Skip to main content

Anomaly Detection Experiment

ADJob.png

An anomaly detection experiment provides a platform for iterating over anomaly detection jobs using the following approaches:

  • KPI-driven approach: This section explains what the AD experiment with the KPI-driven approach contains and how to use it.
  • System-driven approach: This section explains what the AD experiment with the system-driven approach contains and how to use it.

Anomaly detection experiments are designed to support users throughout their journey in finding anomalies that can drive business decision-making. This covers stages such as choosing the right approach, finding the ideal configuration for the use case, examining in-sample (model-building) and out-of-sample (detecting) results, inspecting models in detail, and drilling down to the root cause behind a particular detected anomaly or predicted normal behavior.

You can create an anomaly detection experiment for each use case by adding a new experiment with the Anomaly Detection type:

ADExperiment.png

After creating the experiment, you will be taken to the experiment screen where you can select one of the available approaches:

AdExperimentScreen.png