Silhouette Silhouette website

An R package for silhouette-based diagnostics in standard, soft, and multi-way clustering.

Quantifies clustering quality by measuring both cohesion within clusters and separation between clusters. Implements advanced silhouette width computations for diverse clustering structures, including: simplified silhouette by Van der Laan et al. (2003), Probability of Alternative Cluster normalization methods by Raymaekers & Rousseeuw (2022), fuzzy clustering and silhouette diagnostics using membership probabilities by Campello & Hruschka (2006), Menardi (2011) and Bhat & Kiruthika (2024), and multi-way clustering extensions such as block and tensor clustering by Schepers et al. (2008) and Bhat & Kiruthika (2025). Provides tools for computation and visualization based on Rousseeuw (1987) to support robust and reproducible cluster diagnostics across standard, soft, and multi-way clustering settings.

Note: This package does not use the classical Rousseeuw (1987) calculation directly.


✅ Why This Package?


Installation

You can install the released version of Silhouette from GitHub using:

# Install devtools if needed
if (!requireNamespace("devtools", quietly = TRUE)) {
  install.packages("devtools")
}

# Install from GitHub
devtools::install_github("kskbhat/Silhouette")

From CRAN, install via:

install.packages("Silhouette")

Usage

library(Silhouette)
km <- kmeans(iris[, 1:4], centers = 3, nstart = 20)
sil <- Silhouette(proxy::dist(iris[, 1:4], km$centers))  # n x K distances
summary(sil)
plot(sil)

The package has four vignettes:

The package website has three further articles: silhouette plots for results from this and other packages, a case study comparing soft clusterings of thyroid patients, and a comparison with other packages, including timings.


Report a Bug or Request a Feature

If you encounter a bug or have an idea for a new feature in the Silhouette package, please let us know by opening an issue on GitHub:

Your feedback and suggestions are valuable and help improve the package.