Which is the best (fastest) algorithm for finding no of clusters for k means while clustering a multi-spectral image, Silhouette value , or Dunn Index or Davies Bouldin index ?

I am using Silhouette value for cluster cohesion evaluation, but finding silhouette for all possible no of clusters makes it slow, so i rather want to find the optimum no of clusters first that exists in the data and then use the silhouette value for further cluster compactness evaluation

Answer 1

[Luciano Garim]

Use evalclusters. There are many ways to detect the right number of clusters using this function.

Type help evalclusters for more information.