Predictive Planning is LEAN!

Our data science teams work hard to improve our Predictive Planning time series forecasting algorithm.  

This week will see the delivery of a major improvement, effective on all SAP Analytics Cloud deployments. The internal code name for this project is LEAN and it has three objectives:  

• Favor the predictive models that are the easiest to interpret 
• Improve the model accuracy 
• Reduce the training time  

Favor the predictive models that are the easiest to interpret 

Before LEAN, it was difficult for end-users to understand some model components (e.g., quadratic polynomial trend).  

The LEAN approach reduces the number of tested model components. Time series forecasting models are more often using linear trends, multi-linear trends, exponential smoothing, and cycles.  

Sticking to the principle of parsimony, this leads to simpler and more interpretable models.  

Improve the model accuracy 

Our internal benchmarks measured an increased accuracy of the models when tested on known data. The use of simpler techniques increases the model’s robustness, with lower processing cost.  

Reduce the training time  

Our internal benchmarks measured that the model training time was on mean reduced by 20% to 30%.  

Kudos to our data science teams for meeting these three objectives to the benefit of our SAP Analytics Cloud customers and partners. We hope you will enjoy these enhancements!