CBLOFDetector<T>
Models & Types · AiDotNet.AnomalyDetection.ClusterBased
Detects anomalies using Cluster-Based Local Outlier Factor (CBLOF).
Inheritance: ModelBase<T, Matrix<T>, Vector<T>> → AnomalyDetectorBase<T> → CBLOFDetector<T>
Implements: IGeneratedParameterRegistrar<T>
For Beginners
CBLOF combines clustering with local outlier detection. It first clusters the data, then scores each point based on its distance to its cluster center, weighted by the cluster size.
How It Works
The algorithm works by:
- Cluster data into large and small clusters
- For large cluster points: score = cluster_size * distance_to_center
- For small cluster points: score = cluster_size * distance_to_nearest_large_cluster
When to use:
- When anomalies are expected in small clusters or far from large clusters
- Faster than LOF for large datasets
- When cluster structure is meaningful
Industry Standard Defaults:
- K (clusters): 8
- Alpha (large cluster threshold): 0.9 (90% of data)
- Beta (size threshold): 5 points
- Contamination: 0.1 (10%)
Reference: He, Z., et al. (2003). “Discovering Cluster-Based Local Outliers.” Pattern Recognition Letters.
Constructors
| Constructor | Summary |
|---|---|
CBLOFDetector(int nClusters, double alpha, int beta, double contamination, int randomSeed) |
Creates a new CBLOF anomaly detector. |
Properties
| Property | Summary |
|---|---|
double Alpha { get; } |
Gets the alpha parameter (proportion for large clusters). |
int Beta { get; } |
Gets the beta parameter (minimum size for large clusters). |
int NClusters { get; } |
Gets the number of clusters. |
Methods
| Method | Summary |
|---|---|
Vector<T> ScoreAnomalies(Matrix<T> X) |
|
void Fit(Matrix<T> X) |