AdjustedRandIndex<T>

Models & Types · AiDotNet.Clustering.Evaluation

Adjusted Rand Index for comparing cluster assignments.

Implements: IExternalClusterMetric<T>

For Beginners

ARI measures how well clusterings agree.

Consider all pairs of points:

  • True Positive: Same cluster in both
  • True Negative: Different clusters in both
  • Pairs that disagree: One says same, other says different

Rand Index = (TP + TN) / (all pairs) But random clusterings can have high RI by chance!

Adjusted Rand Index corrects for this:

  • ARI = 1: Perfect match
  • ARI = 0: No better than random
  • ARI < 0: Worse than random (rare)

How It Works

The Adjusted Rand Index (ARI) measures agreement between two clusterings, adjusted for chance. Values range from -1 to 1:

  • 1: Perfect agreement
  • 0: Random labeling (expected by chance)
  • Negative: Less agreement than random

ARI = (RI - Expected_RI) / (max(RI) - Expected_RI) Where RI is the Rand Index (fraction of pairs that agree).

Constructors

Constructor Summary
AdjustedRandIndex() Initializes a new AdjustedRandIndex instance.

Methods

Method Summary
double Compute(Vector<T> trueLabels, Vector<T> predictedLabels)