Clustering Validity Metrics

Computes validity metrics for a partition of the selected numeric columns: per-observation silhouette widths plus the average silhouette, the Davies-Bouldin index, the Calinski-Harabasz index, and the Dunn index. All metrics are computed in base R from the (optionally standardized) analysis matrix and its distances.

Usage

mv_cluster_metrics(
  data,
  cols,
  cluster,
  dist_method = "euclidean",
  scale = TRUE
)

Arguments

data

A data frame.

cols

<tidy-select> Numeric columns used for distances.

cluster

A vector of cluster labels with one value per row of data (factor, character, or numeric).

dist_method

Distance method passed to stats::dist.

scale

Logical. If TRUE (default), columns are standardized first.

Value

An object of class mv_cluster_metrics: a list with components silhouette (tibble with .row, cluster, neighbor, sil_width), summary (long tibble with metric and value: avg_silhouette, davies_bouldin, calinski_harabasz, dunn), data_mat (the standardized analysis matrix), and meta.

Examples

set.seed(1)
km = mv_kmeans(iris, Sepal.Length:Petal.Width, centers = 3)
m = mv_cluster_metrics(iris, Sepal.Length:Petal.Width, km$cluster$cluster)
m$summary
#> # A tibble: 4 × 2
#>   metric               value
#>   <chr>                <dbl>
#> 1 avg_silhouette      0.460 
#> 2 davies_bouldin      0.834 
#> 3 calinski_harabasz 242.    
#> 4 dunn                0.0265
plot_mv_silhouette(m)