In this article
We will discuss the concept of classification models.
Classification Models allow you to build models using multiple seed cohorts (labels). This is particularly valuable when you want to model out user attributes. Compared to lookalike models, there are a few major differences:
- Mutual Exclusivity: Classification Models assign users to a single cohort (label) at a given time. Classification Models are specifically designed for use cases where you want to enforce mutual exclusivity rules.
- Feature Selection: Classification Models are aware of all possible labels and aim to bucket the rest of your audience into one of these groups. When the model is trained, it picks the most relevant features for this classification exercise.
Population Split: Classification Models split your whole audience into distinct groups. This means the reach of your Classification Model is theoretically your entire audience (all your readers have an age for example). This is in contrast to Lookalike models where the reach is usually constrained by the uniqueness of the seed. However, in practice, the reach of Classification Models is usually less than your entire audience, since you choose a minimum confidence threshold.