IAnomalyDetector<T>

Interfaces · AiDotNet.Interfaces

Defines methods for algorithmic anomaly/outlier detection using a fit-predict pattern.

Implements: IFullModel<T, Matrix<T>, Vector<T>>

Known implementations: AnomalyDetectorBase<T>

For Beginners

This interface provides a machine learning-style approach to anomaly detection. Unlike simple statistical methods (like Z-score or IQR), algorithmic detectors learn patterns from your data and can detect more complex types of anomalies.

How It Works

The typical workflow is:

  1. Create a detector with your desired parameters (contamination defaults to 0.1 / 10%)
  2. Call Fit to train the detector on your data
  3. Call Predict to identify anomalies in new data
  4. Optionally use ScoreAnomalies to get anomaly scores for ranking

Anomaly Detection vs Outlier Removal:

  • Anomaly Detection: The primary task - identifying unusual patterns in data
  • Outlier Removal: A preprocessing step that uses anomaly detection to clean data

This interface focuses on anomaly detection. For preprocessing use cases, wrap detectors with DetectorBasedFilter to integrate them into PreprocessingPipeline.

Properties

Property Summary
T Threshold { get; } Gets the threshold used to classify samples as inliers or outliers.
bool IsFitted { get; } Gets a value indicating whether the detector has been fitted to data.

Methods

Method Summary
Vector<T> ScoreAnomalies(Matrix<T> X) Computes the anomaly score for each sample.
void Fit(Matrix<T> X) Trains the anomaly detector on the provided data.