作者投稿和查稿 主编审稿 专家审稿 编委审稿 远程编辑

计算机工程 ›› 2011, Vol. 37 ›› Issue (17): 146-148. doi: 10.3969/j.issn.1000-3428.2011.17.049

• 人工智能及识别技术 • 上一篇    下一篇

基于特征熵相关度差异的KNN算法

周 靖,刘晋胜   

  1. (广东石油化工学院计算机与电子信息学院,广东 茂名 525000)
  • 收稿日期:2011-04-08 出版日期:2011-09-05 发布日期:2011-09-05
  • 作者简介:周 靖(1980-),女,实验师、硕士,主研方向:人工智能,数据挖掘;刘晋胜,实验师、硕士

KNN Algorithm Based on Feature Entropy Correlation Difference

ZHOU Jing, LIU Jin-sheng   

  1. (College of Computer and Electronic Information, Guangdong University of Petrochemical Technology, Maoming 525000, China)
  • Received:2011-04-08 Online:2011-09-05 Published:2011-09-05

摘要: 传统K最近邻(KNN)法在进行样本分类时容易产生无法判断或判断错误的问题。为此,将特征熵与KNN相结合,提出一种新的分类算法(FECD-KNN)。该算法采用熵作为类相关度,以其差异值计算样本距离。用熵理论规约分类相关度,并根据相关度间的差异值衡量特征对分类的影响程度,从而建立距离测度与类别间的内在联系。仿真实验结果表明,与KNN及Entropy-KNN法相比,FECD-KNN在保持效率的情况下,能够提高分类准确性。

关键词: K最近邻算法, 熵, 相关度, 差异

Abstract: The paper ameliorates the method that combined K-Nearest Neighbor(KNN) with entropy, a new improved algorithm that adopting entropy as correlation and taking differences values to calculate distance is proposed, which calls FECD-KNN, based on the research that KNN tested sample in misjudgment and error easily. The impacted algorithm combines information entropy theory used to statute correlation, measures strength of impact on the classification according to difference of correlation, and establishes the intrinsic relation between the distance and class. The contrast simulation experiment shows that, compared with KNN and Entropy-KNN, the impacted algorithm adopting the degree of correlation to optimize distance raised the rate of accuracy enormously in classification, meanwhile it also maintains efficiency of classification.

Key words: K-Nearest Neighbor(KNN) algorithm, entropy, correlation, difference

中图分类号: