SMC method for online prediction in hidden Markov models
Abstract
Purpose
As the conventional multistep‐ahead prediction may be unsuitable in some cases, the purpose of this paper is to propose a novel method based on joint probability distributions, which provides the most probable estimation for the predicted trajectory.
Design/methodology/approach
Many real‐time series can be modeled in hidden Markov models. In order to predict these time series online, sequential Monte Carlo (SMC) method is applied for joint multistep‐ahead prediction.
Findings
The data of monthly national air passengers in China are analyzed, and the experimental results demonstrate that the method proposed and the corresponding online algorithms are effective.
Research limitations/implications
In this paper, SMC method is applied for joint multistep‐ahead prediction. However, with the increasing of prediction step, the number of particles is increasing exponentially, which means that the prediction steps cannot be too large.
Practical implications
A very useful advice for researchers who study time‐series forecasts.
Originality/value
A novel method of multistep‐ahead prediction based on joint probability distribution is proposed and SMC method is applied to prediction time series online. This paper is aimed at those researchers who focus on time‐series forecasts.
Keywords
Citation
Zhang, D., Ning, X. and Liu, X. (2009), "SMC method for online prediction in hidden Markov models", Kybernetes, Vol. 38 No. 10, pp. 1819-1827. https://doi.org/10.1108/03684920910994349
Publisher
:Emerald Group Publishing Limited
Copyright © 2009, Emerald Group Publishing Limited