What are anomalies and how to detect them statistically?

Anomalies are data or events that significantly differ from expected or normal behavior.

How to detect statistically:

• Use distribution-based methods, such as:
  • Checking for values outside a few standard deviations from the mean (e.g., 3σ rule).
  • Using interquartile range (IQR) to identify outliers.
• Apply statistical tests, such as Grubbs' test for detecting single outliers.
• Use forecasting models and look for large deviations of actual data from predicted values.

For example, if the average metric value is 100 and the standard deviation is 10, then values above 130 or below 70 may be considered anomalies.