Understanding Classification Likelihood Results
Where the V1 engine experience used confidence score percentages to calculate accuracy of results, the V2 engine leverages Likelihood buckets to categorize results and indicate probable detection.
For more information on how to manage terms, see Reviewing Recommended Terms.
Likelihood Result Buckets
After a data source is scanned, terms recommended by the classification service will be paired with a likelihood result to indicate the discovered data's probable detection.
|
Likelihood Bucket |
Description of Use |
|---|---|
|
Very Likely |
This result is where users feel confident in the applied label and can apply auto-approvals without manual review of any data sets. The goal for structured and unstructured data is to return 0 false positives. |
|
Likely |
This result could pass without manual review but a quick review for any false positives is always encouraged. Users could feel confident in auto-approvals for a specific data source if it does not contain critical PII. Adding at least one detection rule will raise the likelihood chances closer to the Very Likely result. |
|
Possible |
This result encourages a user to review the samples to validate edge cases where false positives can occur. "Bulk approval options for automating approvals for results with "Possible" likelihood is not recommended. |
|
Unlikely |
This result does require user review for more critical PII even with only a few results in this category. Users could confident to reject these results if there is no context pending review. |
|
Very Unlikely |
This result type is not surfaced unless detection rules match to change the likelihood to Unlikely or above. |