Model selection (GLE models)

Use model selection or regularization. To activate the controls on this pane, select this check box.

Method. Select the method of model selection or (if using Ridge) the regularization to be used. You can choose from the following options:

Lasso. Also known as L1 Regularization, this method is faster than Forward Stepwise if there are a large number of predictors. This method prevents overfitting by shrinking (that is, by imposing a penalty) on the parameters. It can shrink some parameters to zero, performing a variable selection lasso.
Ridge. Also known as L2 Regularization, this method prevents overfitting by shrinking (that is, by imposing a penalty) on the parameters. It shrinks all the parameters by the same proportions but eliminates none and is not a variable selection method.
Elastic Net. Also known as L1 + L2 Regularization, this method prevents overfitting by shrinking (that is, by imposing a penalty) on the parameters. It can shrink some parameters to zero, performing variable selection.
Forward Stepwise. This method starts with no effects in the model and adds or removes effects one step at a time until no more can be added or removed, according to the stepwise criteria.

Automatically detect two-way interactions. To automatically detect two-way interactions, select this option. Note that GLE only detects the interaction of two categorical variables and squared of continuous variables. It doesn't detect the interaction of two continuous variables and the interaction of categorical and numerical variables.

Penalty Parameters: These options are only available if you select either the Lasso or Elastic Net Method.; Automatically select penalty parameters. If you are unsure what parameter penalties to set, select this check box and the node identifies and applies the penalties.; Lasso penalty parameter. Enter the penalty parameter to be used by the Lasso model selection Method.; Elastic net penalty parameter 1. Enter the L1 penalty parameter to be used by the Elastic Net model selection Method.; Elastic net penalty parameter 2. Enter the L2 penalty parameter to be used by the Elastic Net model selection Method.

Forward Stepwise: These options are only available if you select the Forward Stepwise Method.; Include effects with p-value no less than. Specify the minimum probability value that effects can have to be included in the calculation.; Remove effects with p-value greater than. Specify the maximum probability value that effects can have to be included in the calculation.; Customize maximum number of effects in the final model. To activate the Maximum number of effects option, select this check box.; Maximum number of effects. Specify the maximum number of effects when using the forward stepwise building method. To optimize performance, 10 is the highest supported number.; Customize maximum number of steps. To activate the Maximum number of steps option, select this check box.; Maximum number of steps. Specify the maximum number of steps when using the forward stepwise building method.