Arbitrated Ensemble for Time Series Forecasting

Abstract

This paper proposes an ensemble method for time series forecasting tasks. Combining different forecasting models is a common approach to tackle these problems. State-of-the-art methods track the loss of the available models and adapt their weights accordingly. Metalearning strategies such as stacking are also used in these tasks. We propose a metalearning approach for adaptively combining forecasting models that specializes them across the time series. Our assumption is that different forecasting models have different areas of expertise and a varying relative performance. Moreover, many time series show recurring structures due to factors such as seasonality. Therefore, the ability of a method to deal with changes in relative performance of models as well as recurrent changes in the data distribution can be very useful in dynamic environments. Our approach is based on an ensemble of heterogeneous forecasters, arbitrated by a metalearning model. This strategy is designed to cope with the different dynamics of time series and quickly adapt the ensemble to regime changes. We validate our proposal using time series from several real world domains. Empirical results show the competitiveness of the method in comparison to state-of-the-art approaches for combining forecasters.

Big data solar power forecasting based on deep learning and multiple data sources

“…Other avenues for future work include dynamic selection of the best prediction model for the next day or studying seasonal differences (Koprinska, Rana, & Agelidis, ) and building prediction models that are better tuned to the seasonal variations. We also plan to develop dynamic ensembles for big data, motivated by Cerqueira et al ().…”

“…In the last few years, several studies in time series forecasting have focused on creating ensembles of prediction models. Ensembles combine the predictions of several forecasting models and have been shown to be very competitive, and more accurate than single forecasting models in Cerqueira, Torgo, Pinto, and Soares (), Koprinska, Rana, Troncoso, and Martínez‐Álvarez (), and Oliveira and Torgo (), including for PV power forecasting (Z. Wang et al ()). Another ensemble method was proposed by Thorey, Chaussin, and Mallet ()—an online learning method that generates a weighted combination of PV power forecasts for PV plants located in France; this technique was used to predict solar energy up to 6 days in advance.…”

Toward a Framework for Seasonal Time Series Forecasting Using Clustering

“…The order of the regressive process is set to the length of the seasonality of the time series. Finally, the AETSF method has been proposed in [4]. It is an ensemble method that combines prediction of several algorithms based on meta-learners.…”

“…Ensemble forecasting methods and hybrid models are created from several state of the art, independent models, that are mixed to create more complex chains. This strategy is the one proposed in the Arbitrated Dynamic Ensemble [4]. This metalearning method combines different models, regarding to their specifies against target datasets.…”

“…We assume here that time series are univariate, regularly sampled and that there is no missing data. We also assume that there is no trend component in the time series 4 .…”