基于Ranking Loss的多标签分类集成学习算法  

Ensemble learning algorithm of multi-label classification based on Ranking Loss

作　　者：任志博[1,2] 王莉莉[1,2] 付忠良[1,2] 张丹普[1,2] 杨燕霞[1,2] 

机构地区：[1]中国科学院成都计算机应用研究所,成都610041 [2]中国科学院大学,北京100049

出　　处：《计算机应用》2013年第A01期40-42,68,共4页journal of Computer Applications

基　　金：四川省科技支撑计划项目(2011GZ0171)

摘　　要：针对目标可以属于多个类别的多标签分类问题,提出了一种基于Ranking Loss最小化的集成学习方法。算法基于Real AdaBoost算法的核心思想,从Ranking Loss定义出发,以Ranking Loss在样本空间最小化为目标,采取迭代的方法训练多个弱分类器,并将这些弱分类器集成起来构成强分类器,强分类器的Ranking Loss随着弱分类器个数的增加而逐渐减少,并给出了算法流程。通过理论分析和实验数据对比验证了提出的多标签分类算法的有效性和稳定性。An ensemble learning algorithm based on Ranking Loss was proposed to solve the mutil-label classification problem that the instance may belong to several classes at the same time. The algorithm based on the key idea of the real AdaBoost algorithm was proposed from the definition of Ranking Loss. It aimed to minimize the sample space. The algorithm trained weak classifiers by using iterative method, then integrated these weak classifiers to a strong classifier. The Ranking Loss of the algorithm would be gradually reduced as the number of weak classifiers increased. And the algorithm steps were given. Theoretical analysis and experimental results show that this ensemble learning algorithm is effective and stable.

关 键 词：多标签分类 ADABOOST算法 RankingLoss 分类器组合 集成学习 

分 类 号：TP81[自动化与计算机技术—检测技术与自动化装置]