Time series forecasting is an important area of research as there are many prediction problems that involve a time component. However, while the time component adds additional information, it also makes time series problems more difficult to handle compared to many other prediction tasks.
This Special Session aims at evaluating different machine learning methodologies addressing the accuracy of soft computing methods in industrial time series forecasting. Multivariate and univariate machine learning approaches will be considered.
Sub-Session: Copper Price Time Series Forecasting Competition for Soft Computing
The present sub-session includes a competition using a data set that contains the time series of copper price. The competition will focus on forecasting one year of monthly-based copper prices by means of state-of-the-art soft computing methods.
The data set to be used will be the World Bank Commodity Price Data (The Pink Sheet): Copper (LME), grade A, minimum 99.9935% purity, cathodes and wire bar shapes, settlement price: Bloomberg; Engineering and Mining Journal; Platts Metals Week; Thomson Reuters Datastream; World Bank, that can be downloaded from: http://pubdocs.worldbank.org/en/561011486076393416/CMO-Historical-Data-Monthly.xlsx and used under a Creative Commons Attribution 4.0 International License (CC BY 4.0).
Contributions will use as the training dataset monthly copper prices between January 1960 and August 2018, although the selected period can be smaller if adequately justified, in order to forecast monthly prices from September 2018 till August 2019, which are already published in the data set.
As the prices to be forecasted are previously known and in order to avoid trial and error practices, authors will have to clearly and thoroughly justify the methodology used in their soft computing model, e.g. in the case of selecting a smaller training dataset, or in the case of selecting other datasets from the Pink Sheet in order to undergo a multivariate analysis.
The model’s forecasting performance will have to be evaluated by means of the following measures: root mean squared error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE). In case that the three measures cannot be simultaneously rated, the MAPE will be considered the most representative performance measurement..
Session topics include but are not limited to:
- Classical Time Series Forecasting.
- Multivariate Time Series Forecasting.
- Machine Learning strategies for Time Series Forecasting.
- Deep Learning for Time Series Forecasting.
- ETS (Error, Trend, Seasonality) models to make forecasts.
- Forecasting performance errors.
- Copper Price Time Series Forecasting Competition.
- Alicja Krzemień, Central Mining Institute (Poland)
- Fernando Sánchez Lasheras, University of Oviedo (Spain)
- Gregorio Fidalgo Valverde, University of Oviedo (Spain)
- Pedro Riesgo Fernández, University of Oviedo (Spain)
Prof. Pedro Riesgo Fernández
Escuela de Ingeniería de Minas, Energía y Materiales de Oviedo
C/ Independencia, 13. 33004 Oviedo (Spain)