Dunn Index
Dunn Index for Cluster Compactness and Separation · Also known as: Dunn's index, separation coefficient
The Dunn Index, introduced by Joseph C. Dunn in 1974, is a metric that captures cluster quality by measuring the ratio of the minimum between-cluster distance to the maximum within-cluster diameter. Higher values indicate well-separated and compact clusters, with better clustering quality.
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 Dunn Index when you need a metric that directly measures cluster compactness and separation. It is useful for selecting the optimal number of clusters and for comparing clustering results from different algorithms. However, computing all pairwise distances makes it computationally expensive for large datasets. It also assumes that compact, well-separated clusters are desirable, which may not hold for all data distributions.
Strengths & limitations
- Intuitive interpretation: higher values always indicate better clustering
- Directly measures both compactness (small diameter) and separation (large minimum distance)
- Works with any distance metric and clustering algorithm
- No external labels required; purely unsupervised evaluation
- Computationally expensive; requires computing all pairwise distances and cluster diameters
- Biased toward compact, well-separated clusters; performs poorly on non-convex or density-varying clusters
- Sensitive to outliers; a single outlier can increase cluster diameter significantly
- Can be misleading when clusters have very different sizes or densities
Frequently asked
What is a good Dunn Index value?
Higher is better; there is no universal threshold, but values greater than 1.0 generally indicate well-separated clusters. Interpretation should be relative: compare Dunn Index values across different numbers of clusters or different algorithms to select the best solution.
Why is the Dunn Index computationally expensive?
The Dunn Index requires computing the distance between all pairs of points to determine the minimum between-cluster distance and maximum within-cluster diameter. For n points, this requires O(n^2) distance calculations, which becomes prohibitive for large datasets.
How sensitive is the Dunn Index to outliers?
Highly sensitive. A single outlier can significantly increase a cluster's diameter, reducing the Dunn Index. Preprocess data to handle outliers before computing the index, or use robust variants that limit the effect of extreme values.
Can I use the Dunn Index to select the optimal number of clusters?
Yes; compute the Dunn Index for different numbers of clusters and select the k that maximizes the index. However, due to computational expense, this approach is practical only for small to medium datasets.
Sources
- Dunn, J. C. (1974). Well-separated clusters and optimal fuzzy partitions. Journal of Cybernetics, 4(1), 95-104. DOI: 10.1080/01969727408546059 ↗
How to cite this page
ScholarGate. (2026, June 3). Dunn Index for Cluster Compactness and Separation. ScholarGate. https://scholargate.app/en/model-evaluation/dunn-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.
- Calinski-Harabasz IndexModel Evaluation↔ compare
- Davies-Bouldin IndexModel Evaluation↔ compare
- Gap StatisticModel Evaluation↔ compare
- Inertia (Within-Cluster Sum of Squares)Model Evaluation↔ compare
- Silhouette ScoreModel Evaluation↔ compare