RobustPCADetector<T>

Models & Types · AiDotNet.AnomalyDetection.Linear

Detects anomalies using Robust PCA (Principal Component Pursuit).

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

Implements: IGeneratedParameterRegistrar<T>

For Beginners

Robust PCA decomposes data into a low-rank component (normal patterns) and a sparse component (anomalies). Unlike standard PCA, it’s robust to outliers because it explicitly models them as sparse corruptions.

How It Works

The algorithm works by:

  1. Decompose data matrix M = L + S
  2. L is low-rank (captures normal patterns)
  3. S is sparse (captures anomalies)
  4. Solve via convex optimization (ADMM)

When to use:

  • Data corrupted by sparse anomalies
  • When standard PCA is affected by outliers
  • Video surveillance (background subtraction)
  • Network intrusion detection

Industry Standard Defaults:

  • Lambda: 1/sqrt(max(n,m))
  • Max iterations: 1000
  • Tolerance: 1e-7
  • Contamination: 0.1 (10%)

Reference: Candès, E.J., et al. (2011). “Robust Principal Component Analysis?” JACM.

Constructors

Constructor Summary
RobustPCADetector(double lambda, int maxIterations, double tolerance, double contamination, int randomSeed) Creates a new Robust PCA anomaly detector.

Properties

Property Summary
double Lambda { get; } Gets the lambda parameter (sparsity penalty).
int MaxIterations { get; } Gets the maximum iterations.

Methods

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