Time Series Forecasting
Specialeforsvar ved Mathias Christian Svendsen
Titel: Time Series Forecasting
Abstract: The purpose of this present thesis is to obtain sensible model fits with the purpose of forecasting future demand on various products sold in differentWalmart stores. This is done in connection with the M5 forecasting accuracy competition, based on data provided by Walmart . In particular, we will engage this task with two different approaches. The first approach uses traditional statistical time series models such as ARIMA and exponential smoothing, whereas the second approach uses more modern machine learning algorithms. The thesis is composed of several sections. Section 2 feature an introduction to the main goals of the project. Section 3 introduces the dataset, as well as the rules of the competition and how each submission is being evaluated. Section 4 sets out the theoretical, statistical foundation for the analysis carried out in section 5-7. In section 8 the analysis is concluded. In conclusion, we found that a machine learning approach with tree-based methods performed better overall.
Vejleder: Anders Tolver
Censor: Lars Nørvang Andersen, Aarhus Universitet