Recently came across articles on MAD and several variations of ESD test for detecting local vs. global anomalies (outliers) in time series data. Anyone has any experiences with such algorithms and any words of wisdom on how to utilize such algorithms to detect local anomalies? Thank you very much,
Is there any benefit other than MAD being more robust (to outliers) than Standard Deviation? what about ESD?
Can’t comment on ESD as I haven’t used that.
As for MAD vs SD, I think the key thing here is understanding your data.
If I know that there might be some “wrong” data points (f.e., sensor malfunction, or other external disturbances) which are affecting my anomaly detection, or I am more interested in detecting gradual change, I will prefer MAD.
If I am more confident about my data quality, and would like to not discard the data points that are considerably different from what is usual, I will stick with SD.
Enter your E-mail address. We'll send you an e-mail with instructions to reset your password.
Check the
documentation
Ask the
Community
Take a look
at
Academy
Cognite
Status
Page
Contact
Cognite Support
