Hi everyone,
I was wondering what are the current existing internal clustering index for binary data.
I know already the silhouette and Davis Bouldin for euclidian space, i suppose they work as well in binary space using Hamming distance for example, tell me if i'm wrong.
We can use the distance to points from centroid per cluster as an indication compared to the average distance point to point, but the cost of the computation of the similarity matrix is a bit restrictive when it comes to deal with Big data so if it exist :
1 - Any binary internal clustering index
2 - Applicable to Big data
3 - Other internal index even in euclidian space
Thank you in advance for your wise suggestions
Peace
.
quite a few indexes have been studied here
http://ro.uow.edu.au/cgi/viewcontent.cgi?article=1062&context=commpapers
but definitely not from a "big data" perspective
.
use connectivity and dunn index for calculating the distance in the clusters
.
two points in relation with "massive data"
- if you have a lot of cases, you need not use all of them to compute your indexes, at least when your clusters are still massive (that is not a collection of very many small niches)
- if you have lot of features, such datsets are often quite sparse and performing a dimension reduction as a first step may simplify the analysis (and brings you back to a metric space)
.
- Badges
- Science method