Skip to Content

Test phase in statistical forecasting for IBP 1802.


With IBP version 1802, SAP announced splitting of the historical horizon into a "training phase and a test phase".

The documentation says:

"The training phase is used by the system to train the various algorithms in your forecast model based on the ex-post forecast, while the test phase is reserved to assess the prediction accuracy of each algorithm (that is, their forecast accuracy in the historical horizon).

As a result of this separation, real predictions (and not ex-post forecasts) are compared with the historical sales data. Therefore, the system can more reliably choose the algorithm that calculates the best forecast."

The functionality and language here seems to be out of a Machine Learning text. I wonder what is the system doing internally different than before. What are the "Real Predictions" that is compares with historical sales data? How are the values different than the ex-post forecast?

Any ideas anyone? Any pointers to documentation on the math involved?


Shivaditya Singh


Add comment
10|10000 characters needed characters exceeded

  • Get RSS Feed

1 Answer

  • Best Answer
    Oct 23, 2018 at 02:03 PM


    The overall functionality helps to segregate the historical data into phases (training data and test data) and process separately with multiple iterations.

    Real predictions mean comparing historical data multiple iterations with each forecast model and to choose the algorithm that calculates the best forecast.

    If you would like to understand more details technically please raise SAP support incident.

    Best Regards,


    Add comment
    10|10000 characters needed characters exceeded