Predictive rule modeling

While predictive models may deliver all the predictive power you need, there are times when you may want to gain a little more control and insight into the inner workings of the model. In this case predictive rule (interactive) modeling can be used. With predictive rule models, you can manually create rules to segment your data, or automatically find segments with high or low concentrations of the value of your chosen field. For example, you might look for customers who gave a positive response to your campaign and then identify segments with a higher probability of responding.

  1. Returning to the Modeling tab, click Change Model and select Predictive rule model.
  2. Select bank response data for the data source, and select Response as the target field.
  3. Specify 1 as the response sought, indicating you want to look for customers who gave a positive response.
  4. Select Start Build.
  5. Select Grow Model to identify segments with a higher probability of responding. The three best rules that identify segments with the highest probability to respond are created and displayed.
  6. Click the Include/Exclude icons to select whether or not to include or exclude customers selected by each rule. For example, the second and third rule segments show response rates just over 8%, which is still higher than the overall test campaign rate of 3.7%, but not nearly as high as the rate of 15.65% for the first segment. To see how including or excluding these two rules may impact your overall profit, you can run the profit simulation both ways and compare the result, as follows:
  7. With all three rule segments set to Include, click Evaluate.
  8. Select Gains chart and Profit chart for the graph types to build.
  9. Select Simulate maximum profit and specify 3000 for the population, 2.25 for cost, and 60 for revenue. These are the same specifications used when evaluating the predictive model. The response sought is automatically completed.
  10. Click Run and, when the graphs are built, go to the Simulate tab.

    The maximum simulated profit for this model is $3178.90, a bit lower than for the automated model in this case. Optionally, you can close the Evaluate dialog, change the setting for the second and third rule segments to Exclude, and repeat the profit simulation to see how this changes the result. The real benefit of predictive rule (interactive) modeling is the increased control and insight it affords.

  11. If desired, save the project as bank_predictive_rule_model.str.