Suppose we have a clustering problem where data sample is of multi-dimension with a mix of numeric and categorical type.

If the problem is static i.e. we have all the data, then we can solve this problem by using K-prototype algorithm (variant of K-Means algorithm). But what if data comes dynamically, how can we solve this problem in such cases

Possible constraints:

  • Data comes dynamically
  • Number of clusters is not fixed (it will increase with time)
  • If similarity(new_data_sample) < threshold for all the clusters then the new cluster should be created containing new_data_sample

enter image description here

  • $\begingroup$ What you call dynamic is also commonly referred to as online learning $\endgroup$ – mapto Jul 3 '18 at 14:27
  • $\begingroup$ Also, notice that the rule that you introduce in your third bullet point is greedy and might result in unnecessary high number of clusters. Think of examples where redefining one of the existing clusters to move a bit towards the new data point is better than just creating a new cluster. $\endgroup$ – mapto Jul 3 '18 at 14:44

See the existing methods on

data stream clustering



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.