Improving Time Series Forecasts

[former_member613955]

As stated in the book “SAP Predictive Analytics, The Comprehensive Guide”, Time series models can be improved by adding additional explanatory variable. In my case I have a forecast with a MAPE of 0, 435. If I include only one extra predictive variable the model improves, but by adding multiple variables (they all individually improved the model) I receive a worse or equal result.

Shouldn’t SAP PA automatically remove those variables, that reduce the quality? What is really the best approach to improve a model?

Best regards

Rene

[marc_daniau]

I confirm that the approach is to provide extra variables to reduce the MAPE.Could you share the csv file (if the data is not confidential) and describe briefly the workflow to reproduce your case ?

Thanks

[achab]

Hello Rene, first of all thanks so much for buying the book “SAP Predictive Analytics, The Comprehensive Guide” as one of the 4 co-authors. On your question, please get in touch directly with Marc on the same, Marc is very close to our internal data science teams and we would like to understand more on your specific case. Usually adding extra variables indeed improve the MAPE.

Thanks & kind regards,

Antoine Chabert, Augmented Analytics Product Manager