Store-based Demand Forecasting of a Company via Ensemble Learning

Abstract

Demand forecasting is a topic that is frequently used in the literature and is applied in almost every field. For companies, being able to predict future demand provides a strategic advantage. Especially companies with more than one store have to make demand forecasts for their products on a store basis. The reason for this is that the demand for each product can differ based on the store. For this purpose, although there are many examples of traditional approaches in the literature, machine learning methods have been widely used for demand forecasting in recent years. The use of ensemble learning algorithms along with traditional algorithms in machine learning problems has also positively affected demand forecasting success. In this study; Demand forecasting with store-based historical sales data of a company's products was estimated by machine learning method, and the results of ensemble learning algorithms and traditional machine learning algorithms were compared. To improve the results obtained, hyperparameter optimization was applied to the most successful algorithms and increased prediction success.