Predicting Sales

Now use Sales in thousands [sales] as the target, to get an idea of what kind of sales figures you might expect once these prototypes are established models.

  1. Recall the Nearest Neighbor Analysis dialog and click the Variables tab.
    Figure 1. Variables tab
    Nearest Neighbor Analysis, Variables tab with Sales in thousands selected as the target
  2. Deselect Vehicle type [type] and select Sales in thousands [sales] as the target.
  3. Click the Features tab.
  4. Select Perform feature selection.
  5. In the Stopping Criterion group, type 5 as the number of features to select.
  6. Click the Partitions tab.

    When feature selection is requested in addition to k selection, cross-validation is no longer used because of performance concerns. Instead, a set of features are selected for each value of k, and then the value of k (and attendant feature set) that minimizes the error in predicting the target is chosen.

    This also means there is no need to set the seed for the random number generator.

  7. Deselect Set seed for Mersenne Twister.
  8. Click the Save tab.
  9. Select Predicted value or category.
  10. Click OK.

Next