INFLODetector<T>

Models & Types · AiDotNet.AnomalyDetection.DistanceBased

Detects anomalies using Influenced Outlierness (INFLO).

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

Implements: IGeneratedParameterRegistrar<T>

For Beginners

INFLO combines the concepts of k-nearest neighbors and reverse nearest neighbors. It considers not only which points are close to a given point, but also which points consider the given point as their neighbor.

How It Works

The algorithm works by:

  1. Find k-nearest neighbors (kNN) for each point
  2. Find reverse k-nearest neighbors (RkNN) - points that have this point as a neighbor
  3. Compute influence space = kNN union RkNN
  4. Compare point’s density to its influence space’s density

When to use:

  • When boundary between clusters have outliers
  • When LOF fails at cluster boundaries
  • Similar scenarios to LOF but with better boundary handling

Industry Standard Defaults:

  • K (neighbors): 10
  • Contamination: 0.1 (10%)

Reference: Jin, W., et al. (2006). “Mining Top-n Local Outliers in Large Databases.” KDD.

Constructors

Constructor Summary
INFLODetector(int k, double contamination, int randomSeed) Creates a new INFLO anomaly detector.

Properties

Property Summary
int K { get; } Gets the number of neighbors used for detection.

Methods

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