计算机应用 ›› 2014, Vol. 34 ›› Issue (8): 2273-2278.DOI: 10.11772/j.issn.1001-9081.2014.08.2273

• 人工智能 • 上一篇    下一篇

案例推理分类器属性权重的内省学习调整方法

张春晓,严爱军,王普   

  1. 北京工业大学 电子信息与控制工程学院,北京100124
  • 收稿日期:2014-03-04 修回日期:2014-04-16 出版日期:2014-08-01 发布日期:2014-08-10
  • 通讯作者: 严爱军
  • 作者简介:张春晓(1983-),女,山东日照人,博士研究生,主要研究方向:案例推理;严爱军(1970-),男,湖北当阳人,副教授,博士,主要研究方向:人工智能、过程建模与优化控制;王普(1962-),男,安徽合肥人,教授,博士生导师,主要研究方向:信息处理、智能控制。
  • 基金资助:

    国家自然科学基金资助项目

Introspective learning adjustment approach for attribute weights of case-based reasoning classifier

ZHANG Chunxiao,YAN Aijun,WANG Pu   

  1. College of Electronic Information and Control Engineering, Beijing University of Technology, Beijing 100124, China
  • Received:2014-03-04 Revised:2014-04-16 Online:2014-08-01 Published:2014-08-10
  • Contact: YAN Aijun

摘要:

针对案例推理(CBR)分类器中案例属性权重的分配问题,提出一种基于内省学习的属性权重迭代调整方法。该方法可根据CBR分类器对训练案例分类的结果调整属性的权重。基于成功驱动的权重学习策略,若当前训练案例分类成功,则首先根据权重调整公式增加匹配属性的权重并减少不匹配属性的权重;然后对所有权重进行归一化从而得到当次迭代的新权重。实验结果表明,所提方法的CBR分类器在UCI数据集PD、Heart和WDBC的准确率比传统CBR分类器分别提高1.72%、4.44%和1.05%。故成功驱动的内省学习权重调整方法可以提高权重分配的合理性,进而提高CBR分类器的准确率。

Abstract:

Aiming at the optimal allocation problem of attribute weights in Case-Based Reasoning (CBR) classifier, an introspective learning-based iterative adjustment approach for the attribute weights was proposed. The attribute weights could be adjusted according to the classification result of the training case by CBR classifier. Based on the success-driven weight learning strategy, if the current training case was classified successfully, the weights of matched attributes would be increased and the weights of mismatched attributes would be decreased according to weight adjustment formulas, then all of the weights would be normalized as the new weights of the current iteration. The experimental results show that the accuracy on UCI dataset PD, Heart and WDBC of CBR classifier with the proposed method are respectively 1.72%, 4.44% and 1.05% higher than the traditional CBR classifier. This illustrates that success-driven introspective learning method for the weights adjustment can improve the rationality of weight allocation, and then improve the accuracy of CBR classifier.

中图分类号: