Simplify the Use of Hyperparameter Tuning for Improved Forecast Accuracy Through Additional Feedback and Control

Hyperparameter tuning allows you to tune your forecasts and potentially improve forecast accuracy automatically. Oracle Demand Management determines the values for forecasting parameters for your historical demand information at the lowest level of your forecast tree. This update provides more feedback to you and additional control options for the tuning processes that run in the background as part of running the plan.

Some examples of improvements are as follows:

  • Information about the tuning results: The tuned setting measures now provide textual information, such as the tuning results for the selected forecasting node and the level of the forecast tree for the tuned parameter values.
  • Improved control logic: Several improvements have been made in areas such as the handling of forecasting nodes for which the mean absolute percentage error (MAPE) of the forecast is null

This update includes improvements to the feature titled “Use Automated Hyperparameter Tuning to Improve Forecast Accuracy” introduced in Update 24B.

  • Improved user feedback in tuned results for each node: The Tuned Settings measures now contain explanations of the tuning results. This table lists some explanations of the tuning results:

Text in Tuned Settings Measures

 Explanation of Text
Tuned - Improvement found above threshold Hyperparameter tuning identified forecasting parameter changes that resulted in an improvement to the forecast accuracy that meets or exceeds the value in the HypertuneMAPEThreshold forecasting parameter.

Tuned - Base is optimal

Hyperparameter tuning determined that the existing forecasting parameter settings are optimal.

Tuned - Prior is best

Hyperparameter tuning determined that the recommended forecasting parameter settings from the previous tuning run are optimal.

Tuned - No improvement above threshold, Base is best but only by N%

Hyperparameter tuning identified forecasting parameter changes that resulted in an improvement to the forecast accuracy, but the improvement didn’t meet the threshold specified in the HypertuneMAPEThreshold forecasting parameter. N indicates the percentage improvement.

Tuned - No improvement, all params have no impact

Hyperparameter tuning determined that there weren’t any forecasting parameter changes that affected the forecast accuracy.

The Tuned Settings measures now also indicate the level in the forecast tree that the tuning was performed at. This information is shown in the FL=n# format where FL is the initialism for forecast level, and n is the level in the forecast tree.

Example of Information in Shipments Forecast: Tuned Settings Measure

Example of Information in Shipments Forecast: Tuned Settings Measure

  • Expanded tuning: If the base forecast couldn’t be generated with the forecasting parameters in the forecasting profile but a tuned forecast was generated, then the tuned forecast and settings are shown with the text “Tuned - Improvement found, base failed” in the Tuned Settings measures. In this case, the measure for the base MAPE won’t have a value. Previously, tuning wasn’t done if the base forecast couldn’t be generated.
  • Aggregate tuning according to sample percent: Tuning is performed at an aggregate level in the forecast tree only when tuning at the forecast level node wasn’t successful, and the forecast level node is within the current sample as determined by the value in the HypertuneSamplePercent forecasting parameter. Previously, tuning at the aggregate level didn’t consider whether the forecast level node was within the current sample.
  • More intuitive use of full run: When the value in the HypertuneSamplePercent forecasting parameter is 100, and nodal tuning isn’t enabled, then all the hyperparameter tuning results from prior tuning runs will be cleared after the next full run.
  • Advanced Tuning Mode: You now have better handling of multiphase tuning runs when nodal tuning is enabled, and parameter sets are changed between runs. When nodal tuning is enabled, and the set of forecasting parameters is changed from the prior run, then tuning results from prior runs will be retained in the Tuned Settings measures for those forecasting parameters not in the current set of forecasting parameters.

Steps to Enable

You don't need to do anything to enable this feature.

Key Resources

  • Refer to the Cloud Applications Readiness content for the following features for Oracle Fusion Cloud Supply Chain Planning:
    • Use Automated Hyperparameter Tuning to Improve Forecast Accuracy (Update 24B)
    • Control Automated Hyperparameter Tuning at the Plan Level (Update 24C)
    • Receive Planning Advisor Notifications for Automated Hyperparameter Tuning (Update 24C)

Access Requirements

Users who are assigned a configured job role that contains these privileges can access this feature:

  • Edit Forecasting Profiles (MSC_EDIT_FORECASTING_PROFILES_PRIV)
  • Edit Plan Options (MSC_EDIT_PLAN_OPTIONS_PRIV)
  • Edit Plans (MSC_EDIT_PLANS_PRIV)

These privileges were available prior to this update.