I have the goal to cluster about 150 quantitative variables according to their correlations. I used hierarchical clustering and obtained a number of clusters (blocks with good significant level). The next step I'd like to perform - measuring of intra block interactions. Are there any statistics more interesting than mean pairwise correlations?
you have cluster to quantitative variables for hierarchical clustering and obtained a number of clusters and make up group optimize working then simulation group. Beginning fuzzy clustering test about 150 quantitative variables and loop all group interesting.
Dear Alexander
in fact you are asking for a quality measure of your clustering
regardless where the clustering came from.
Be aware, since you are NOT judging a possible "classification performance"
only inner quality measures apply.
There are several approaches known in literature to do this.
Some of the popular ones (not necessarily the most adequate) are:
- Silhouette coefficient
- Gap Statistics
- Calinsky Harabassz coefficient
- non-parametric mutual information
I think, especially the gap statistics should be helpful for your task.
regards
Nils
Maybe you could try this:
http://www.sciencedirect.com/science/article/pii/S0169743910000663
Code is in https://github.com/josecamachop/MEDA-Toolbox
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Applied Statistics