ML-Based Apdex Alerts ¶
Overview ¶
Apdex is a time-series metric. Traditional anomaly alerting on time series often relies on fixed thresholds, which require strong prior knowledge and do not adapt to changing patterns. To automatically detect anomalies, Apdex alerting supports a prediction-based mechanism: when the difference between the actual value and the predicted value exceeds a user-defined threshold, an alert is triggered.
How it works ¶
The alert compares the actual Apdex at each evaluation point with the predicted Apdex. If the deviation crosses the configured threshold, the alert fires.
Threshold types ¶
Supported deviation types:
- Absolute difference
- Relative difference (percentage)
- Z-score (number of standard deviations)
You can configure the deviation type and threshold on the Apdex page. See the following example: 
Prediction method ¶
The prediction for Apdex is derived from statistical time-series modeling to capture trend and seasonality. In dashboards, the predicted series is shown as the "predict" line for easy comparison against actuals: 