一种基于迭代Bagging的回归算法  

A Regression Algorithm Based on Iterated Bagging

作　　者：王立[1] 朱学峰[1] 

机构地区：[1]华南理工大学自动化科学与工程学院,广东广州510640

出　　处：《控制工程》2009年第1期59-61,79,共4页Control Engineering of China

基　　金：广东省科技攻关基金资助项目(2005B10201005)

摘　　要：在集成算法中,Bagging算法能够在回归预测中有效地减少方差,但在减少偏置方面却没有明显的效果。针对这一问题,提出一种迭代Bagging回归算法,在每个独立阶段的学习中通过Bagging算法和个体学习机初始权值的随机化设置来减小方差;同时,又通过减小方差后的回归残差进行多个阶段的迭代计算,并将各阶段结果叠加,在减小方差的同时达到偏置的减小,从而使得泛化误差更大程度的减小,得到更精确的预测结果。通过对标准数据集和实际数据的仿真实验证明,此方法可以达到比单纯Bagging算法更好的回归预测效果。Among the ensemble algorithm, Bagging can effectively reduce the variance of regression predictions, while the bias left almost unchanged. An iterated Bagging regression method based on multi-networks is proposed. By setting the single network more randomly and the residual iteration, the variance can be reduced and meanwhile the bias is decreased effectively. The generalization error is decreased and more accurate prediction is obtained. The simulations on the standard and real data set show that the proposed method works.

关 键 词：迭代 偏置 BAGGING 回归 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程]