DDM<T>

Models & Types · AiDotNet.DriftDetection

Implements DDM (Drift Detection Method) for concept drift detection.

Inheritance: DriftDetectorBase<T> → DDM<T>

For Beginners

DDM is one of the simplest and most popular drift detectors. It monitors the error rate of a classifier and triggers when the error rate increases significantly compared to the minimum observed error rate.

How It Works

How DDM works:

  • Track the error rate (p) and its standard deviation (s)
  • Remember the minimum p + s observed (the “best” state)
  • Warning: Current p + s exceeds minimum by more than 2σ
  • Drift: Current p + s exceeds minimum by more than 3σ

Key Concepts:

  • Error rate (p): Running average of errors (0 = correct, 1 = error)
  • Standard deviation (s): sqrt(p * (1-p) / n)
  • Minimum p + s: The best observed performance
  • Warning zone: Performance degraded but not enough for drift
  • Drift zone: Significant performance degradation detected

Advantages: Simple, fast, low memory, works well with gradual drift.

Disadvantages: Only works with error rates (0/1), may miss small drifts.

Reference: Gama et al., “Learning with Drift Detection” (2004)

Constructors

Constructor Summary
DDM(double warningThreshold, double driftThreshold, int minimumObservations, int warningDelay) Creates a new DDM drift detector.

Properties

Property Summary
double DriftThreshold { get; } Gets the drift threshold (default: 3.0 standard deviations).
double WarningThreshold { get; } Gets the warning threshold (default: 2.0 standard deviations).
int WarningDelay { get; } Gets the delay between warning and drift detection (for gradual drift).

Methods

Method Summary
bool AddObservation(T value) Adds a new observation (error indicator) to the detector.
double GetErrorRate() Gets the current error rate.
double GetMinimumPsi() Gets the minimum error rate observed (the baseline).
void Reset()