基于粗糙集的决策树集成学习算法  

Decision Tree Ensemble Learning Algorithm Based on Rough Set

作　　者：时雷[1] 段其国 张娟娟[1] 熊明阳 席磊[1] 马新明[1] SHI Lei;DUAN Qiguo;ZHANG J uanjuan;XIONG Mingyang;XI Lei;MA Xinming(College of Information and Management Science,Henan Agricultural University/Collabora tive Innovation Center of Henan Grain Crops,Zhengzhou,Henan,450002,China;Zhengzhou Commodity Exchange,Zhengzhou,Henan,450008,China)

机构地区：[1]河南农业大学信息与管理科学学院,河南粮食作物协同创新中心,河南郑州450002 [2]郑州商品交易所,河南郑州450008

出　　处：《广西科学》2018年第4期423-427,共5页Guangxi Sciences

基　　金：国家自然科学基金(31501225);河南省高等学校重点科研项目(16A520055);河南省现代农业产业技术体系(S2010-01-G04);国家重点研发计划(2016YFD0300609);粮食丰产增效科技创新专项(SQ2017YFNC050081);国家留学基金资助(201709160005);河南省科技攻关项目(162102110120)资助

摘　　要：【目的】为提高决策树集成的泛化能力和效率,解决集成全部决策树的情况下有时并不显著提高精度、反而导致额外存储和计算开销的问题,提出一种基于粗糙集的决策树集成学习算法。【方法】该算法基于粗糙集理论,从训练的全部决策树中选择一部分进行集成。【结果】与目前流行的集成学习算法Bagging和Boosting相比,本文提出的算法有效地减小了集成规模,并获得更好的泛化能力。【结论】该算法提高了决策树集成的泛化能力和效率。[Objective]The research of the paper focuses on the improvement of the generalization ability and efficiency of ensemble, and resolves the problems that aggregating all decision trees in ensemble usually improves the accuracy of classification slightly, but leads to extra memory costs and computational times. A decision tree ensemble learning algorithm based on rough set is proposed in this paper. [Methods]The algorithm is based on the rough set theory and selects a part from all the decision trees of the training for integration. [Results]The experiment re suits show that compared with the currentpopular ensemble learning algorithm IBaggmg and Boosting, the proposed algorithm not only effectively reduces the scale of ensemble but also obtains stronger generalization ability. [Conclu- sion]The algorithm improves the generalization ability and efficiency of decision tree integration.

关 键 词：集成学习 粗糙集 决策树 BAGGING BOOSTING 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]