文章摘要
关李晶,何洁帆,张立勇,闫晓明.基于单输出子网迭代学习的缺失值填补方法[J].,2022,62(4):427-432
基于单输出子网迭代学习的缺失值填补方法
Missing value imputation method based on single output sub-network with iterative learning
  
DOI:10.7511/dllgxb202204012
中文关键词: 不完整数据  缺失值填补  单输出子网  基于模型的填补  迭代学习
英文关键词: incomplete data  missing value imputation  single output sub-network  model-based imputation  iterative learning
基金项目:国家自然科学基金资助项目(62076050);国家重点研发计划资助项目(2018YFB1700205).
作者单位
关李晶,何洁帆,张立勇,闫晓明  
摘要点击次数: 70
全文下载次数: 66
中文摘要:
      现实世界中不完整数据广泛存在,通常会增加数据分析与挖掘的难度,影响分析结果的可靠性.合理填补不完整数据的缺失值已经成为当前数据分析和挖掘中一个非常重要的环节.采用不完整数据属性关联建模的方法填补缺失值,鉴于不完整数据属性关联关系的复杂性,使用具有强大学习能力的单输出子网模型对不完整数据的缺失值进行填补,并针对由于缺失值的存在所导致的模型输入不完整问题,从缺失值的对待与描述切入,提出一种基于单输出子网迭代学习的缺失值填补方法.实验结果表明,通过单输出子网迭代学习能够取得更精确的填补结果,验证了所提方法的有效性.
英文摘要:
      Incomplete data exist widely in the real world, which usually increases the difficulty of data analysis and mining, and affects the reliability of analysis results. Therefore, it has become an important issue in data analysis and mining that missing values could be reasonably imputed. The method of attribute association modeling of incomplete data is used for missing value imputation. In view of the complexity of the relations among attributes, a kind of single output sub-network model with strong learning ability is adopted for missing value imputation. Aiming at the problem of incomplete model input caused by the existence of missing values, focusing on the treatment and description of missing values, a missing value imputation method based on single output sub-network with iterative learning is proposed. The experimental results show that the single output sub-network assisted by iterative learning can achieve more accurate imputation results, which verifies the effectiveness of the proposed method.
查看全文   查看/发表评论  下载PDF阅读器
关闭