Abstract
We examine hidden Markov models in which the probability density function
of every observed variable, given a state of the Markov chain, is gaussian. The aim of
this paper is to show a methodology to obtain the maximum likelihood estimators of
the parameters of this class of models that can be computed also when the sequence
of observations contains unrecorded data (i.e. missing observations). Our methodology,
based on the EM algorithm, will be explained analysing a time series about the duration
of geyser eruptions, which presents missing observations
Year
2002
Category
Refereed journal