SUODDetector<T>

Models & Types · AiDotNet.AnomalyDetection.Ensemble

Detects anomalies using SUOD (Scalable Unsupervised Outlier Detection).

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

Implements: IGeneratedParameterRegistrar<T>

For Beginners

SUOD is an acceleration framework that combines multiple anomaly detection algorithms efficiently. It uses approximation techniques to speed up detection while maintaining accuracy.

How It Works

The algorithm works by:

  1. Train multiple diverse base detectors
  2. Use random projection for dimensionality reduction
  3. Combine scores using robust averaging

When to use:

  • Large datasets where speed matters
  • When you want ensemble benefits with good performance
  • As a general-purpose robust detector

Industry Standard Defaults:

  • Base detectors: LOF, k-NN, Isolation Forest
  • Contamination: 0.1 (10%)

Reference: Zhao, Y., et al. (2021). “SUOD: Accelerating Large-Scale Unsupervised Heterogeneous Outlier Detection.” MLSys.

Constructors

Constructor Summary
SUODDetector(bool useRandomProjection, int nProjectedFeatures, double contamination, int randomSeed) Creates a new SUOD anomaly detector.

Properties

Property Summary
bool UseRandomProjection { get; } Gets whether random projection is used.

Methods

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