Selecting dynamic graphical models with hidden variables from data

Abstract

Selecting graphical models for a set of variables from data consists of finding the graphical structure and its associated probability distribution which best fit the data. In this paper we propose a new method for selecting Markovian dynamic graphical models from data and, in particular, we develop a new Bayesian technique for selecting graphical hidden Markov models, depicted by a chain graph, from an incomplete data set where values corresponding to hidden or latent variables are not present in data. The proposed method is illustrated by a case study.