OneClassSVM<T>

Models & Types · AiDotNet.AnomalyDetection.Linear

Implements One-Class SVM for novelty/outlier detection using the RBF kernel.

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

Implements: IGeneratedParameterRegistrar<T>

For Beginners

One-Class SVM learns a “boundary” around your normal data points. New points that fall outside this boundary are considered outliers. It’s like drawing a flexible shape around your data - anything outside the shape is unusual.

How It Works

The algorithm works by:

  1. Mapping data to a high-dimensional space using the RBF (Radial Basis Function) kernel
  2. Finding a hyperplane that separates most data points from the origin
  3. Points far from this hyperplane (on the wrong side) are outliers

When to use: One-Class SVM is particularly effective for:

  • Novelty detection (training only on “normal” data)
  • When you have a clear notion of what “normal” looks like
  • Data with complex, non-linear boundaries

Industry Standard Defaults:

  • Nu: 0.1 (upper bound on outlier fraction)
  • Gamma: Auto-detect (1/n_features)
  • Max iterations: 1000
  • Tolerance: 1e-3

Reference: Schölkopf, B., Platt, J. C., Shawe-Taylor, J., Smola, A. J., and Williamson, R. C. (2001). “Estimating the Support of a High-Dimensional Distribution.” Neural Computation.

Constructors

Constructor Summary
OneClassSVM(double nu, double gamma, int maxIterations, double tolerance, double contamination, int randomSeed) Creates a new One-Class SVM anomaly detector.

Properties

Property Summary
double Gamma { get; } Gets the gamma parameter for the RBF kernel.
double Nu { get; } Gets the nu parameter (upper bound on outlier fraction).
int NumSupportVectors { get; } Gets the number of support vectors after fitting.

Methods

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