Anomaly Advisor Tab

The Anomaly Advisor tab helps you identify potentially anomalous metrics and charts by focusing on a highlighted timeframe of interest. This feature uses Netdata's Anomaly Rate ML scoring to rank metrics based on unusual behavior.

Each chart in the Metrics tab also features an Anomaly Rate ribbon for anomaly visibility.

For configuration details, see the ML documentation.

---

How Anomaly Advisor Works

The Anomaly Advisor leverages Netdata’s machine learning to evaluate anomaly rates across your nodes. It provides three key visualizations:

Chart Name	Purpose	Why It Matters
Anomaly Rate	Shows the percentage of anomalous metrics over time per node.	Helps you quickly spot nodes behaving abnormally.
Count of Anomalous Metrics	Displays raw counts of anomalous metrics per node.	Useful when nodes have different numbers of collected metrics.
Anomaly Events Detected	Indicates when the anomaly rate has triggered a node-level event.	Focuses your attention on meaningful spikes, not just noise.

note

Anomaly Events Detected appear slightly after anomaly rates rise, as they require a significant portion of metrics on the node to show anomalous behavior.

---

Workflow Overview

1. Highlight a timeframe of interest on the anomaly charts.
2. An ordered list of related charts appears, ranked by anomaly level.
3. The Anomaly Rate ribbon (purple) is visible on each chart.
4. Use the right-hand anomaly index to sort metrics from most to least anomalous.
5. Click an entry in the index to navigate directly to the corresponding chart.

tip

Use the node filter to focus on specific nodes before highlighting a timeframe.

---

Usage Tips

Tip	Why It Matters
Filter to specific nodes before highlighting.	Reduces noise by limiting averaging across unrelated nodes.
Highlight close to the anomaly spike.	Improves ranking accuracy by focusing on the relevant timeframe.

---

Anomaly Advisor Diagram

tip

This diagram shows the Anomaly Advisor flow: highlight, rank, and explore. Use the ranking to prioritize which charts to investigate.