Forecasting Parameters for Feature-Based Forecasting

This topic lists the forecasting parameters that are available for feature-based forecasting.

Forecasting parameters provide control over the forecasting process.

Forecasting Parameter	Description
AccuracyStratificationLevel	Specifies a parent level of the Item level in the Product dimension. The parent level is used for out-of-sample calculation of the forecast accuracy by the XG Boost forecasting method. For example, if you select product family as the level, out-of-sample calculation of the forecast accuracy is done on a sample of products selected across multiple product families. Ensure that the hierarchy for the parent level is also in the dimension catalog for the demand plan. Note: You must also add the selected level as a feature in a feature group that's based on levels and attributes.
AccuracyThreshold	Determines the minimum acceptable accuracy for the forecast. If the value is low, the validations performed are less, and the forecast accuracy is low. Tip: Oracle recommends that you set the AccuracyThreshold forecasting parameter to a value between 0.4 (40%) and 0.6 (60%).Initially, specify a low value, and gradually increase it on the basis of the feedback provided by demand planners on the accuracy of the feature-based forecasts.
ExternalStorageServiceName	Stores the name of the storage connection you created on the OCI Object Storage Connection tab on the Configure External Storage page in the Oracle Business Intelligence Cloud Connector. Data from Oracle Demand Management is extracted through the Business Intelligence Cloud Connector to the storage location in the Oracle Cloud Infrastructure Object Storage service.
PlanningAdvisorNotificationLayout	Stores the name of the page layout that's opened when you click a link to the suggested forecast in the Planning Advisor. For the predefined Feature-Based Bookings Forecast for New Products forecasting profile, the default is the predefined Feature Based Bookings Forecast page layout. For the predefined Feature-Based Shipments Forecast for New Products forecasting profile, the default is the predefined Feature Based Shipments Forecast page layout. You can also specify a page layout that you created.
TrainingPercent	Specifies the percentage of combinations that's used for machine learning by the XG Boost forecasting method. The remaining combinations are used by the forecasting method for testing the machine learning results. These combinations are determined by the levels selected for the forecasting table. If the value is low, less data is available for machine learning, and if the value is high, less data is available for testing. Tip: Oracle recommends that you set the TrainingPercent forecasting parameter to a value between 0.75 (75%) and 0.85 (85%).