Kibana result machine learning job contain negative values

[Oskr (Oscar Bohorquez)]

Hi,
I have a question about the results obtained when I run my machine learning job, when I check the anomalies I find that in typical values there are negative values but in my index I don't have negative values.

Why this behavior is occurring?

[walterra (Walter Rafelsberger)]

Hi Oscar,

If there aren't any erroneous negative values in the source data, still, not all distribution models are constrained to have only positive values. If the algorithm selects one of these models as the most appropriate it can report negative values even if the data being fitted are all positive, for example, if the values are very skewed and many are close to zero. The unsupervised model is technically not aware of any domain knowledge of the source data like a range of values all above zero.