Skip to content

Threshold-based anomaly alertingΒ #1933

@milhtp

Description

@milhtp

Is your feature request related to a problem? Please describe.
In anomaly detection, it would be helpful to have a configuration available where anomaly alerts are only triggered if the metric value exceeds a specific threshold. For example, I might want to detect anomalies in row counts but only alert if the count is over 20 (to reduce noise from small fluctuations in low-volume tables).

Describe the solution you'd like
Optional config that allows you to set upper / lower thresholds. When set, values will only be flagged as anomalies when they cross those thresholds within the detection period, AND have a significant anomaly_score.

        data_tests:
          - elementary.column_anomalies:
              upper_anomaly_threshold: 20
              lower_anomaly_threshold: 5

Describe alternatives you've considered
where_expression applies to the whole model, not just the detection period

Would you be willing to contribute this feature?
No

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions