第7期828-832,共5页Control and Decision

基　　金：国家自然科学基金项目(69732010,60272095)

摘　　要：为了有效提升支持向量机的泛化性能,提出两种集成算法对其进行训练.首先分析了扰动输入特征空间和扰动模型参数两种方式对于增大成员分类器之间差异性的作用;然后提出两种基于二重扰动机制的集成训练算法.其共同特点是,同时扰动输入特征空间和模型参数以产生成员分类器,并利用多数投票法对它们进行组合.实验结果表明,因为同时缩减了误差的偏差部分和方差部分,所以两种算法均能显著提升支持向量机的泛化性能.For improving the generalization performance of support vector machine （SVM） effectively, two ensemble algorithms are proposed to train SVM. Firstly, the effectivity of two different disturbance mechanisms on augmenting the diversities among member classifiers, disturbing feature subspace and disturbing model parameters is analyzed. Then, two ensemble algorithms are proposed based on the double disturbance mechanism. The common character of them is that, member classifier is generated by disturbing feature subspace and model parameters, and the finial decision is made by the majority voting procedure. The experimental results show that both algorithms have the ability of improving the generalization performance of SVM significantly because they reduce the bias part and the variance part of the error simultaneously.

关 键 词：支持向量机 集成算法 二重扰动机制 成员分类器 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]