基于AdaBoost改进的多分类器动态集成算法  

Dynamic integration algorithm of multiple classifiers based on improved AdaBoost

作　　者：姜季春[1] 马丹[1] 

机构地区：[1]贵州大学计算机科学与技术学院,贵州贵阳550025

出　　处：《计算机工程与设计》2015年第11期3000-3004,共5页Computer Engineering and Design

基　　金：贵州大学青年教师科研基金项目(2012017)

摘　　要：为提高分类准确率,研究一种改进的多分类器动态集成算法。调整AdaBoost,使其适用于加权训练集;引入属性相关度来标记待分类样本和训练集决策属性之间的相似程度,实现以动态筛选的方式组合最终的分类模型。该算法避免了在分类模型集成过程中对训练集的重复抽取,弥补了模型中单分类器位置固定不变的不足。实验结果表明,该算法能有效提高分类精度和泛化能力。To improve the accuracy rate of classification,an improved dynamic integration algorithm of multiple classifiers was studied.The AdaBoost algorithm was redefined,so that it was applicable to the weighted training set.The definition of the attribute correlation between the sample to be tested and decision attributes of the training set was introduced,and the final classification model was assembled by means of dynamic selection.The improved algorithm avoids re-sampling of training sets,and resolves the problem that the improved AdaBoost generates the aptotic array of classifiers to all the samples.Experimental results show that the proposed algorithm effectively improves the classification precision,and gets better classification results.

关 键 词：多分类器集成 分类 ADABOOST 动态选择 属性相关度 

分 类 号：TP311.13[自动化与计算机技术—计算机软件与理论]