Forecast Engine Best Fit: Align Information Criterion calculation

The Information Criterion calculation for the BATS algorithm in the Best Fit forecasting process is aligned with other algorithms in relation to the degrees of freedom.

Degrees of freedom represents the number of data points used in the model fitting process and is used in the Information Criterion calculation as part of the model selection process. All data points are used with BATS, but with Holt-Winters the initialization methods used in Best Fit uses one ‘periodicity’ of less data, as part of model initialization.

We have aligned the calculations for Holt-Winters and BATS with the degrees of freedom starting from the end of the initialization period. Regardless of where the model initialization ends and the modelling starts, the degrees of freedom are now comparable. BATS models are reasonably penalized in the information criterion calculation, and hence the best model selection.

This change to the Information Criterion calculation in relation to BATS models is now applied to the execution of any forecast engine configuration with Default Algorithm = Best, including the default Demand Planning engine (Best ML Combination Months). The general result is anticipated that expect more BATS models to be returned by these forecast engines. Several forecast results are now produced by these forecast engines than in the previous version and this is considered as a disruptive change.

Executing forecast engines where the Default Algorithm = BATS apply the new information criterion calculation but retain other forecast results.

Executing forecast engines where the Default Algorithm is a value other than Best or BATS remain unaffected by this change.

Note: By default, Forecast Engine configuration with Default Algorithm = Best or BATS is automatically applied the updated logic and no additional role or privilege is required to access this feature.