Splunk Search

Detecting anomaly in session eventcount

Communicator

We have data set which aggregated sessions with it's eventcount for each event.
We are looking at setting up an alert for sessions where eventcount exceeded "normalcy".

For Bell-curved data we'd setup an alert for 2x or 3x STDEV. But in our case eventcount is not really Bell-curved - as it starts right away very high at low eventcount and then gradually gets lower in this manner
x
x
xx
xxxx

xxxxxxxxxxxxxxx

Does Splunk has built-in ways to handle deviations for other types of non-Bell curved data sets?

Tags (2)
0 Karma

Path Finder

You're very astute to recognize that using a "bell curve" Gaussian model (average and std. deviation) is not the most appropriate method to approach this. You could try the Prelert app (https://splunkbase.splunk.com/app/1306/) to detect anomalies instead - it uses machine learning to automatically pick an appropriate probability distribution that best models your data, thus giving more accuracy to outlier detection.

0 Karma

Esteemed Legend

Check out Splunk's "cluster" command (I assume you have already tried the "stdev" function of the "stats" command).

0 Karma
State of Splunk Careers

Access the Splunk Careers Report to see real data that shows how Splunk mastery increases your value and job satisfaction.

Find out what your skills are worth!