What is an anomaly?
An anomaly is an observation that deviates so much from the expected behavior that it raises suspicions that it was generated by a different data-generating process or mechanism. In other words, an anomaly is an observation that is not expected to happen and that is not expected to happen again.
Anomalies can be due to noise or erroneous data, or they can be due to a change in the data-generating process. A good example of an anomaly due to noise or erroneous data is a sudden drop in conversions from a marketing campaign due to a loss in web tracking. You, as an analyst, want to be able to detect this anomaly and understand that it is due to a loss in tracking, and not due to a change in the data-generating process. Another type of anomaly could be a steep drop in the conversion rate due to a change in the rules that allow your ads to be shown. You still want to detect this anomaly, but here you also want to understand that the data-generating process has changed...
