Abstract: The problem of reconstructing the feature values in samples of objects given in terms of numerical features is considered. A numerical study of different approaches of solving this problem on one model and three practical problems at different levels of data incompleteness is held. A modification of the model calculation of estimates, not requiring metrics for signs, is suggested. The advantage of a local and recognition approaches over filling gaps with sample averages is shown.

Keywords: data mining, missing data, classification, pattern recognition.

ACM Classification Keywords: I.5 Computing Methodologies – Pattern recognition

Link:

CLASSIFICATION OF INCOMPLETE DATA

Vasily V.Ryazanov, Vladimir V.Ryazanov

http://www.foibg.com/ijima/vol02/ijima02-03-p01.pdf