Advanced settings

The fine-tuning of anomaly detection sits in a collapsed section of the form. All fields are optional: left blank, the default values apply, which vary with the interval and the aggregation function.

Setting	Content	Default
Sensitivity	Number of standard deviations from which a point is considered anomalous. Lower values make detection more sensitive	3
Analysis window (intervals)	Number of historical intervals used to train the model	depends on the interval and on seasonality
Minimum standard deviation	Floor for the standard deviation used in the calculation, avoiding excessive sensitivity when variation is very low	depends on the aggregation function
Minimum distance (%)	Minimum distance the value must exceed the threshold by, as a percentage of the width of the expected band. Entered as a whole number, between 0 and 100	10
Minimum distance (absolute)	Minimum distance in the units of the aggregation. On an aggregation that results in a rate, entered in percentage points	not required
Measurements to close	Consecutive normal measurements required to close an alert, avoiding flapping	depends on the interval
Seasonality adjustment	Removes seasonal patterns from the series before detection, reducing false alarms on cyclical metrics	enabled on count aggregations

The two minimum distances are combined: when both are configured, the deviation must satisfy both to generate an alert.

The form takes the relative distance as a whole percentage. The API and an agent’s queries keep sending and returning the same value as a fraction, so 10 on the form corresponds to 0.1 on the API.

The default values per interval are listed in limits and parameters.