GAUSSIAN HIDDEN MARKOV MODELS: PARAMETERS ESTIMATION AND APPLICATIONS TO AIR POLLUTION DATA
Roberta Paroli, Luigi Spezia · 1999
Hidden Markov models (HMMs) are sequences of conditionally independent random variables and every observed variable depends only on the contemporary state of an unobserved Markov chain. HMMs in which the probability density function of every observed variable, given a state of the Markov chain, is gaussian are examined here. The aim of this paper is to show how the maximum likelihood estimators of the parameters of these models may be suitably obtained using the EM algorithm. An application of gaussian HMMs to air pollution data will be presented.