Skip to content

OutlierDetectionEvents¤

Processes time series data to detect outliers based on specified statistical methods.

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

When to Use¤

Use when you need to find anomalous data points in sensor signals before further analysis. Works with any numeric timeseries column. Choose Z-score for normal distributions, IQR for skewed data, MAD for noisy/heavy-tailed signals, or IsolationForest for complex multivariate patterns.

Quick Example¤

from ts_shape.events.quality.outlier_detection import OutlierDetectionEvents

detector = OutlierDetectionEvents(df, value_column="value_double")

# Z-score method for normally distributed signals
outliers_z = detector.detect_outliers_zscore(threshold=3.0)

# IQR method for skewed distributions
outliers_iqr = detector.detect_outliers_iqr(threshold=(1.5, 1.5))

# ML-based detection for complex patterns
outliers_if = detector.detect_outliers_isolation_forest(contamination=0.1)

Key Methods¤

Method Purpose Returns
detect_outliers_zscore(threshold=3.0) Z-score outlier detection for normally distributed data DataFrame with outlier flags
detect_outliers_iqr(threshold=(1.5, 1.5)) IQR-based detection for skewed distributions DataFrame with outlier flags
detect_outliers_mad(threshold=3.5) Median Absolute Deviation detection for heavy-tailed signals DataFrame with outlier flags
detect_outliers_isolation_forest(contamination=0.1) ML-based detection for complex multivariate patterns DataFrame with outlier flags

Tips & Notes¤

Choose the right method for your distribution

Z-score assumes normality and works best for Gaussian signals. For process data with occasional spikes or skew, prefer IQR or MAD. Use Isolation Forest when you suspect multivariate interactions between features.

Related modules

  • Anomaly Classification — classify detected outliers by type (spike, drift, oscillation)
  • Signal Quality — check data completeness before running outlier detection
  • SPC Rules — rule-based detection using Western Electric Rules

See Also¤