HampelDetector<T>

Models & Types · AiDotNet.AnomalyDetection.Statistical

Detects anomalies using Hampel identifier (median-based outlier detection).

Inheritance: ModelBase<T, Matrix<T>, Vector<T>> → AnomalyDetectorBase<T> → HampelDetector<T>

For Beginners

The Hampel identifier is a robust outlier detection method that uses a sliding window to compute local median and MAD. It’s particularly effective for time series data where local context matters.

How It Works

The algorithm works by:

  1. For each point, compute median and MAD in a local window
  2. Compute the deviation from local median in MAD units
  3. Points exceeding the threshold are anomalies

When to use:

  • Time series with local patterns
  • When global statistics don’t capture local anomalies
  • As a robust alternative to moving average methods

Industry Standard Defaults:

  • Window size: 7 (3 on each side)
  • Threshold: 3 MAD units
  • Scale factor: 1.4826
  • Contamination: 0.1 (10%)

Reference: Hampel, F.R. (1974). “The Influence Curve and its Role in Robust Estimation.”

Constructors

Constructor Summary
HampelDetector(int windowSize, double threshold, double scaleFactor, double contamination, int randomSeed) Creates a new Hampel anomaly detector.

Properties

Property Summary
double ThresholdMAD { get; } Gets the threshold in MAD units.
int WindowSize { get; } Gets the window size.

Methods

Method Summary
Vector<T> ScoreAnomalies(Matrix<T> X)
void Fit(Matrix<T> X)