The ‘Output’ tab in the ‘Model Build Settings’ dialog provides additional options for both controlling the model building process and determining how results are stored and reported.
- Prediction Format - Use this option to set the format in which predictions from the model are output. The options available are different depending upon the model type.
- Create Final Models - This drop-down determines whether PREDICT will build final models for all iterations or just the single "best" iteration.
- Automatically Build Just the Best Model - Use this option to have PREDICT build and report a final model for a single iteration - the one that performs the best according to the primary performance metric (see below). All model iterations will still be validated and have performance metrics reported for comparison. This option determines outcome based solely upon quantitative criteria, identifying final coefficients and other output that is unnecessary.
- Build All Models for Review - With this option set, PREDICT will not only compute performance metrics for all iterations of the model, but will also build a final model for each iteration. Both quantitative performance metrics (such as lift and error rates) and qualitative criteria (such as model coefficients and structure) can be reviewed, prior to deciding upon an iteration to implement. This option will require the longest run times, but it also gives the user complete control of model implementation.
- None: Performance Metrics Only - Use this option to choose for PREDICT to build no final models. With this option selected, PREDICT will only compute comparative performance metrics. This option will require the shortest run times, but it will create no final implementable model.
- Primary Performance Metric - Select the performance metric that will be used to compare and rank the order of different model iterations. Relevant options for the selected model type will appear. The main use of this option is when Automatically Build the Best Model is selected above. The "Best" model will be decided based on the metric selected here.
- Prediction Lower/Upper Bounds – Checkboxes and bound setting are used to bound the model with lower and/or upper bounds. The bound setting input boxes will only be available if the relevant checkbox is checked.
- Set Population Average for Target - Check this box if the modeling dataset has a different target variable average than the full population. This is useful with stratified sampling, to construct the modeling dataset. After checking this box, enter the true population average in the data entry box that becomes available. Even if the population average is different than the modeling dataset, it is not necessary to use these settings. The effect will be that predictions resulting from the model will be scaled so that they align more closely to the population average, rather than the modeling dataset average.