Using classification trees to explore the predictions of a Neural Network

Neural Nets have the reputation of being a black box technique; that is, that they are not highly revelatory of the reasoning behind their predictions. Compared to other techniques, information regarding what variables played the most important role in the model is fairly thin. It would be an exaggeration to say, however, that the Neural Net algorithm in Modeler provides no information; it does. Neural Nets are sometimes strong performers, and when they are the top performer they might be (and should be) a tempting option for Deployment. Is it possible to use other techniques to get a deeper insight into what the Neural Net has done behind the scenes? It is possible and one method for doing so is the subject of this recipe. We will be using CHAID to explore Neural Net predictions.

Getting ready

We will start with the Look Inside NN.str stream that uses the TELE CHURN MERGED data set.

How to do it...

To use classification...