Silhouette Score
Also known as: silhouette coefficient, silhouette index
The Silhouette Coefficient, introduced by Peter Rousseeuw in 1987, is a metric that measures how similar an object is to its own cluster compared to other clusters. It ranges from -1 to 1, where values close to 1 indicate well-separated and cohesive clusters, values near 0 suggest overlapping clusters, and negative values indicate misclustered points.
Key highlights
- Ranges from -1 to 1, making it intuitive to interpret
- Works with any distance metric
- Computationally efficient for moderate dataset sizes
- Provides both global (average) and per-point assessments of cluster quality
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 Silhouette Score when you need a simple, interpretable metric to validate clustering quality. It works well for spherical, well-separated clusters and is computationally efficient. However, it can be biased toward convex clusters and may not perform well when clusters have complex shapes or significantly different densities. It is also sensitive to the choice of distance metric.
Strengths & limitations
- Ranges from -1 to 1, making it intuitive to interpret
- Works with any distance metric
- Computationally efficient for moderate dataset sizes
- Provides both global (average) and per-point assessments of cluster quality
- Biased toward spherical, well-separated clusters
- Poor performance on clusters with varying density or complex shapes
- Assumes that inter-cluster distance is meaningful
- Computational cost increases quadratically with dataset size
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
What does a negative silhouette coefficient mean?
A negative coefficient for a point indicates that the point is, on average, closer to the nearest other cluster than to its assigned cluster. This suggests potential misclassification and that the point might belong to a different cluster or be an outlier.
How do I interpret the overall silhouette score for my clustering?
Compute the average silhouette coefficient across all points. A score greater than 0.5 is generally considered strong, 0.25 to 0.5 is reasonable, and below 0.25 suggests weak cluster structure. However, context and the nature of your data should always guide interpretation.
Does silhouette score require labeled data?
No; silhouette is an unsupervised metric. It evaluates cluster quality based on the clustering assignment itself, not external labels. This makes it useful when ground truth is unavailable.
Can I use silhouette score to choose the optimal number of clusters?
Yes; compute the average silhouette score for different numbers of clusters and select the k that maximizes the score. However, combine this with domain knowledge and other validation metrics for robust cluster number selection.
Sources
- 1.Rousseeuw, P. J. (1987). Silhouettes: a graphical aid to the interpretation and validation of cluster analysis. Journal of Computational and Applied Mathematics, 20, 53-65.
You have read it. What now?
Cite this page
ScholarGate. (2026, June 3). Silhouette Score. ScholarGate. https://scholargate.app/model-evaluation/silhouette-score