OCSVMDetector<T>

Models & Types · AiDotNet.AnomalyDetection.DistanceBased

Detects anomalies using One-Class SVM with simplified SGD training.

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

Implements: IGeneratedParameterRegistrar<T>

For Beginners

One-Class SVM finds a boundary that encompasses the normal data. It learns the shape of normal data in a high-dimensional kernel space and flags points outside this region as anomalies.

How It Works

The algorithm works by:

  1. Map data to kernel space using RBF kernel
  2. Find hyperplane that separates data from origin with maximum margin
  3. Points on the negative side of the hyperplane are anomalies

When to use:

  • When you have only normal examples
  • High-dimensional data
  • When you need a decision boundary

Industry Standard Defaults:

  • Nu: 0.1 (roughly proportion of outliers)
  • Gamma: auto (1/n_features)
  • Kernel: RBF
  • Contamination: 0.1 (10%)

Reference: Schölkopf, B., et al. (2001). “Estimating the Support of a High-Dimensional Distribution.”

Constructors

Constructor Summary
OCSVMDetector(double nu, double gamma, int maxIterations, double contamination, int randomSeed) Creates a new OCSVM anomaly detector.

Properties

Property Summary
double Gamma { get; } Gets the gamma parameter.
double Nu { get; } Gets the nu parameter.

Methods

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