Study: multivariate time series anomalies are mostly univariate

Arxiv CS.LG (machine learning)

Evaluation on eight multivariate time series benchmarks reveals that most anomalies affect only a single channel, calling into question the need for cross-channel modeling. The finding suggests recent MT-SAD model designs may be overcomplicated.

Arxiv CS.LG (machine learning)·Marc Pinet (LIG)Julien Cumin (LIG)Samuel Berlemont (LIG)Dominique Vaufreydaz (LIG)

8 days ago