Detecting outliers automatically
In these examples so far, we detected outliers with a simple visual inspection of the data and applied common sense. In a fully automated setting, defining logical rules for what we as humans do intuitively can be difficult. Outlier detection is a good use of an analyst's time as we humans are able to use much more intuition, domain knowledge, and experience than a computer can. But as Prophet was developed to reduce the workload of analysts and automate as much as possible, we'll examine a couple of techniques to identify outliers automatically.
Winsorizing
The first technique is called Winsorization, named after the statistician Charles P. Winsor. It is also sometimes called clipping. Winsorization is a blunt tool and tends not to work well with non-flat trends. Winsorization requires the analyst to specify a percentile; all data above or below that percentile is forced to the value at the percentile.
Trimming is a similar technique...
