Generating a Test Design

As a final step before actually building the model, you should take a moment to consider again how the model's results will be tested. There are two parts to generating a comprehensive test design:

• Describing the criteria for "goodness" of a model
• Defining the data on which these criteria will be tested

A model's goodness can be measured in several ways. For supervised models, such as C5.0 and C&R Tree, measurements of goodness typically estimate the error rate of a particular model. For unsupervised models, such as Kohonen cluster nets, measurements may include criteria such as ease of interpretation, deployment, or required processing time.

Remember, model building is an iterative process. This means that you will typically test the results of several models before deciding on the ones to use and deploy.