Anomaly detection via z-score analysis on residuals from a trend model.
Fits the same polynomial + seasonal model as TimelessMetrics.Forecast, computes
residuals (actual - predicted), and flags points where the z-score exceeds
a threshold based on the chosen sensitivity level.
Sensitivity thresholds:
:low— z > 3.0 (flags only extreme outliers):medium— z > 2.5 (default, good balance):high— z > 2.0 (flags more subtle anomalies)
Summary
Functions
Detect anomalies in time series data.
Functions
Detect anomalies in time series data.
Parameters
data- List of{timestamp, value}tuplesopts- Options::sensitivity-:low,:medium, or:high(default::medium):periods- Seasonal period lengths in seconds (default: auto-detected)
Returns {:ok, [%{timestamp, value, expected, score, anomaly}, ...]} or {:error, reason}.