Databases Reference
In-Depth Information
In the dialog that opens, aggregate on ClusterName; check Purchases for
aggregation; check “Avg”.
Name the aggregation “AvgPurchases”; click “OK”.
View the newly created AvgPurchases.csv dataset in a parallel plot.
Which cluster of customers has the highest average purchase amount? Which
cluster has the lowest? Based on the results, which magazine (or magazines)
should receive the heaviest advertising during the upcoming promotion?
Notice that via aggregation of clusters we have achieved the last objective of
cluster analysis, reducing the 5,000 observations down to nine.
Summary
Cluster analysis is the process of grouping similar observations. It has three
primary purposes:
to identify and isolate sub-populations
to improve understanding of the dataset relationships
to aggregate observations for reduction purposes.
The result of a cluster analysis is called a clustering. The quality of a
clustering is defined with respect to its cohesiveness within clusters and
separation between clusters.
VisMiner implements a three-dimensional version of the self-organizing map
(SOM) algorithm for cluster analysis. It provides for synchronized visualiza-
tions of the overall clustering, individual clusters, and clusters selected for
comparison. It also supports extraction of clusters for use by other data mining
algorithms.
