Machine LearningMethods for Forecasting Product Demand: A case study with telecommunications software
Abstract: There is a lack of evidence pointing to an optimal method for demand forecasting. This paper joins the collection of studies that forecast demand using a combination of machine learning methods. Demand forecasting literature from economics and supply chain management lead to a selection of machine learning models for this paper: random forest (RF), extreme gradient boosting (XGB), and support vector regression (SVR). These models are developed within a walk-forward validation process, alongside a benchmark seasonal auto-regressive moving average (SARIMA) model, to forecast demand of a telecommunications software product with data from Ericsson AB. A stacked ensemble hybrid model is constructed from the forecasts of the SARIMA and RF models. The SARIMA model, with a root mean square errors (RMSE) of 1.38, was outperformed by the three single machine learning models, which had RMSE between 0.98 and 1.05. The stacked ensemble hybrid model, SARIMA-RF, showed high predictive capabilities, with a RMSE of 0.43.
AT THIS PAGE YOU CAN DOWNLOAD THE WHOLE ESSAY. (follow the link to the next page)