LOCIDetector<T>

Models & Types · AiDotNet.AnomalyDetection.DistanceBased

Detects anomalies using Local Correlation Integral (LOCI).

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

Implements: IGeneratedParameterRegistrar<T>

For Beginners

LOCI is a density-based outlier detection method that automatically determines the appropriate neighborhood size. It computes a Multi-Granularity Deviation Factor (MDEF) that indicates how much a point’s local density deviates from its neighborhood’s density.

How It Works

The algorithm works by:

  1. For multiple radius values, compute local density
  2. Compare point’s density to average density of neighbors (MDEF)
  3. Flag points with MDEF exceeding a threshold based on standard deviation

When to use:

  • When you don’t want to manually tune neighborhood size
  • Data has varying local densities
  • You want automatic threshold determination

Industry Standard Defaults:

  • Alpha: 0.5 (neighborhood multiplier)
  • Contamination: 0.1 (10%)

Reference: Papadimitriou, S., et al. (2003). “LOCI: Fast Outlier Detection Using the Local Correlation Integral.” ICDE.

Constructors

Constructor Summary
LOCIDetector(double alpha, int kMax, double contamination, int randomSeed) Creates a new LOCI anomaly detector.

Properties

Property Summary
double Alpha { get; } Gets the alpha parameter (sampling neighborhood ratio).
int KMax { get; } Gets the maximum number of neighbors to consider.

Methods

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