VERIFIED SOLUTION i
X

How is the “Separation” calculated in the clustering algorithm implemented in Spectrum Miner?

UPDATED: September 4, 2017


The maths formulae for Clustering is covered in the product Help documenatation, (press F1 in Miner),
under the 'Building clustering models for segmentation and predictive modeling - About clustering models in Portrait Miner' section.

The separation for a record is '1 - d1/d2', where d1 is the distance between the record and the nearest cluster center, and d2 is the distance between the record and the second-nearest center. Values range between 0 and 1; a higher value indicates that the record has a stronger attachment to it's assigned cluster.

The separation for a cluster is the minimum separation for a record in the cluster. A well separated cluster has a value close to 1.

The separation for a clustering model is the minimum spearation for a record in the dataset. A good model has a value close to 1.

 

Environment Details

Product Feature: Cluster Builder

 

Downloads

  • No Downloads