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:
- Train multiple diverse base detectors
- Use random projection for dimensionality reduction
- 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) |