Outliers Multivariate

Prophet-based outlier detection, multivariate

/home/runner/work/strom/strom/.venv/lib/python3.10/site-packages/plotly/io/_json.py:558: UserWarning:

Discarding nonzero nanoseconds in conversion.

Adding these two relevant predictors -without much feature-engineering really-, substantially improves the fit and reduces the uncertainty band. This could actually be useful for outlier detection. Only concern I had, I would have expected the uncertainty band to be wider in the cold season (due to extreme values and higher variability). Additional concern, it can predict negative values, so we sould add some contraint about that.

Yeah, I think this is actually better. Still, the thing about the uncertainty during the cold periods being not so big, I think it is concerning. It end up detecting some outliers that are probably legit. It also may end up failing to catch some outliers in the warm period.

TODOs