基于错分样本权重约束的AdaBoost算法改进  

Improved AdaBoost Algorithm Based on Weight Restriction of Wrongly Classified Samples

作　　者：严晓明[1] 

机构地区：[1]福建师范大学数学与计算机科学学院,福建福州350108

出　　处：《福建师范大学学报（自然科学版）》2011年第6期23-26,共4页Journal of Fujian Normal University：Natural Science Edition

基　　金：福建省教育厅资助项目(JB11036);福建省自然科学基金资助项目(2009J01273)

摘　　要：提出一种AdaBoost算法中错分样本权重约束的方法,该方法对AdaBoost算法中多次错分样本数据的权重进行约束,当其超过一个阈值时,就限制这些样本权重的增加,使其在下一轮迭代中的权重比例不至于过大,以提高强学习算法的泛化能力及精度.仿真实验表明该方法在分类问题和回归问题上都得到了较好的结果.An improved AdaBoost algorithm which restricts wrongly classified samples weight is proposed.The method restricts increasing weight of samples if their weight are beyond a specific threshold because of more misclassified times than normal.Generalization ability and classification accuracy of strong learning algorithm are enhanced after these restricted weight proportion samples sent to next iterative.Experiment results show that the proposed method has better performance in classification and regression.

关 键 词：ADABOOST 样本权重 约束 泛化能力 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]