I have a dataset for clustering including numerical and nominal variables. I would like to compare the k-means and k-medoids clustering algorithms and I would also like to find the optimal k-value (number of clusters).

I cant use the Davies Bouldin method, because my data contains nominal values. Is there any other way I can compare these algorithms to see which one performs better?

  • $\begingroup$ You might use clustering criterions other than those based on ANOVA ideology. Such as point-biserial correlation, Gamma, AIC\BIC, C-Index, Silhouette (just to mention those considered in the description document from my web-page) and some other. $\endgroup$ – ttnphns May 25 '15 at 13:52
  • $\begingroup$ Hello, thank you for your response. I don't really understand your suggestion. I'm using the rapidminer tool, can I use aic/bic after lets say k-means? $\endgroup$ – Nab May 25 '15 at 13:55
  • $\begingroup$ My comment listed some of criterions which can be used to select the "better" cluster partition among partitions of different "k" when data are not continuous. I cannot take care about what specific software you are using. You say you use rapidminer. Does it provide any clustering criterions I've listed? $\endgroup$ – ttnphns May 25 '15 at 14:44
  • $\begingroup$ No it does not contain any of the once you listed. $\endgroup$ – Nab May 26 '15 at 17:26

Your Answer

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

Browse other questions tagged or ask your own question.