Decide what to predict

When defining a model, each attribute in your Tealium AudienceStream CDP profile is reviewed. This process automatically determines the top attributes that have a predictive relationship for the action you want to predict.

Target attributes

A target attribute is an AudienceStream attribute that represents the visitor behavior that you want to predict with any Tealium Predict model. For example, for the boolean visit attribute Has Purchased, a value of true indicates that a purchase event has occurred during a visit while a value of false means a purchase event did not occur during a visit.

The target attribute must be either a boolean or a badge attribute, and be visit or visitor-scoped.

We recommend that you use a visit-scoped boolean attribute as the target. Enrich this attribute to false at the beginning of each visit. Then, enrich it to true when the target event occurs during the visit (for example, when the purchase event occurs).

We also recommend adding a visitor number attribute if you do not already have one. Increment this attribute by one using the same rule that sets the target boolean attribute mentioned in the preceding sentence.

Exclusion attributes

You can exclude attributes that are not relevant for your model.

For your first round of model training, we recommend including all attributes except for output attributes from other Predict models, or attributes based on them.

Note that this model’s output visit attribute is automatically excluded, so you don’t need to manually exclude it again.

By including all other attributes during your initial training, you’ll gain insight into which ones are most relevant. This can also highlight new visit or visitor attributes you might want to add for future training.

After you complete the initial training, you might choose to exclude attributes with values that do not appear outside your training period. After making these changes, you may notice that your F1 score drops a bit in retraining. However, your model’s performance in deployment becomes more consistent and reliable.

Examples of attributes to consider excluding are:

• Attributes based on dates of visit or dates of purchase. These attributes do not repeat their values outside of the training period.
• Attributes based on other information, such as one-off campaign landing parameters, that do not repeat their values outside of the training period.

The target attribute is automatically excluded and does not need to be added manually as an excluded attribute.

Output attributes

The output attribute is created by default when a new model is created. It is a numeric visit-scoped attribute that stores the prediction value generated by a corresponding deployed model.