Common Demand Prediction Methods

This chapter covers several common methods for demand prediction. We start by presenting the basic linear regression method applied to one SKU. We then explain how to properly structure the dataset. We next discuss two approaches in terms of data aggregation: the centralized approach (which combines the data across all SKUs and estimate a single model) and the decentralized approach (which estimates a different model for each SKU by solely relying on its data). We then cover the topics of feature selection and regularization. Finally, we present several practical concepts, including log transformations and various configurations with fixed effects. We implement each prediction method using the accompanying dataset and we compute the resulting prediction accuracy. We provide comprehensive notebooks to guide the entire process and allow readers to replicate all the steps in their setting of interest.