A Reuse-based Environment to Build Ensembles for Time Series Forecasting

摘要

Several works show that ensembles improve the performance of time series forecasting solutions. However, developing an ensemble is not an easy task. Usually, the analyst has to develop each ensemble as a separate project, designing, implementing and configuring the individual and the ensemble methods for each experiment. This paper proposes a change to this common view. It argues that it is possible and necessary to also look from a reuse perspective. Combining ideas from reuse and time series forecasting requirements, this paper proposes an environment to enable reusability for ensemble development. The environment intends to provide a flexible tool for the analyst to include, configure and execute individual methods and to build and execute ensemble experiments.