Can clustering-based methods, rather than neural networks, be fitted on the entire time series ？
I noticed that the unsupervised models in this repo seem to be fit on the full sequence.
If that is the case, we propose a new training-free model that only uses clustering and requires no backpropagation.It achieves an impressive VUS-PR of 0.57 for univariate anomaly detection
Can clustering-based methods, rather than neural networks, be fitted on the entire time series ？
I noticed that the unsupervised models in this repo seem to be fit on the full sequence.
If that is the case, we propose a new training-free model that only uses clustering and requires no backpropagation.It achieves an impressive VUS-PR of 0.57 for univariate anomaly detection