基于属性组合的集成学习算法  

Ensemble learning algorithm on attribute combination

作　　者：付忠良[1] 赵向辉[1] 苗青[1] 姚宇[1] 

机构地区：[1]中国科学院成都计算机应用研究所,成都610041

出　　处：《计算机应用》2010年第2期465-468,475,共5页journal of Computer Applications

基　　金：四川省科技支撑计划基金项目(2008SZ0100;2009SZ0214)

摘　　要：针对样本由数字属性构成的分类问题,在AdaBoost算法流程基础上,改传统的基于单属性分类器构造方法为基于组合属性分类器构造方法,提出了一种基于样本属性线性组合的集成学习算法。对属性组合系数的构造,提出了一般性的构造思路,按照该思路,提出了几种具体的组合系数构造方法,并对构造方法的科学合理性进行了分析。利用UC I机器学习数据集中的数据对提出的方法进行了实验与分析,结果表明,基于属性组合的集成学习算法不仅有是有效的,而且比传统AdaBoost算法好。Concerning the classification of samples being composed of digital attributes, an ensemble learning algorithm based on linear combination of samples attributes was proposed. It constructed classifiers based on combined attributes instead of single attribute by traditional AdaBoost algorithm. The general construction idea for attribute combination coefficients was put forward. In accordance with the idea, several concrete construction methods for combination coefficients were given and analyzed to be scientific and reasonable. The experimental results on UCI machine learning dataset illustrate that the ensemble learning algorithm based on attribute combination is effective and outperforms AdaBoost algorithm.

关 键 词：ADABOOST算法 属性组合 集成学习 分类器组合 

分 类 号：TP181[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程]