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:
- Find k-nearest neighbors (kNN) for each point
- Find reverse k-nearest neighbors (RkNN) - points that have this point as a neighbor
- Compute influence space = kNN union RkNN
- 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) |