Learn to build a Clustering Model in Portrait Miner

Product Feature: Cluster Builder

Clustering is an undirected (objective-free) modeling technique, used to segment a population into clusters of similar records based on the observed values of one or more analysis fields.
  • Each cluster in a clustering model has a center, which can be a record containing a value for each analysis candidate (except that a categorical field is treated as though it were a set of indicator fields).
  • Each record in the focus belongs to the cluster whose center is closest to the record, according to a normalized sum-of-squares distance metric.
Steps to build a clustering model:
  • Optionally interpret some fields as analysis candidates.
  • Optionally interpret some fields as analysis candidates. If there are fields with an analysis candidate interpretation, the Cluster Builder only considers those fields when creating clustering models.
  • Choose Tools > Cluster Builder Clustering_Model.png or click on  on the toolbar.
  • If necessary, change the default build options.
  • Choose Clustering > Build to End or click on Play_Button.png on the toolbar, to build a clustering model. The Cluster Builder continues until it satisfies one or more of the termination criteria for a clustering model.
  • Optionally, to create a further iteration of the clustering model, choose Clustering > Do Next Iteration or click on Forward_Button.png on the toolbar. (If the clustering model has converged or the build parameters have changed, this option is not available.)
  • The Cluster Builder is an undirected model-building tool. It therefore takes no account of any objective that you have set.
  • The panes on the left of the Cluster Builder window show included fields (analysis candidates) and excluded fields (other fields in the focus), with fields that appear in the current clustering model shown in boldface. A field can be interpreted as an analysis candidate by moving it from the lower pane to the upper pane (and vice versa), by using the arrow buttons between the panes.
  • After clustering model is started, the Cluster Builder prompts us to interpret any string fields that do not have a categorical interpretation, so that they can be binned appropriately. Either choose the default categorical interpretation or use a saved categorical hierarchy. If Cancel is selected, when prompted to choose an interpretation for a string field, the Cluster Builder ignores that analysis candidate, along with any remaining string fields without a categorical interpretation.
  • Statistics for clustering models appear in the Clustering Results pane. These include measures of model quality (total ESS and separation) as well as model-building parameters.
  • Statistics for the clusters in the last model to be built appear in the Cluster Details pane (except for the outlier and null clusters). These include measures of cluster quality (ESS, separation, and radius) as well as the number of records in each cluster and the field values that describe each cluster's center.
  • To clear the history of clustering models produced along with the current model details, choose Clustering > Clear or click on Cancel_Button.png on the toolbar.

For detailed information about the clustering model, click on Help or press on F1 and search for 'Clustering' in the Portrait Miner Help window.

UPDATED:  August 18, 2017