Êtes-vous un étudiant de l'EPFL à la recherche d'un projet de semestre?
Travaillez avec nous sur des projets en science des données et en visualisation, et déployez votre projet sous forme d'application sur Graph Search.
Demand forecasting consists of using data of the past demand to obtain an approximation of the future demand. Mathematical approaches can lead to reliable forecasts in deterministic context, extrapolating regular patterns in series. However unpredictable events that do not appear in the historical data can make the forecasts obsolete. As forecasters have a partial knowledge of the context and probable future events (such as strikes, promotions), this work investigates structuring the implicit as well as the explicit knowledge in order to be easily and fully integrated into final forecasts. This paper presents a judgmental-based approach in forecasting where mathematical forecasts are considered as a basis and the structured knowledge of the experts is provided to adjust the initial forecasts. This is achieved using the identification of four factors characterizing specific events that could not have been considered in the initial forecasts. The validation of this approach has been conducted on 2 industrial case studies, a plastic bag manufacturer and a distributor on the food market. The results show that structuring the expert knowledge could lead not only to high improvements of forecasts accuracy but also to a better initial data cleaning and outlier identifications.
Joseph Chadi Benoit Lemaitre, Pan Xu, Weitong Zhang, Yijin Wang, Wei Cao, Myungjin Kim, Shan Yu, Xinyi Li, Lei Gao, Yuxin Huang
Ekaterina Krymova, Nicola Parolini, Andrea Kraus, David Kraus, Daniel Lopez, Yijin Wang, Markus Scholz, Tao Sun