Combining Automatic Prediction Strategies Using Out-of-Sample Evaluations
Sergio David Madrigal Espinoza, Javier Morales-Castillo · Computing in Science & Engineering · 2025
Sometimes, it is necessary to predict hundreds or thousands of time series quickly and efficiently. Currently, there are computational automations (automatic prediction strategies) of some forecasting methodologies that can perform this task relatively easily. However, it is not possible to know in advance which of these automations should be employed, and once one has been chosen, all the series in the set must be predicted using the same forecasting methodology. In this paper, we discuss the aspects that should be considered in order to propose a combination of previous existing automations. Our approach enables users to select a mixture of existing automations automatically, consisting of the best automation for each series according to out-of-sample evaluations, rather than being limited to choosing only one for the entire data set.