Feature & Model Selection; Regression - Advanced Tab

Select the Advanced tab of the Feature & Model Selection; Regression dialog box to access options to select the specific types of models you want to fit ("try"), and the results you want to compute.

Models

In this group box, select the types of models you want to fit, to determine the variables and methods that are most important for the prediction of the dependent variable(s) of interest. In general, Process Optimization will compute the predictor statistics by the specified method, and then rank the predictors based on the method-specific measure of predictor importance. Refer to Feature and Method Selection Computational Details for more information; see also General Regression Models (GRM) and General Discriminant Analysis (GDA)], classification and regression trees [see General Classification and Regression Trees (GC&RT)], Multivariate Adaptive Regression Splines (MARSplines), boosted tree models (see Boosted Trees; this is an implementation of stochastic gradient boosting), and various neural network architectures. The default model that will always be computed is the linear model.

Interactive trees (C&RT)

Select this check box to compute a regression tree using the methods described in the documentation for the General Classification and Regression Trees (GC&RT) module.

Stochastic gradient boosting trees

Select this check box to compute a boosted tree using the methods described in the documentation for the Boosted Trees module. Note that the computations for the boosted trees method are complex, and the overall speed of computations may be noticeably slower when selecting this option.

MARSplines

Select this check box to compute a multivariate adaptive regression splines (MARSplines) model using the methods described in the documentation for the Multivariate Adaptive Regression Splines (MARSplines) module.

Neural networks (various architectures)

Select this check box to compute a best neural network. Note that the computations for the neural networks are complex, and in this particular case, the program may compute and evaluate a large number of networks and network architectures to find a best model; the overall speed of computations may be noticeably slower when selecting this option.

Reports

Use the options in this group box to determine whether to compute predictor (importance) rankings and method rankings (which method seems to work best for the current data), and the detail of the results that are to be reported.

Predictor and method ranking

Select this check box to compute importance rankings for the predictors and the methods. Summary results spreadsheets will be created that contain rankings for the predictor variables for each selected method (as selected in the Models group box), and a summary for each method (squared Pearson correlation value for the correlation of the predicted values with the observed values, and rankings of those correlations). See also, Feature and Method Selection Computational Details for more information.

Summary for each method

Select this check box to create spreadsheets with detailed results for each method, i.e., with the results spreadsheets that were used by Process Optimization to rank the importance of the predictors. Refer to the documentation suggested for each method (see above) for details regarding these statistics; see also Feature and Method Selection Computational Details.