MatrixProfileDetector<T>

Models & Types · AiDotNet.AnomalyDetection.TimeSeries

Detects anomalies using Matrix Profile for time series discord detection.

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

Implements: IGeneratedParameterRegistrar<T>

For Beginners

Matrix Profile computes the distance to the nearest neighbor subsequence for every subsequence in a time series. Subsequences with no similar matches (discords) are anomalies. It’s one of the most powerful time series anomaly detection methods.

How It Works

The algorithm works by:

  1. Extract all subsequences of length m
  2. For each subsequence, find the nearest neighbor (excluding trivial matches)
  3. Store these distances as the Matrix Profile
  4. High values in the Matrix Profile indicate discords (anomalies)

When to use:

  • Time series pattern anomaly detection
  • Finding unusual subsequences
  • Discord discovery in time series

Industry Standard Defaults:

  • Subsequence length: 50 (or approximately one period)
  • Exclusion zone: m/4
  • Contamination: 0.1 (10%)

Reference: Yeh, C.C.M., et al. (2016). “Matrix Profile I: All Pairs Similarity Joins for Time Series: A Unifying View.” IEEE ICDM.

Constructors

Constructor Summary
MatrixProfileDetector(int subsequenceLength, int exclusionZone, double contamination, int randomSeed) Creates a new Matrix Profile anomaly detector.

Properties

Property Summary
int ExclusionZone { get; } Gets the exclusion zone size.
int SubsequenceLength { get; } Gets the subsequence length.

Methods

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