Application of Deep Learning in the Supply Chain Management: A comparison of forecasting demand for electrical products using different ANN methods

Abstract

The task of demand forecasting has become more important recently due to the strong economic and health crisis caused by covid 19. This phenomenon has brought uncertainty to the business environment and a great fluctuation on the customers’ demand. Therefore, using a method that can produce accurate forecasts has become paramount for companies, in order to take the necessary precautions to avoid stock-outs and waste of resources and improve competitiveness. Deep learning (DL) is an abstraction technique that has proven to be superior to traditional neural networks (ANN), machine learning (ML) techniques and traditional statistical approaches to time series forecasting because of its ability to model massive data sets and solve high-level problems. However, its application in the manufacturing industry is limited. In this paper, we propose a new deep learning method Gated Recurrent Unit (GRU). We compare the proposed method with simple Recurrent Neural Networks (RNN), Long Short-Term Memory (LSTM) models, using real data from a Moroccan electrical products manufacturing company. We conFigure all the models used in this study with the Gridsearch technique to automatically select the most appropriate hyperparameters for each model, and then evaluate the results with the symmetric mean absolute percentage error (SMAPE) and root mean square error (RMSE) methods. Comparison of the results suggests that the GRU method produces the most accurate forecasts than those obtained by the other LD models.