Adjusted Rand Index
Also known as: ARI, adjusted Rand coefficient
The Adjusted Rand Index (ARI), developed by Hubert and Arabie in 1985, is an external clustering evaluation metric that measures the agreement between a predicted clustering and a ground truth labeling. It ranges from -1 to 1, where 1 indicates perfect agreement, 0 indicates random clustering, and negative values indicate performance worse than random chance.
Key highlights
- Accounts for chance agreement; ARI of 0 truly indicates random performance
- Ranges from -1 to 1, making it intuitive and bounded
- Works with any number of clusters in both predicted and true partitions
- Symmetric: ARI(A, B) = ARI(B, A)
Intuition
This section is available to Pro members. Upgrade to Pro
How it works
This section is available to Pro members. Upgrade to Pro
When to use it
Use the Adjusted Rand Index when you have ground truth labels and want to evaluate how well your clustering matches the true labels. It is particularly useful for comparing different clustering algorithms or parameter settings on labeled data. However, it requires that ground truth labels be available, limiting its use in truly unsupervised settings.
Strengths & limitations
- Accounts for chance agreement; ARI of 0 truly indicates random performance
- Ranges from -1 to 1, making it intuitive and bounded
- Works with any number of clusters in both predicted and true partitions
- Symmetric: ARI(A, B) = ARI(B, A)
- Requires ground truth labels; not suitable for purely unsupervised evaluation
- Assumes categorical labels; cannot directly handle hierarchical or overlapping clusters
- Can be sensitive to class imbalance
- Less informative when ground truth is uncertain or subjective
Common pitfalls
This section is available to Pro members. Upgrade to Pro
Applications
This section is available to Pro members. Upgrade to Pro
Frequently asked
When should I use ARI versus other external metrics?
Use ARI when you need a well-established metric that accounts for chance agreement. Alternatives include the Fowlkes-Mallows Index and V-measure, which have different mathematical properties. Choose based on your specific requirements and the nature of your label comparison.
What does a negative ARI mean?
A negative ARI indicates that your clustering is worse than random assignment. This suggests that the predicted clusters are actively misaligned with the ground truth, not just similar-to-random. This outcome signals a need to reconsider your clustering approach or parameters.
Can I use ARI on multi-label or hierarchical data?
Standard ARI assumes exclusive cluster membership; it is not directly designed for multi-label or hierarchical structures. For such data, consider alternative metrics like normalized mutual information or variants designed for specific label structures.
How is ARI different from the unadjusted Rand Index?
The unadjusted Rand Index does not account for chance agreement, making it high even for random clustering. ARI corrects this by subtracting the expected value under random clustering, providing a more meaningful measure of true clustering quality.
Sources
- 1.Hubert, L., & Arabie, P. (1985). Comparing partitions. Journal of Classification, 2(1), 193-218.
- 2.Rand, W. M. (1971). Objective criteria for the evaluation of clustering methods. Journal of the American Statistical Association, 66(336), 846-850.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Adjusted Rand Index. ScholarGate. https://scholargate.app/model-evaluation/adjusted-rand-index