计算机应用 ›› 2010, Vol. 30 ›› Issue (1): 156-158.

• 人工智能 • 上一篇    下一篇

基于相似关系粗糙集模型的数值属性约简算法

吴敏   

  1. 合肥工业大学
  • 收稿日期:2009-07-06 修回日期:2009-08-27 发布日期:2010-01-01 出版日期:2010-01-01
  • 通讯作者: 吴敏
  • 基金资助:
    合肥工业大学科学研究发展基金

Algorithm of numerical attributes reduction based on similarity rough set

Min WU   

  • Received:2009-07-06 Revised:2009-08-27 Online:2010-01-01 Published:2010-01-01
  • Contact: Min WU

摘要: 针对数值属性数据包含大量噪声而经典粗糙集方法易受噪声干扰的问题,提出一种属性度量指标综合衡量属性在样本上的差异性和相似性。以这种属性度量指标为启发式设计了相似关系粗糙集框架下的数值属性约简算法,并推广到经典粗糙集。在车牌字符集和UCI手写体数字字符集上和常用约简算法进行了比较,实验结果显示这种方法产生的约简属性可以导出规则数少并且具有较好分类能力的规则集。

关键词: 字符识别, 粗糙集, 属性约简, 特征选择

Abstract: As to the problem of interferential or noisy data reduction, an attribute significance evaluation principle was proposed based on the difference and similarity of attributes within objects. A numerical attributes reduction algorithm was constructed based on similarity rough set model, and it was extended to canonical rough set too. Experiments were carried out on two data sets, one is of license plate characters and the other is of UCI handwritten number, the experimental results show that the proposed algorithm can generate simpler but more powerful rules set than other reduction algorithms.

Key words: character recognition, rough set, attributes reduction, feature selection