HDBSCANDetector<T>

Models & Types · AiDotNet.AnomalyDetection.ClusterBased

Detects anomalies using HDBSCAN (Hierarchical DBSCAN).

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

Implements: IGeneratedParameterRegistrar<T>

For Beginners

HDBSCAN is an improved version of DBSCAN that automatically finds clusters of varying densities without requiring an epsilon parameter.

How It Works

The algorithm works by:

  1. Compute core distances for each point
  2. Build a mutual reachability graph
  3. Construct a minimum spanning tree
  4. Extract a cluster hierarchy
  5. Points not in any cluster are anomalies

When to use:

  • When clusters have varying densities
  • When you don’t want to tune epsilon
  • Complex data structures with hierarchical cluster patterns

Industry Standard Defaults:

  • Min cluster size: 5
  • Min samples: 5
  • Contamination: 0.1 (10%)

Reference: Campello, R., et al. (2013). “Density-Based Clustering Based on Hierarchical Density Estimates.” PAKDD.

Constructors

Constructor Summary
HDBSCANDetector(int minClusterSize, int minSamples, double contamination, int randomSeed) Creates a new HDBSCAN anomaly detector.

Properties

Property Summary
int MinClusterSize { get; } Gets the minimum cluster size.
int MinSamples { get; } Gets the minimum number of samples for core points.

Methods

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