Loading...
Publication
Improving forecasting by resampling STL decomposition
| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 423.17 KB | Adobe PDF |
Advisor(s)
Abstract(s)
The development of new forecasting algorithms has shown an increasing interest due to the emerging of new fields of application like machine learning and forecasting competitions. Although initially intended for independent random variables, bootstrap methods can be successfully applied to time series. The Boot.EXPOS procedure, which combines bootstrap and exponential smoothing methods, has shown promising results for forecasting. This work proposes a new approach to forecasting, which is briefly described as follows: using Seasonal-Trend decomposition by Loess (STL), the best STL fit is selected by testing all possible combinations of parameters. The best combination of smoothing parameters is chosen based on an accuracy measure. The time series is then decomposed into components according to the best STL fit. The Boot.EXPOS procedure is employed to forecast the seasonal component and the seasonally adjusted time series. These forecasts are aggregated to obtain a final forecast. The performance of this combined forecast is evaluated using real datasets and compared with other established forecasting methods.
Description
Keywords
Boot.EXPOS Forecast Time series Seasonal-Trend decomposition by Loess
Citation
Cordeiro, C., Ramos, M.R., Neves, M.M. (2023). Improving Forecasting by Resampling STL Decomposition. In: Gervasi, O., et al. Computational Science and Its Applications – ICCSA 2023 Workshops. ICCSA 2023. Lecture Notes in Computer Science, vol 14112. Springer, Cham. https://doi.org/10.1007/978-3-031-37129-5_12
Publisher
Springer