Statistical Forecasting Methods and Machine Learning Models in Hierarchical Forecasting for Supply Chain Applications

Abstract. Growing amounts of data result in the ability to generate forecasts that are more accurate and more explainable to decision-makers. In multiple settings, however, statistical forecasting methods are too simple to capture the complexity present in datasets – in these cases, a need for machine learning arises. Additionally, many forecasting initiatives touch hierarchical processes – forecasting has to be performed at multiple levels of hierarchies and then aggregated and corrected to yield accurate results. This paper aims to provide an overview of statistical forecasting ideas, prepare data for machine learning models, and tie everything together to form a hierarchical forecasting solution supplemented by a review of hierarchical forecasting use cases within the supply chain planning area.