CloudWatch Anomaly Detection Flow

CloudWatch Anomaly Detection Flow A workflow diagram generated by Archify. 01 / Learning Phase 02 / Detection Phase EX / Anomaly Alert Historical Data · historical metrics · Learning Phase Historical Data historical metrics ML Model · learns normal pattern · Learning Phase ML Model learns normal pattern Expected Band · ANOMALY_DETECTION_BAND · Learning Phase Expected Band ANOMALY_DETECTION_BAND Current Metrics · current metrics · Detection Phase Current Metrics current metrics Within Band? · evaluation · Detection Phase · decision Within Band? evaluation decision Normal · within band · Detection Phase Normal within band Anomaly Alert · outside band · Anomaly Alert Anomaly Alert outside band baseline compare train produce band No Yes Legend Processing Decision Metric data Cloud component Notification

Expected band width

  • • ANOMALY_DETECTION_BAND(m1, 2): default width
  • • A larger value widens the band; no guaranteed 95% or 99.7% interval
  • • Alarm evaluation also applies configured M-of-N and missing-data handling

Adjusting the Training Period

  • • The model is created automatically with the first alarm
  • • Exclude maintenance or incident periods from training with ExcludedTimeRanges in put-anomaly-detector