I used an example from the KNIME Example server *1) to set up a workflow that demonstrates how this could be done. I am not yet satisfied with the graphics…
The principle is that you use a time window of historical GDP data (Spain) for a time window of eg 5-10 years and then keep the last 2 (or 1 or 3) years as an unknown (to be predicted).
I am not sure if you could use that to predict to the year 2048 that seems like a very far stretch. The models rely on historical GDP data from the years before.
XGBoost gives the best prediction for 2016 and 2017, but still the results are not very good. Ideas could be to add relative growth from year to year or adapt for inflation since the model will use absolute numbers.
cluster ^= GDP
*1) knime://EXAMPLES/04_Analytics/07_Time_Series/02_Example_for_Predicting_Time_Series
kn_example_gdp_prediction.knar (3.2 MB)