Consistent model selection of discrete Bayesian networks from incomplete data

DOI10.1214/13-EJS802zbMATH Open1336.62087arXiv1105.4507MaRDI QIDQ1951147FDOQ1951147

Authors: Nikolay Balov 

Publication date: 29 May 2013

Published in: Electronic Journal of Statistics (Search for Journal in Brave)

Abstract: A maximum likelihood based model selection of discrete Bayesian networks is considered. The model selection is performed through scoring function $S$, which, for a given network $G$ and $n$-sample $D_n$, is defined to be the maximum log-likelihood $l$ minus a penalization term $lambd{a}_{n}h$ proportional to network complexity $h(G)$, ${\displaystyle }$ S(G|D_n) = l(G|D_n) - lambda_n h(G). ${\displaystyle }$ The data is allowed to have missing values at random that has prompted, to improve the efficiency of estimation, a replacement of the standard log-likelihood with the sum of sample average node log-likelihoods. The latter avoids the exclusion of most partially missing data records and allows the comparison of models fitted to different samples. Provided that a discrete Bayesian network is identifiable for a given missing data distribution, we show that if the sequence $lambd{a}_n$ converges to zero at a slower rate than $n^{-1/2}$ then the estimation is consistent. Moreover, we establish that BIC model selection ($lambd{a}_n=0.5log(n)/n$) applied to the node-average log-likelihood is in general not consistent. This is in contrast to the complete data case where BIC is known to be consistent. The conclusions are confirmed by numerical examples.

Full work available at URL: https://arxiv.org/abs/1105.4507

Recommendations

• Learning Bayesian networks from incomplete data with the node-average likelihood
• Bayesian Selection of Decomposable Models With Incomplete Data
• Normative selection of Bayesian networks
• scientific article; zbMATH DE number 1222294
• scientific article; zbMATH DE number 1522721

zbMATH Keywords

model selectionpenalized maximum likelihoodBayesian networksmissing completely at randomcategorical data

Mathematics Subject Classification ID

Asymptotic properties of parametric estimators (62F12) Bayesian inference (62F15) Estimation in multivariate analysis (62H12) Graphical methods in statistics (62A09)

Cites Work

• Estimating the dimension of a model
• Asymptotic Statistics
• A Bayesian method for the induction of probabilistic networks from data
• Bayesian analysis in expert systems. With comments and a rejoinder by the authors
• Title not available (Why is that?)
• On the choice of a model to fit data from an exponential family
• 10.1162/153244303321897717
• Stratified exponential families: Graphical models and model selection
• The EM algorithm for graphical association models with missing data

Cited In (8)

• Greedy structure learning from data that contain systematic missing values
• Bayesian Selection of Decomposable Models With Incomplete Data
• Bayesian network models for incomplete and dynamic data
• Being Bayesian about learning Gaussian Bayesian networks from incomplete data
• Normative selection of Bayesian networks
• Learning Bayesian networks from incomplete data with the node-average likelihood
• Title not available (Why is that?)
• Strong limit theorems for the Bayesian scoring criterion in Bayesian networks

Uses Software

• catnet

This page was built for publication: Consistent model selection of discrete Bayesian networks from incomplete data

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q1951147)