Dealing with missing data through mixture models
Abstract
Many data sets have missing values, however the majority of statistical methods need a complete dataset to work. Thus, practitioners often use imputation or multiple imputations to complete the data as a pre-processing step. In this talk it will be shown how mixture models can be used to naturally deal with missing data in an integrated way depending on the purpose. Especially, it will be shown how they can be used to classify the data or derive estimates for the distances. Results on real data will be shown.
Domains
Statistics [stat]
Fichier principal
vandewalle.pdf (249.41 Ko)
Télécharger le fichier
slides2.pdf (869.96 Ko)
Télécharger le fichier
Origin : Files produced by the author(s)