Does anybody know about the acceptable values for Silhouette Coefficient (or maybe Calinski-Harabasz and Davies-Bouldin index) in K-means clustering?. I know that Silhouette Coefficients close to 1 are supposed to be better but is there any benchmark? for example: from 0.6 a model is considered to be good, from 0.2 to 0.6 is acceptable, and so on?

  • 1
    $\begingroup$ Generally, there is no "acceptable value" or "threshold" for internal clustering criteria. They are used typically in a comparative, relative fashion. Criterion's value are plotted and compared for the clustering solutions you want to choose the best from. Still, classic Silhouette index, because it has fixed bounds of variation (-1 to 1), has heuristic, tentative benchmarks: <0.3 "bad", >0.7 "good" clustering. $\endgroup$ – ttnphns Feb 8 at 12:59
  • $\begingroup$ See stats.stackexchange.com/q/52838/3277 $\endgroup$ – ttnphns Feb 8 at 13:01
  • $\begingroup$ Thank you @ttnphns, I think I will use these metrics just to choose the correct number of clusters in a given dataset. $\endgroup$ – Jony Zambrano Feb 9 at 8:17

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.