MicroStrategy ONE

Segmentation Example (Using Cluster Analysis)

Clustering attempts to segment items so that members of one group are more similar to each other than to members of other groups. These items are most often customers, but can also be products, patients, prescriptions, phone calls, emails, or any other item relevant to the enterprise. Clustering algorithms do the segmentation by analyzing the characteristics of the items and finding the best ways to group them by similarities.

You will use cluster analysis in the following example to group your customers into exactly five segments based on demographics and psychographics.

Each of the following will be used as inputs into the training metric:

  • Age Range
  • Education
  • Gender
  • Housing Type
  • Marital Status

As before, you must create a metric for each attribute form, which can then be used to create a predictive metric. Examples are:

	Max([Customer Age Range]@ID) {Customer}
	Max([Customer Gender]@DESC) {Customer}
	Max([Housing Type]@DESC){Customer}

The example Tutorial project includes reports, metrics, and other objects created for this segmentation example (search the project for "Cluster Analysis"). You can use the objects in the Tutorial project to step through the example and determine how it can be applied to your reporting environment.

Use the Training Metric Wizard to design a training metric, following the procedure below.

To Create a Training Metric for Cluster Analysis

  1. In MicroStrategy Developer, choose Tools > Training Metric Wizard.

    To skip the Introduction page when creating training metrics in the future, select the Don't show this message next time check box.

  1. Click Next.
  2. Select Cluster as the type of analysis.
  3. Specify exactly 5 clusters for Model Specifications.
  4. Click Next.
  5. Add the Age Range, Education, Gender, Marital Status, and Housing Type metrics to the list of Independent Metrics.
  6. Click Next.
  7. Select the Automatically create on report execution check box.
  8. Select Predicted Value.
  9. Click Finish to save and create the metric. You can now include the metric in a training metric to create a predictive metric, as described in Creating a Predictive Model Using MicroStrategy.
  10. Often, training reports do not require a large number of rows of data to formulate an acceptable result. You will need to reduce the number of rows in your training report by sampling the data.

    For this example, your sample will include a random set of 20% of the customers. Create a filter to define this random set. Use the filter on a new report with the Customer attribute and the training metric created above.

  11. Execute the report.

    A predictive metric is created in the folder you specified in the Training Metric Wizard. The default location is the My Objects folder.

    Add the predictive metric to a new report with Customer, Age Range, Gender, Education, Housing Type, and Marital Status to see the relationships between segments.

    A custom group can be created based on each segment to further segregate the groupings.

    In some cases, the ideal number of clusters is not known before performing an analysis. To overcome this obstacle, Data Mining Services offers a feature to determine the optimal number of clusters while training the cluster model. Retrain the model above, allowing Data Mining Services to determine how many clusters to create, following the procedure below.

To Retrain the Model to Automatically Determine the Number of Clusters

  1. Double-click the existing training metric to open the Training Metric Wizard.
  2. Click Next.
  3. Specify the maximum number of clusters as 10 in Model Specifications.
  4. Click Next.
  5. Click Next.
  6. Rename the predictive metric to prevent overwriting the existing cluster predictive metric.
  7. Save the training metric with a new name, specifying that it is optimal.
  8. Add the optimal training metric to a new report that contains the Customer attribute. Filter the report on a random 20% sample of customers.
  9. Execute the report.

Notice that Data Mining Services creates a predictive model with only two clusters, the optimal number of clusters less than the maximum allowed. Recall that the maximum specified in the Training Metric Wizard was 10.

A predictive metric is created in the folder you specified in the Training Metric Wizard. The default location is the My Objects folder.