Forecasting Problem:

Forecast a dataset of given time series as accurately as possible, using methods from the computational intelligence area and applying a consistent methodology. This year, all time series in the competition are of a monthly frequency.

Methods:

The prediction competition is open to all methods of computational intelligence, incl. fuzzy method, artificial neural networks, evolutionary algorithms, decision & regression tress, support vector machines, hybrid approaches etc. used in all areas of forecasting, prediction & time series analysis, etc. Ensemble techniques are also allowed, if they employ any CI method. The contestants will use a unique consistent methodology for all time series.

Evaluation:

The only evaluation criterion is the Symmetric Mean Absolute Percentage Error (SMAPE): where F_t is t-th forecasted value, A_t is t-th actual (real) value, and n is the forecast horizon. The results will be uncovered during the IEEE WCCI 2016 conference at the end of the special session IJCNN-13 Advances in Computational Intelligence for Applied Time Series Forecasting (ACIATSF).

Competition Participation:

Organizers as well as their department/institute colleagues are exempted form the competition evaluation but they are welcome to submit their results that will serve as comparison benchmarks. Their paper submission are also welcome.

Conference Participation:

While we do not require this, we strongly encourage the participants to register for IEEE WCCI 2016 and to submit a contribution to the related special session IJCNN-13 Advances in Computational Intelligence for Applied Time Series Forecasting (ACIATSF).



We look forward to your submission and also to meet you in Vancouver on the IEEE WCCI 2016 conference.


Kind regards & good luck!
Martin Štěpnička and Michal Burda