Bayesian Networks for Imputation

Abstract

Bayesian networks are particularly useful for dealing with high dimensional statistical problems. They allow a reduction in the complexity of the phenomenon under study by representing joint relationships between a set of variables through conditional relationships between subsets of these variables. Following Thibaudeau and Winkler we use Bayesian networks for imputing missing values. This method is introduced to deal with the problem of the consistency of imputed values: preservation of statistical relationships between variables ("statistical consistency") and preservation of logical constraints in data ("logical consistency"). We perform some experiments on a subset of anonymous individual records from the 1991 UK population census. Copyright 2004 Royal Statistical Society.

Bayesian network data imputation with application to survival tree analysis

“…We end this section by discussing on other methods to perform data imputation. Di Zio et al (2004) also use Bayesian networks for data imputation, but they force the network (and the imputation procedure) to follow a pre-defined order among the variables in the domain. This constraint is known to restrict the learning to sub-optimal Bayesian networks.…”

“…These ideas seem well suitable when the number of samples are over a thousand (Riggelsen, 2006), which is rarely the case of most medical datasets. Finally, to the best of our knowledge, all previous attempts have used local search methods to deal with the structure learning problem (Di Zio et al, 2004;Romero and Salmerón, 2004;Ramoni and Sebastiani, 1997;Riggelsen and Feelders, 2005;Riggelsen, 2006), while we use a globally optimal procedure.…”

“…Standard belief updating (and/or maximum probable explanation) methods are used (Koller and Friedman, 2009). We use Bayesian networks in a similar way to previous studies (Romero and Salmerón, 2004;Di Zio et al, 2004), but differently from them, we implement a global structure learning method that directly works with the expected sufficient statistics instead of completing the data at each step of the structure learning process, thus achieving more accurate models. Our method for data imputation is freely available online at http://code.google.com/p/csda-dataimputation/.…”

Calibrated Hot-Deck Donor Imputation Subject to Edit Restrictions

“…That is, we impute the variable(s) with the least number of missing values first, and end with the variable(s) with the most missing values. Possibly better orders for the variables to be imputed can be developed (see, e.g., Di Zio et al 2004).…”

An Extended Version of the CPT-based Estimation for Missing Values in Nominal Attributes

“…Marco et al used the dependency relationship between attributes to improve estimation performance, when a causal network is available for the attributes [6]. In the causal network of Fig.…”