Parimputation: From Imputation and Null-Imputation to Partially Imputation

IEEE Computer Society
Publication Type:
Journal Article
The IEEE Intelligent Informatics Bulletin, 2008, 9 (1), pp. 32 - 38
Issue Date:
Filename Description Size
Thumbnail2009005049OK.pdf365.69 kB
Adobe PDF
Full metadata record
Missing data imputation is an important step in the process of machine learning and data mining when certain values are missed. Among extant imputation techniques, kNN imputation algorithm is the best one as it is a model free and efficient compared with other methods. However, the value of k must be chosen properly in using kNN imputation. In particular, when some nearest neighbors are far from a missing data, the kNN imputation algorithms are often of low efficiency. In this paper, a new imputation framework is designed. The imputation uses the left or right nearest neighbor for a missing data in a given dataset. Furthermore, a parimputation (partially imputation) strategy is proposed for dealing with the issue of missing data imputation. Specifically, some missing data are imputed when there are some complete data in a small neighborhood of the missing data and, other missing data without imputation are given up in applications, such as data mining and machine learning.
Please use this identifier to cite or link to this item: