This disclosure discusses a data-driven framework to predict an error of time-series forecasting models. Monitoring and management of time-series forecasting models are crucial tasks in industrial and business contexts, where multiple models are deployed and used over time to guarantee their correct operation. However, the large amount of available data, together with the large availability of forecasting models complicate the performance of these tasks. A novel supervised learning approach is presented to predict the error of time-series forecasting models, exploiting information on the forecasting error on other time-series available at training time. The estimated forecasting error represents a surrogate measure of a model’s future performance and can be used to jointly perform model monitoring and selection over time.
Predicting error metrics of time-series forecasting
MOMI 2019, Phd Seminar sur "Le Monde des Mathématiques Industrielles", 25-26 February 2019, Sophia Antipolis, France
Poster / Demo
© EURECOM. Personal use of this material is permitted. The definitive version of this paper was published in MOMI 2019, Phd Seminar sur "Le Monde des Mathématiques Industrielles", 25-26 February 2019, Sophia Antipolis, France and is available at :
PERMALINK : https://www.eurecom.fr/publication/6185