R: Clustering validation methods (mixed data)

1

votes

I have clustered mixed dataset contains numerical and categorical features (heart dataset from UCI) using two clustering methods k-prototype and PAM

My question is: how to validate the results of clustering?

I have found different methods in R such as Rand Index, SSE, Purity, clValid, pvclust all of them works with numeric data.

Is there any method can be used in the case of mixed data

r validationcluster-analysis

0

votes

Yeah u can compare the clustering result with, CV index. For more u can read this Cv index CV formula contains of CU (Category Utility) for categorical attributes, and varians for numeric attributes

0

votes

You can still use the Adjusted Rand Index. This index only compares two partitions. It does not matter if the partition is build from categorical or continuous features

0

votes

How many observations (n) and dimensions (d) are you particularly studying? Probably you are in the n>>d case, but more recently d>>n is a hot topic.

Variable selection is something that needs to be done before-hand. Check for feature correlation, this can affect the number of clusters that you detect. If the features are correlated and they happen to be linear, you can use the gradient instead of the two variables.

There is no absolute answer to your question. Many methods exist because of this. Clustering is explorative by nature. The better you know your data the better you can design tests.

Need to define what you want to test: stability of the partition, or, the stability of the clustering recipe. There are different ways to deal with each of these problems. For the first one, resampling is a key, and, for the second one, the use of comparison indexes to measure how many observations were left out of certain partition is often used.

R: Clustering validation methods (mixed data)

3 Answers