针对传统缺失值检测方法缺少对多维特性数据全面立体的分析及难以从众多缺失值填补算法中选择合适方法的问题,通过设计缺失值检测方法,在目前常见的数据点缺失度基础上,首次提出数据总体缺失度和加权数据总体缺失度的概念,实现对数据集缺失程度的全面检测,进而通过实验对比分析不同缺失值填补方法性能.实验结果表明,在不同缺失度的情况下,不同缺失值填补算法的性能不同,所提出的方法可为缺失值填补算法的选择提供有效依据.
Aiming at the problems that traditional missing value detection methods are not comprehensive enough to analyze the multidimensional feature data and it is difficult to select the most appropriate missing value algorithm among numerous methods,this paper first designs a missing value detection method and then proposes three different concepts of missing degree to achieve the comprehensive detection of the data with multidimensional features.On this basis,it compares and analyzes the performance of different missing value imputation methods.The results show that the proposed detection method can evaluate the data with multidimensional features effectively and provide basis for the selection of missing value imputation methods.
乔非;翟晓东;王巧玲
同济大学 电子与信息工程学院,上海 201804
计算机与自动化
数据预处理;缺失值检测;缺失度;缺失值填补方法
data preprocessing;missing value detection;missing degree;missing value imputation methods
《同济大学学报（自然科学版）》 2023 (012)
1972-1982 / 11
科技创新 2030"新一代人工智能"重大项目(2018AAA0101704);国家自然科学基金(62133011,61973237,61873191)
10.11908/j.issn.0253-374x.22166
评论