ABODDetector<T>

Models & Types · AiDotNet.AnomalyDetection.AngleBased

Detects anomalies using Angle-Based Outlier Detection (ABOD).

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

Implements: IGeneratedParameterRegistrar<T>

For Beginners

ABOD detects outliers by analyzing the angles formed between a point and all pairs of other points. Outliers tend to have angles that point in similar directions (low variance), while inliers have diverse angle patterns.

How It Works

The algorithm works by:

  1. For each point p, compute angles between p and all pairs of other points
  2. Calculate the variance of these angles (weighted by distance)
  3. Low variance indicates the point is an outlier

When to use:

  • High-dimensional data (works better than distance-based methods)
  • When the “curse of dimensionality” affects other methods
  • Data has complex structure

Industry Standard Defaults:

  • Contamination: 0.1 (10%)
  • For large datasets, use FastABOD variant

Reference: Kriegel, H., et al. (2008). “Angle-Based Outlier Detection in High-dimensional Data.” KDD.

Constructors

Constructor Summary
ABODDetector(double contamination, int randomSeed) Creates a new ABOD anomaly detector.

Methods

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