Automatically detect anomalies

[atom-box]

(A user migrating from another solution asked whether Matomo has the following) Add an option in the tool to help to find anomalies.

[Findus23]

Just a few links about this topic before I forget them again:

• https://learn.netdata.cloud/docs/agent/collectors/python.d.plugin/anomalies
• https://github.com/netdata/community/tree/main/netdata-pandas-examples
• https://nbviewer.org/github/netdata/community/blob/main/netdata-pandas-examples/anomalies_collector_deepdive.ipynb
• https://nbviewer.org/github/netdata/community/blob/main/netdata-pandas-examples/anomaly_detection.ipynb
• https://nbviewer.org/github/netdata/community/blob/main/netdata-pandas-examples/correlation_heatmap.ipynb

I'm sure this concept can be easily mapped to e.g. pageviews (or other metrics) per day. And I didn't look it up, but there is probably also some existing research in that space.

This could probably even be implemented in PHP, similar to how Nextcloud did their neural-network-based plugin:

• https://apps.nextcloud.com/apps/suspicious_login
• https://github.com/RubixML/ML

[Findus23]

See also https://github.com/matomo-org/matomo/issues/5959, https://github.com/matomo-org/matomo/issues/14646 But I'm not sure if one would even need machine learning/neural networks for this or if something like PCA might already be good enough and easier to reason about.