Machine Learning versus Statistical Methods in Demand Planning for Energy-Efficient Supply Chains

Abstract

The research project "E2-Design" intends to integrate energy efficiency as a planning parameter in the design of production and logistics networks. In order to achieve this objective, models and methods are developed that enable an appropriate approach in the strategic and tactical planning of supply chains. It was observed that models and methods for designing supply chains are always based on accurate demand forecasting. This is equally true for the design of supply chains with the core objective of energy efficiency. The better and more granular the demand forecast can be performed, the more valid recommendations for action can be provided to improve energy efficiency. To accomplish this, this paper aims at identifying promising models for the selection and implementation of a demand forecasting algorithm. By an initial comparison of statistical methods with machine learning methods, high potentials in the context of machine learning will be identified. Subsequently, several process models for implementing a suitable machine learning algorithm to improve forecasting quality are analyzed and the most suitable procedure will be extracted. In order to validate the individual process phases, the importance and presence of the individual phases will be analyzed with the help of a literature research and the need for further research in the area of the development of a standardized procedure will be formulated.