-1

In my case, All the time-series observations are with high dimensions. Very likely, they will fall in multiple clusters(meaning multiple distribution patterns) other than a single cluster(meaning single distribution pattern). Will anomaly detection be able to cope with this scenario?

Shengjie
  • 231
  • 1
  • 2
  • 7

1 Answers1

1

This question is difficult to answer because anomaly detection is a whole field and not just one technique. Some simple techniques would not be able to pick up on these types of patterns but most advanced techniques would be able to pick up on them. It depends on the underlying techniques that are used, if it uses a local density estimation technique then time series that lie close to a lot of other time series will not be annotated as anomalies. If you would look at certain statistics only due to high dimensions this might not be the case.

Jan van der Vegt
  • 9,368
  • 35
  • 52