基于支持向量域描述的稀疏Bagging算法  

Sparse Bagging via support vector domain description

作　　者：闫文真 李建武[1] 

机构地区：[1]北京理工大学计算机学院,北京100081

出　　处：《中国科技论文》2016年第20期2363-2367,共5页China Sciencepaper

基　　金：国家自然科学基金资助项目(61271374)

摘　　要：提出了1种基于SVDD(support vector domain description)的集成剪枝算法。首先通过有放回的随机采样训练出若干个学习模型,接着通过支持向量域描述算法寻找1个最小超球面,使其包含不少于一定数量的预测模型;然后得到1个可以确定球心位置的稀疏权重向量;最后选取该向量中非零元素所对应的学习模型解决二分类问题。通过多组实验将基于SVDD的集成剪枝算法与Bagging以及其他集成剪枝算法进行比较,验证了所提出算法的准确性和高效性。An ensemble pruning algorithm based on support vector domain description （SVDD） is proposed. Firstly, a number of predictors are trained using bootstrap sampling, and the SVDD is employed to find a minimum hypersphere enclosing the trained predictors whose number is larger than a fixed threshold. Then, a sparse weight vector which represents the center of hyper-sphere can be achieved. Finally, the models whose corresponding elements in the vector are non-zero are used to solve the binary classification problems. Experimental results demonstrate the efficiency and accuracy of this algorithm, compared to Bagging and others stat^of-the-art methods.

关 键 词：集成学习 支持向量域描述 最小超球面 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程]