We are trying to use a clustering technique to isolate and analyze bugs when the software is in use in the production system. Here, features of the data are whether a user used certain features/capabilities or not; each representing a boolean column. Some columns are numerical as well. Each row would be an independent user session.

Once we identify the clusters where an error occurred, we would recognize which combination of features used together can potentially cause the error.

The challenge for us is to interpret the results once the cluster forms. Is there any technique/algorithm where we can interpret the given cluster as a nature of issues involved? For example, to say that "when widget x has specific actions A,B,C it fails"

In other words, I am trying to find out, that which feature-set and its value range has the defining influence of that cluster.

I am looking for a generic algorithm/technique to interpret what constitutes a certain cluster in terms of the original feature set.


You don't want clustering.

Use getting itemset mining to find which feature combination best predicts failure.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.