Configure anomaly detection
Anomaly based alerts use a statistical data model to identify atypical website performance.
With the mPulse anomaly detection feature, you don’t have to set static thresholds for alerts. mPulse employs a statistical model that represents a normal behavior range based on historical data for your web users, and triggers an alert when conditions fall outside of that range. When you configure anomaly detection it can take up to 30 minutes for the alert to be created. In order to create an alert with anomaly detection mPulse needs a minimum of 14 days of data to create the initial model, then automatically retrains its detection model based on new data every 30 days.
Anomaly based alerts are useful when you don’t have a specific target static threshold value to monitor. You can implement fewer alerts to monitor your site, because mPulse employs logic that accounts for traffic trend fluctuations throughout the day or week.
- Enter a name and description for the alert condition.
- Select the Add an annotation checkbox, if you want to see notifications on the mPulse Home page indicating that the alert fired.
- Click Anomaly Detection.
- Select the app you want to monitor from the drop-down list.
- Select Timer or Metric.
- Specify how long the anomaly must persist in order to trigger an alert.
- Use the alert sensitivity
slider to indicate whether you want the alert tolerance
thresholds to be more or less sensitive.Tip: Keep the default setting initially. If you find the alert firing too frequently or not frequently enough, use the slider to adjust sensitivity.
- Add filters by dimension if you want to limit the report data.
- Choose an action for notification when the anomaly is detected.
- Orange - Pending creation. When you first configure anomaly detection it may take up to 30 minutes for the alert to be created. During this time the icon is orange.
- Red - The incoming data does not lie within the determined range of normal behavior. The alert stays red until the condition no longer applies.
- Green - The incoming data falls within the determined range of normal behavior.