Cluster Measurement Measures

Does Matlab provide any means to evaluate clustering methods? (compactness of compactness and cluster separation ....) Or is there any set of tools?

+6
source share
4 answers

Not in Matlab, but ELKI (Java) provides a dozen cluster quality parameters for evaluation.

+4
source

Matlab provides the Silhouette index, as well as the CVAP toolbar: a cluster validation analysis platform for Matlab. Which include the following confidence indicators:

  • Davis-Bouldin
  • Calinski-harabasz
  • Dunn Index
  • R-squared index
  • Hubert-Levin (C-index)
  • Krzhanowski Lai Index
  • Hartigan Index
  • Standard Deviation Index (RMSSTD)
  • Half-R-Square Index (SPR)
  • The distance between two clusters (CD) index
  • weighted cross-industry index
  • Homogeneity index
  • Split index

Please note that for your platform you may need a precompiled

+5
source

You can try Silhouette from the Statistical Tools toolbox.

See this documentation for an example.

+4
source

Keep in mind that Silhouette in Matlab has weird behavior for single clusters. He assigns a score of 1 for single numbers, when for me a more reasonable approach would be to give 0 for these clusters. In the Matlab implementation, if you specify the number of clusters as the number of objects, Silhouette will give you a rating of 1.

+3
source

Source: https://habr.com/ru/post/914420/


All Articles