SQL Server Data Mining—Looking at the Clusters Created

Microsoft .NET Framework, ASP.NET, Visual C# (CSharp, C Sharp, C-Sharp) Developer Training, Visual Studio

Looking at the Clusters Created

The Mining Model Viewer tab in the model designer enables you to view the model that has been processed. Each algorithm produces a different type of model, so there are specific viewers for each model. The initial view for clusters is the Cluster Diagram, which shows all the clusters in the model with lines connecting them. Each cluster is positioned closer to other similar clusters, and the darkness of the line connecting two clusters shows the level of similarity. The shading of each cluster by default is related to the population of the cluster (that is, how many customers it contains—the darker clusters have the most customers).

For our Customer Internet Segmentation model, we can see ten clusters named Cluster 1 through Cluster 10. Each cluster represents a group of customers with similar attributes, such as customers who are fairly new to our Internet site and have not made a lot of purchases yet. Our task at this stage is to understand the kinds of customers in each cluster and hopefully come up with some more meaningful names for the clusters.

We can start by using the Cluster Diagram's shading variable and state parameters to look at each attribute and see which clusters contain the most customers with the selected attribute. For example, if I select Sales Amount > 1275 in Figure 10-6, I can see that Cluster 5 and Cluster 8 contain the most customers who have total sales of more than $1,275, as shown in Figure 10-6.

Figure 10-6 Cluster diagram

You can use the cluster diagram to help you comprehend each cluster by looking at one variable at a time. To really understand and compare the composition of clusters (that is, what types of customers are in each group), you need to use the Cluster Profiles and Cluster Discrimination views. We can see in the diagram that Cluster 1 contains a fairly high percentage of customers with high sales and is arranged near to Cluster 2 and Cluster 6, but we need more complete information to be able to assign a meaningful name to these clusters.