Anomaly detection for cumulative metrics

Some metrics increase over time, only resetting occasionally. A common example is the total number of requests a server has handled since it started. These are known as cumulative metrics.

When using EYER, the recommended approach for handling cumulative metrics is to subtract consecutive values and feed the resulting differences into the system. This allows EYER to monitor the rate of change and provides the most accurate anomaly detection.

If you don’t perform this subtraction, EYER can still detect and work with cumulative metrics automatically—under certain conditions. Specifically:

• The metric must reset no more than once per week.

• It must have increased at least six times in the past two weeks.

Warnings:

• Resets will be flagged as anomalies.

• Around midnight UTC, some cumulative metrics may produce false positives as new daily baselines are being calculated.