Comparing Boosting and Deep Learning Methods on Multivariate Time Series for Retail Demand Forecasting - Inria - Institut national de recherche en sciences et technologies du numérique Accéder directement au contenu
Communication Dans Un Congrès Année : 2022

Comparing Boosting and Deep Learning Methods on Multivariate Time Series for Retail Demand Forecasting

Georgios Theodoridis
  • Fonction : Auteur
  • PersonId : 1406763
Athanasios Tsadiras
  • Fonction : Auteur
  • PersonId : 1014215

Résumé

Retail demand forecasting is an inherently complex problem as many different time-related factors as well as the correlation of demands in between each and every retail product have to be taken into account. In technical terms, retail demand forecasting is a multivariate timeseries forecasting problem where every single timeseries has to be not only analyzed but also predicted. Hence, added complexity is introduced necessitating the use of advanced methods with machine/deep learning backgrounds. Boosting models, such as XGBoost and LightGBM, are perfect choices with extensive bibliographic background and have been widely used to tackle multivariate timeseries forecasts. Simultaneously, recent advancements in deep neural networks have introduced new promising architectures that are yet to be applied on many different scenarios. Therefore, within this paper, two of those architectures with different core components are introduced, analyzed and applied. The Temporal Convolutional Network based on Convolution and the Temporal Fusion Transformer based on the Transformer architecture, which uses self-attention, are compared to boosting methods as well as standard statistical approaches, namely Exponential Smoothing and Seasonal ARIMA. The results indicate that the deep learning networks are the better choices contributing to the notion that deep learning has extraordinary capabilities in relation to large scale, complex and noisy data and that the aforementioned newly adopted designs are excellent choices for multivariate timeseries forecasting.
Fichier sous embargo
Fichier sous embargo
0 4 6
Année Mois Jours
Avant la publication
mercredi 1 janvier 2025
Fichier sous embargo
mercredi 1 janvier 2025
Connectez-vous pour demander l'accès au fichier

Dates et versions

hal-04668668 , version 1 (07-08-2024)

Licence

Identifiants

Citer

Georgios Theodoridis, Athanasios Tsadiras. Comparing Boosting and Deep Learning Methods on Multivariate Time Series for Retail Demand Forecasting. 18th IFIP International Conference on Artificial Intelligence Applications and Innovations (AIAI), Jun 2022, Hersonissos, Greece. pp.375-386, ⟨10.1007/978-3-031-08337-2_31⟩. ⟨hal-04668668⟩
0 Consultations
0 Téléchargements

Altmetric

Partager

Gmail Mastodon Facebook X LinkedIn More