摘要
为了平衡集成学习中差异性和准确性的关系并提高学习系统的泛化性能,提出一种基于AdaBoost和匹配追踪的选择性集成算法.其基本思想是将匹配追踪理论融合于AdaBoost的训练过程中,利用匹配追踪贪婪迭代的思想来最小化目标函数与基分类器线性组合之间的冗余误差,并根据冗余误差更新AdaBoost已训练基分类器的权重,进而根据权重大小选择集成分类器成员.在公共数据集上的实验结果表明,该算法能够获得较高的分类精度.
To balance the diversity and the accuracy in ensemble learning and improve the generalization performance of learning system, a selective ensemble algorithm based on AdaBoost and matching pursuit is proposed. In the algorithm, matching pursuit is fused into the training of AdaBoost, in which the residual between the target function and the linear combination of basis classifiers is minimized with a greedy iterative idea. Then the weight of each basis classifier is updated bythe residual which is generated during the last iteration and then the optimal weights for every classifier are gained, by which the component classifiers are selected. Experimental results on common data sets show that the algorithm can get higher classification accuracy.
出处
《控制与决策》
EI
CSCD
北大核心
2014年第2期208-214,共7页
Control and Decision
基金
国家自然科学基金项目(60975026
61273275)
