计算机工程

• 人工智能及识别技术 • 上一篇    下一篇

一种改进的带参数AdaBoost算法

邱仁博,娄震   

  1. (南京理工大学 计算机科学与工程学院,南京 210094)
  • 收稿日期:2015-04-10 出版日期:2016-07-15 发布日期:2016-07-15
  • 作者简介:邱仁博(1990-),男,硕士,主研方向为模式识别;娄震,副教授、博士。

An Improved Parameterized AdaBoost Algorithm

QIU Renbo,LOU Zhen   

  1. (School of Computer Science and Engineering,Nanjing University of Science and Technology,Nanjing 210094,China)
  • Received:2015-04-10 Online:2016-07-15 Published:2016-07-15

摘要: 基于间隔理论的AdaBoost改进算法大多通过直接优化基于间隔的损失函数,以提高算法的泛化能力。通过改进AdaBoost算法的权值调整策略,增加间隔增量从正到负变化的样本权重,进而抑制训练样本间隔的负向移动,优化损失函数。通过100轮5倍交叉验证结果表明,与PAB,IPAB等算法相比,该算法的分类准确性和稳定性都有一定的提高。

关键词: 间隔理论, AdaBoost算法, 权重调整策略, 泛化性能, 交叉验证

Abstract: Most of the improved AdaBoost algorithms based on margin theory diverty optimize the loss function to improve the gereralization ability of the algorithm.This paper proposes an improved parameterized AdaBoost algorithm.It improves the weight adjusting strategy of AdaBoost algorithm and increases the weights of the training samples of which the margin increment moves from positive to negative,so as to resist the negative movemen of the margin in crement and better optimize the loss function.Through 100 times of 5-fold cross-validation,results show that,compared with other algorithms,such as PAB,IPAB,etc.,the proposed algorithm has better classification accuracy and stability.

Key words: margin theory, AdaBoost algorithm, weights adajusting strategy, gernalization capability, cross-validation

中图分类号: