Enable Anomaly Detection

You need to enable Anomaly Detection separately for each application.

  1. In Alert & Respond > Anomaly Detection, select one of the following applications from the dropdown:
    • Applications

    • Databases

    • User Experience: Browser Apps
    • User Experience: Mobile Apps
  2. Toggle Anomaly Detection ON.
    After you enable Anomaly Detection, it takes 48 hours for Anomaly Detection and Automated Root Cause Analysis to become available. During that time, the machine learning models train on your application.
  3. Select Model Training to view the training status for your application servers, business transactions, base pages, databases, and network requests as applicable.

    The following table explains the training statuses:

    StatusMeaning
    In TrainingModel training is in progress for the application server, business transaction, base page, database, or network request.
    ReadyModel training is complete and the application server, business transaction, base page, database, or network request is healthy.
    WarningModel training is complete, but the application server, business transaction, base page, database, or network request has experienced one or more Warning level anomalies during the training period.
    CriticalModel training is complete, but the application server, business transaction, base page, database, or network request has experienced one or more Critical level anomalies during the training period.
    Not AvailableModel training is incomplete and the application server, business transaction, base page, database, or network request is not visible to Anomaly Detection.
The models continue training as long as Anomaly Detection is enabled. For example, if traffic to a Business Transaction is interrupted for long enough duration preventing training that day, Anomaly Detection continues to function using the models from the previous seven days.
Note: No machine learning models are trained for Business Transactions that have very low calls per minute (CPM), because the sample size will be so small that the resulting model will be unreliable.