What is it about?
This study proposes the use of a meta-learning approach for time-series forecasting. We use the Gaussian process regression model for multiple-input multiple-output forecasting. By extending conventional local modelling approaches we are able to model the cross-dependencies between a given set of time series, returning a vectorial forecast. The structure of the proposed model, consists of a batch of univariate forecasting modules based on Gaussian regression, followed by a linear regression that takes into account the cross-influences between the different forecast. Making use of the existing correlations in international tourism demand to all seventeen regions of Spain, the performance of the proposed model is assessed in a multiple-step-ahead forecasting comparison.
Featured Image
Photo by Ali Maah on Unsplash
Why is it important?
The results of the experiment in a multivariate setting show that the Gaussian process regression model significantly improves the forecasting accuracy of a multi-layer perceptron neural network used as a benchmark.
Perspectives
This paper presents a meta-learning approach for time-series forecasting. This approach allows to preserve the stochastic properties of the training series in multiple-step ahead prediction, and to model the cross-dependencies between a given set of input variables in order to generate a vectorial prediction.
Oscar Claveria
AQR-IREA, Univeristy of Barcelona
Read the Original
This page is a summary of: Modelling cross-dependencies between Spain’s regional tourism markets with an extension of the Gaussian process regression model, SERIEs, June 2016, Springer Science + Business Media,
DOI: 10.1007/s13209-016-0144-7.
You can read the full text:
Resources
Contributors
The following have contributed to this page
