In this article

We walk you through the steps to create a new classification model in the Permutive Platform.

To create a new Classification Model:

1. Click on the “+Add Model” button:
2. Select “Classification” as the model type.
3. Next, select the type of seed you want to use:
   • You can either leverage your own first-party cohorts (Custom Cohorts) as seed labels
   • You can work with a partner dataset.

Custom Cohort Seeds

You can select any of your custom cohorts as seed labels for your Classification Model. However, please note the below:

• There is a minimum size requirement of 1,000 users per seed cohort. This is a hard requirement, if this is not met the model will not train. The dashboard will show you the live size of audiences as you select them to build the model to ensure requirements are met. 

Note: Once the model is built you can click into the model and select 'Validate seeds' at any point, this will return all of the cohort live audiences sizes even once the model is built. 

• The custom cohorts should not be overlapping, i.e. a single user should not be in more than one of the cohorts. If when creating the model there is an overlap of more than 5% with the selected cohorts a message will appear in the dashboard to alert you of the overlap between cohorts. 
• You have to select a minimum of 2 cohorts and a maximum of 5 cohorts.

Partner Dataset

Permutive allows you to leverage partner data for Classification Models, in case you don’t have suitable first-party data for the model. Before selecting a partner dataset, Permutive needs to enable that partner for your project. Please get in touch with your Permutive contact to understand the availability of data partners for your region and the process of enabling them.

Note: Classification-modelled cohorts automatically generate a name that cannot be edited due to this automation. However, this is dependent on the dataset, confidence or model.