Feature Groups for Feature-Based Forecasting

This topic provides information about feature groups for feature-based forecasting.

A feature group is a set of attributes or characteristics, such as those that define demographics or seasonality, that are used to analyze effects on product demand.

Feature groups can be based on measures or levels and attributes.

Additionally, you can classify features as being categorical (based on qualitative data such as color, genre, rating, and promotion event type) or quantitative (based on numeric data such as price, cost, and discount percent).

A forecasting profile that uses feature-based machine learning supports up to 370 features of the numeric data type and 25 features of the string data type, and these features can vary by item. An additional 100 features are supported that can vary by item, organization, customer site, demand class, and planning time level.

Note: You must create a feature group that's based on levels and attributes and add the parent level you select for the AccuracyStratificationLevel forecasting parameter as a feature.

Predefined Feature Groups for Measures

This table lists the predefined feature groups for measures:

Feature Group	Measure	Usage
Product Launch Date	Product Launch Date	Before generating the feature-based forecast, you must select the Active check box for this measure. Forecast recommendations are generated for the new product from the product launch date.
Demand Curve Shape for Product Launch	Product Launch Date: Relative Time Period Number	This feature group is used for predicting the shape of the demand curve during the launch of the new product. This measure is based on the Product Launch Date and Product Launch Date: Relative Week Number measures and is meant for use in a demand plan with the week planning time level.
Product Launch Month	Product Launch Date: Month Number	This feature group is used for analyzing the impact of the launch month on the product sales.
Monthly Seasonality	Seasonality	–
Christmas Holiday	Christmas Holiday	–
New Year Holiday	New Year Holiday	–
Easter Holiday	Easter Holiday	–

If you don't want to use any predefined feature group, clear the Active check box for the associated measure.

Create a Feature Group for Feature-Based Forecasting

Follow these steps to create a feature group for a forecasting profile that uses feature-based machine learning:

1. On the Feature Groups tab on the Manage Forecasting Profiles page, click Actions > Create > Create Feature Group from Measures to create a feature group using measures.

   Click Actions > Create > Create Feature Groups from Levels and Attributes to create a feature group using levels and attributes.

2. If you're creating a feature group using measures, follow these steps:

   1. In the Create Feature Group from Measures dialog box, provide a name and description for the feature group.

   2. Under Available Measures, search for the measures that you want to include in the feature group, select them, and move them under Selected Measures.

      Later, you must ensure that these measures are also in the measure catalog of the demand plan that uses the forecasting profile.

   3. Click OK.

      For the added measures, the Active check box is selected, and Quantitative is selected under Type. If required, you can select Categorical under Type.

   If you're creating feature groups using levels and attributes, follow these steps:

   1. In the Create Feature Groups from Levels and Attributes dialog box, on the Levels tab, under Available Levels, search for the levels that you want to include in the feature group, select them, and move them under Selected Levels. If you select and move a hierarchy, all of its levels are moved.

      The levels displayed are for hierarchies of the Product dimension.

      Later, you must ensure that the hierarchies for these levels are also in the dimension catalog for the demand plan that uses the forecasting profile.

   2. On the Attributes tab, under Available Attributes, search for the attributes that you want to include in the feature group, and move them under Selected Attributes.

      The attributes displayed are those that are configured at the Item level for hierarchies of the Product dimension on the Levels and Attributes tab on the Configure Planning Analytics page. Later, you must ensure that the hierarchies for these attributes are also in the dimension catalog for the demand plan that uses the forecasting profile.

   3. Click OK.

      A feature group is added for every level and attribute that you selected. The name and description assigned to the feature group are the same as the name of the level or attribute. For the added levels and attributes, the Active check box is selected, and Categorical is selected under Type. If required, you can select Quantitative under Type.

Configure Predefined Measures for the Day or Month Planning Time Level

If your demand plan is at the day or month planning time level, take these steps:

1. Create a feature group.

2. Add the Product Launch Date: Relative Day Number or Product Launch Date: Relative Month Number measure to the feature group if your demand plan has the day or month planning time level respectively.

3. For the added measure, select the Active check box, and select Categorical under Type.

4. Edit the expression of the Product Launch Date: Relative Time Period Number measure so that it refers to the Product Launch Date: Relative Day Number or Product Launch Date: Relative Month Number measure.

   Note: If you don't want to use the predefined Product Launch Date: Relative Time Period Number measure, deselect the Active check box for the measure. Then, make a copy of the measure, and edit the expression of the copied measure so that it refers to the Product Launch Date: Relative Day Number or Product Launch Date: Relative Month Number measure. Add the copied measure to a feature group, select the Active check box, and select Quantitative under Type.