TY - JOUR
T1 - Predictive Analytics for Demand Forecasting - A Comparison of SARIMA and LSTM in Retail SCM
AU - Falatouri, Taha
AU - Darbanian, Farzaneh
AU - Brandtner, Patrick
AU - Udokwu, Chibuzor
N1 - Publisher Copyright:
© 2022 The Authors. Published by Elsevier B.V.
PY - 2022
Y1 - 2022
N2 - The application of predictive analytics (PA) in Supply Chain Management (SCM) has received growing attention over the last years, especially in demand forecasting. The purpose of this paper is to provide an overview of approaches in retail SCM and compare the quality of two selected methods. The data used comprises more than 37 months of actual retail sales data from an Austrian retailer. Based on this data, SARIMA and LSTM models were trained and evaluated. Both models produced reasonable to good results. In general, LSTM performed better for products with stable demand, while SARIMA showed better results for products with seasonal behavior. In addition, we compared results with SARIMAX by adding the external factor of promotions and found that SARIMAX performed significantly better for products with promotions. To further improve forecasting quality on the store level, we suggest hybrid approaches by training SARIMA(X) and LSTM on similar, pre-clustered store groups.
AB - The application of predictive analytics (PA) in Supply Chain Management (SCM) has received growing attention over the last years, especially in demand forecasting. The purpose of this paper is to provide an overview of approaches in retail SCM and compare the quality of two selected methods. The data used comprises more than 37 months of actual retail sales data from an Austrian retailer. Based on this data, SARIMA and LSTM models were trained and evaluated. Both models produced reasonable to good results. In general, LSTM performed better for products with stable demand, while SARIMA showed better results for products with seasonal behavior. In addition, we compared results with SARIMAX by adding the external factor of promotions and found that SARIMAX performed significantly better for products with promotions. To further improve forecasting quality on the store level, we suggest hybrid approaches by training SARIMA(X) and LSTM on similar, pre-clustered store groups.
KW - Demand Forecasting
KW - Machine Learning
KW - Predictive Analytics
KW - Supply Chain Management
UR - http://www.scopus.com/inward/record.url?scp=85127770491&partnerID=8YFLogxK
U2 - 10.1016/j.procs.2022.01.298
DO - 10.1016/j.procs.2022.01.298
M3 - Conference article
AN - SCOPUS:85127770491
VL - 200
SP - 993
EP - 1003
JO - Procedia Computer Science
JF - Procedia Computer Science
T2 - 3rd International Conference on Industry 4.0 and Smart Manufacturing, ISM 2021
Y2 - 19 November 2021 through 21 November 2021
ER -