摘要
针对回归问题,通过对bagging集成中的每个个体进行重新排序给出了一种修剪bagging集成的方法.该方法使用回归树作为基学习机,从排序后的回归树中选择一部分预测性能较好的个体构建集成.试验结果表明,基于排序后的大约20%的个体构建的集成除了占用较少的存储空间和具有较快的预测速度外,其预测性能也比基于所有的个体构建的集成好.
This paper presents a novel pruning method based on reordering the regressors generated by bagging, which adopts the regression tree as the base learner and selects a subset of the ordered regressors that have good prediction accuracy to construct the pruned ensemble. The experimental results show that the primed ensemble containing about 20% of the initial pool of regressors, besides being smaller and having faster execution speed, performs better than or as well as the full hagging ensemble in the investigated regression problems.
出处
《系统工程理论与实践》
EI
CSCD
北大核心
2008年第7期105-110,共6页
Systems Engineering-Theory & Practice
基金
河南省软科学研究项目(072400421600)
