Do you know if I can add a threshold information to calculate anomaly score?
I'm analysing a metric with values between 1 to 450. Generaly values are between 435-450. But actually all values higher than 400 must be considered "normal". So I need to add this information to the job .
How can I do this please?
Thanks for your help.
The ML jobs currently don't allow threshold logic on top of the existing anomaly detection. If you need this now you have a few choices:
- Filter the input data by customizing the query that the ML datafeed uses. In other words, exclude data between 400-450.
- Post-process the anomalies (i.e. with Watcher/Alerting) and exclude any anomalies that have an actual value between 400-450.
Or, do you need to use ML at all for this use case? Just use a static threshold for anything below 400?
Thanks for your clear answer.
Maybe 2nd option will be OK for me.
And yes I need ML for this use case, because my question was simple (intentionally) but in practice it is a bit more complicated . I just needed to understand how it works.