MCDDetector<T>

Models & Types · AiDotNet.AnomalyDetection.Linear

Detects anomalies using Minimum Covariance Determinant (MCD).

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

Implements: IGeneratedParameterRegistrar<T>

For Beginners

MCD is a robust method for estimating the center and spread of data. Unlike standard mean and covariance, MCD is resistant to outliers by finding the subset of points that minimizes the covariance determinant. Points far from this robust estimate are anomalies.

How It Works

The algorithm works by:

  1. Find the subset of h points with minimum covariance determinant
  2. Compute robust mean and covariance from this subset
  3. Compute Mahalanobis distances using robust estimates
  4. Points with large distances are anomalies

When to use:

  • Data with outliers that corrupt standard statistics
  • When you need robust location/scatter estimates
  • Multivariate anomaly detection

Industry Standard Defaults:

  • Support fraction: 0.5 (use 50% of data for robust estimate)
  • Contamination: 0.1 (10%)

Reference: Rousseeuw, P.J. (1984). “Least Median of Squares Regression.” Rousseeuw, P.J., Driessen, K.V. (1999). “A Fast Algorithm for the Minimum Covariance Determinant Estimator.”

Constructors

Constructor Summary
MCDDetector(double supportFraction, double contamination, int randomSeed) Creates a new MCD anomaly detector.

Properties

Property Summary
double SupportFraction { get; } Gets the support fraction.

Methods

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