Summary

In this chapter, you have learned how anomaly detection works, both in streaming and non-streaming contexts. This category of machine learning models takes a number of variables about a situation and uses this information to detect whether specific data points or observations are likely to be different from the others.

You have gotten an overview of different use cases for this. Some of those are the monitoring of IT systems, or production line sensor data in manufacturing. Whenever it is problematic to have a data point that is too different from the others, anomaly detection is of great added value.

You have finished the chapter by implementing a model benchmark in which you have benchmarked two online anomaly detection models from the River library. You have seen one model being able to detect a part of the anomalies, and the other model having much worse performances. This has introduced you not only to anomaly detection but also to model benchmarking and model evaluation...