When using Decision Trees to do variable selection for neural networks, does anyone have suggestions on settings to use to get the best results? What would lead to more variables being selected, or more interaction effects being found?
Try setting "number of surrogate rules" to 1 and "Method" to largest in the properties panel of the tree node. These settings should allow for several variables to be picked by the tree and allows variables not explicitly picked by the tree (surrogates) to be used as well.