Microsoft 365 compliance center offers two types of classifiers: Trainable classifiers and Exact Data Match (EDM) classifiers.
- Trainable classifiers: These are machine learning models that you can train to recognize specific types of data. You can use the built-in classifiers provided by Microsoft or you can create your own. Once trained, you can use them in policies to automatically classify and label data. To test a trainable classifier, you can use the "Test" feature in the compliance center. You provide some sample content, and the classifier will predict how it would classify that content. Here is the related documentation: https://learn.microsoft.com/en-us/purview/trainable-classifiers-get-started-with
- EDM classifiers: These are used to detect sensitive data that is specific and unique to your organization. They work by creating a sensitive information type entity which is based on a database of sensitive information you provide. To test an EDM classifier, you'll need to create a DLP policy that uses the EDM classifier, then test the policy by creating and uploading files that contain data which should be detected by the classifier. Here is the related documentation: Create, test, and tune an EDM sensitive info type
In terms of usage, the choice between using trainable classifiers or EDM classifiers depends on your specific needs and the nature of the data you're trying to classify and protect.
For example, if you're trying to detect and classify general data types (like harassment or resumes), the trainable classifiers would be a good choice. But if you're trying to detect very specific data that's unique to your organization (like customer IDs or project codes), then the EDM classifier would be a better choice.