Specify the interactions for Fit Model with TreeNet® Classification

Predictive Analytics Module > TreeNet® Classification > Fit Model > Interaction

This command is available with the Predictive Analytics Module. Click here for more information about how to activate the module.

Select the predictor interaction options.

Predictor interaction control
Choose which interactions to allow during tree growth.
  • Allow all order interactions : Allow all predictor interactions.
  • Do not allow any interactions (Additive model) : Do not allow predictor interactions. In this case, Minitab uses the additive model and builds each tree using only one variable.
  • Select specific predictor interactions up to order: Allow predictor interactions up to the order that you specify. Order specifies the number of variables that can be splitters on the same branch of a tree. For example, an order of 2 indicates that a branch that splits on variable X1 can split on the X1 variable and 1 other variable. Different branches can split on different sets of variables. Thus, a tree that allows interactions of order 2 can split on more than 2 variables, so long as no single branch contains more than 2 variables.

    In Predictors, enter the columns that contain the predictors that you want to allow for the interactions. If you specify no specific predictors, then the analysis considers interactions among all of the predictors.

Display two-way interaction strength table
Display the percentage of the total square error for the strongest two-way interactions. This table is only available with a binary response.
Strength as percentage of total squared error
Select to display the percentage of the total squared error for the main effects of two predictors and their interaction.
Strength as percentage of squared error due to specific pair only
Select to display the percentage of squared error of predictor pairs and the pair interaction.
For example, let X1 and X2 be two predictors. The percentage is the squared error for the interaction divided by the squared error for the X1*X2 interaction and the main effects of X1 and X2.