search

sintel-dev / Orion

A machine learning library for detecting anomalies in signals.
https://sintel.dev/Orion/
MIT License
1.01k stars 158 forks source link

multivariate output #438

Closed khadijakhaldi closed 10 months ago

khadijakhaldi commented 1 year ago

  1. I have a time series data of 4 sensors, an anomaly happens if one or two sensors have a peak value. If all of them have a peak at the same timestamp it is not an anomaly. As far as I know your models detect an anomaly in only one column. Is that correct ?

  2. If it is the case, does it make sense to run the model 4 times on each of my sensors data columns? because in my case an anomaly depends on all the 4 sensors values at the same timestamp. Thanks in advance!

sarahmish commented 1 year ago

Hi @khadijakhaldi – thank you for using Orion!

  1. yes, all currently available pipelines detect anomalies in only one column.
  2. It definitely does. I recommend that you use all 4 sensors as input the pipeline. However, you alter the target column in each time you run the pipeline.

Hope this helps, let me know if you have any further questions!