Back to all skills
AnalyticsVersion 1.0.0Low

Anomaly Detection

Detects data anomalies and proposes checks. Use when a number deviates from the norm.

Pack: Reports & Decisions

Procedure

  1. 1Compute mean and standard deviation of the latest period
  2. 2Flag values beyond 2 deviations from the mean
  3. 3Compare with the same period last year to rule out seasonality
  4. 4Classify: system anomaly, isolated event, data error
  5. 5Propose the concrete check for each

Rules

  • · Every anomaly carries the check that would confirm it
  • · No alarm without the comparison value
  • · Possible data error: always check the source

Expected output

Anomaly list (metric, value, mean, deviation, classification, proposed check)

When not to use it

  • · Unexplained anomalies: never treat them as certain
  • · Seasonal data (holidays, sales): compare with the same period last year

Example requests

«February collapsed, what happened?»