Fowlkes-Mallows Index
Fowlkes-Mallows Index for Clustering Agreement · Also known as: Fowlkes Mallows, FM index
The Fowlkes-Mallows Index, introduced by Fowlkes and Mallows in 1983, is an external clustering evaluation metric based on the geometric mean of precision and recall. It measures agreement between two partitions by examining pairs of points and how they are grouped in both the predicted and ground truth clusterings. Values range from 0 to 1, with 1 indicating perfect agreement.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use the Fowlkes-Mallows Index when you want a metric based on precision and recall of pairwise relationships. It is useful when the consistency of pairwise groupings is more important than overall partition structure. FM Index works well for comparing different clustering algorithms or parameter settings on labeled benchmarks.
Strengths & limitations
- Intuitive extension of precision-recall to clustering
- Ranges from 0 to 1, making it easy to interpret
- Directly measures agreement on pairwise relationships
- Balances precision and recall through geometric mean
- Works with any distance metric and clustering algorithm
- Requires ground truth labels; not suitable for unsupervised evaluation
- Computationally expensive for large datasets; O(n^2) pair counting required
- Can be sensitive to class imbalance and cluster size differences
- Less commonly used than ARI or NMI, so fewer reference benchmarks available
Frequently asked
How is the Fowlkes-Mallows Index different from the Adjusted Rand Index?
Both are pair-counting metrics but differ in their formulas. FM Index uses precision and recall with a geometric mean; ARI corrects for chance agreement. FM Index ranges from 0 to 1; ARI ranges from -1 to 1. Both are valid; choose based on whether you prefer correcting for chance (ARI) or balancing precision-recall (FM).
Is the Fowlkes-Mallows Index symmetric?
No, the FM Index is not symmetric; FM(A, B) may differ from FM(B, A). Treat one partition as the reference and the other as predicted for consistent interpretation.
Can FM Index be used for hierarchical clustering?
Yes, the FM Index was originally designed for hierarchical clusterings. For hierarchical methods, compare clusterings at the same cutting height or level of granularity to get meaningful comparisons.
How do I handle missing or uncertain ground truth labels?
If ground truth is uncertain or partially available, FM Index may not be appropriate. Consider semi-supervised or unsupervised metrics instead, or use FM Index only on confidently labeled subsets of data.
Sources
- Fowlkes, E. B., & Mallows, C. L. (1983). A method for comparing two hierarchical clusterings. Journal of the American Statistical Association, 78(383), 553-569. DOI: 10.1080/01621459.1983.10478008 ↗
How to cite this page
ScholarGate. (2026, June 3). Fowlkes-Mallows Index for Clustering Agreement. ScholarGate. https://scholargate.app/en/model-evaluation/fowlkes-mallows-index
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Adjusted Rand IndexModel Evaluation↔ compare
- Davies-Bouldin IndexModel Evaluation↔ compare
- Normalized Mutual InformationModel Evaluation↔ compare
- Silhouette ScoreModel Evaluation↔ compare
- V-measureModel Evaluation↔ compare