计算机工程 ›› 2012, Vol. 38 ›› Issue (9): 197-198,201.doi: 10.3969/j.issn.1000-3428.2012.09.059

• 人工智能及识别技术 • 上一篇    下一篇

基于图的特征选择算法

潘 锋1,2,王建东1,顾其威2,牛 奔2   

  1. (1. 南京航空航天大学计算机科学与技术学院,南京 210016;2. 深圳大学管理学院,广东 深圳 518060)
  • 收稿日期:2011-09-19 出版日期:2012-05-05 发布日期:2012-05-05
  • 作者简介:潘 锋(1977-),男,讲师、博士研究生,主研方向:数据挖掘,机器学习;王建东,教授、博士生导师;顾其威,教授;牛 奔,副教授、博士
  • 基金项目:
    国家自然科学基金资助项目(71001072);广东省自然科学基金资助项目(9451806001002694)

Feature Selection Algorithm Based on Graph

PAN Feng   1,2, WANG Jian-dong   1, GU Qi-wei   2, NIU Ben   2   

  1. (1. College of Computer Science and Technology, Nanjing University of Aeronautics and Astronautics, Nanjing 210016, China; 2. College of Management, Shenzhen University, Shenzhen 518060, China)
  • Received:2011-09-19 Online:2012-05-05 Published:2012-05-05

摘要: 针对数据挖掘与模式识别领域中的高维数据处理问题,通过分析样本类间距离与类内距离,给出一种基于图理论的特征排序框架。根据该框架,提出使用类内-类间和K近邻相似度定义的2种快速特征选择算法,能避免复杂度较高的广义特征分解过程。实验结果表明,该算法具有较高的分类精度。

关键词: 数据挖掘, 模式识别, 特征选择, 图模型, 特征分解, K近邻

Abstract: The high dimensionality of the data samples often makes the data mining or pattern recognition tasks intractable, through analyzing both the within-class distance and between-class distance, it presents a fast feature ranking framework, from which the computationally expensive feature decomposition is avoided. Two similarity measures of within-class and between-class similarity and K nearest neighbor similarity are employed to derive efficient feature selection algorithms. Experimental results demonstrate that these algorithms have higher classification precision.

Key words: data mining, pattern recognition, feature selection, graph model, feature decomposition, K nearest neighbor

中图分类号: