Skip to content

AnomalyClassificationEvents¤

Classify anomalies in a numeric signal by type: spike, drift, oscillation, flatline, or level shift.

Module: ts_shape.events.quality.anomaly_classification
Guide: Quality Control & SPC

When to Use¤

Use after initial anomaly detection to understand the nature of anomalies. Goes beyond "is this an outlier?" to tell you what kind of anomaly it is, enabling targeted corrective action. Works best when combined with an upstream outlier detection step to narrow the signal regions of interest.

Quick Example¤

from ts_shape.events.quality.anomaly_classification import AnomalyClassificationEvents

classifier = AnomalyClassificationEvents(df, value_column="value_double")

# Full multi-type classification in one pass
classified = classifier.classify_anomalies(window="10m", z_threshold=3.0)

# Or detect specific anomaly types individually
flatlines = classifier.detect_flatline(min_duration="1m", tolerance=1e-6)
oscillations = classifier.detect_oscillation(window="1m", min_crossings=6)
drifts = classifier.detect_drift(window="1h", min_slope=0.01)

Key Methods¤

Method Purpose Returns
detect_flatline(min_duration='1m', tolerance=1e-6) Detect stuck/frozen signal segments DataFrame with flatline event intervals
detect_oscillation(window='1m', min_crossings=6) Detect excessive oscillation (e.g., control valve hunting) DataFrame with oscillation flags
detect_drift(window='1h', min_slope=0.01) Detect gradual drift in signal mean DataFrame with drift events
classify_anomalies(window='10m', z_threshold=3.0) Multi-type classification: spike, drift, oscillation, flatline, level shift DataFrame with anomaly type labels

Tips & Notes¤

Run outlier detection first, then classify

Use OutlierDetectionEvents to flag anomalous regions, then pass those regions to classify_anomalies() for root-cause identification. This two-step approach reduces false positives and makes classification more meaningful.

Related modules

  • Outlier Detection — detect anomalies before classifying them
  • Sensor Drift — specialized drift detection for calibration monitoring
  • Signal Quality — distinguish data quality issues from true process anomalies

See Also¤