Juletip #7 – Knowledge is power, interpretability is king!

[Mike_j]

Have you ever dreamt about predicting the future?

Yes?

Then you have come to the right place - Let us create a predictive model.

Wait wait wait, someone said something?

“Predictive models are black boxes that cannot be interpreted”

Good news! Christmas is coming, and Santa is bringing a present to you - Model interpretability. And it even works on imputed and transformed variables!

Yes, that’s right! There are capabilities built into SAS that can assist in interpreting models. The best thing? I will show you how to use it.

My tool of choice for this purpose is SAS Model Studio on SAS Viya. If you haven't used it before, it's a drag-and-drop interface, where you don't have to code anything, unless you really want to. One of my favorite tools for creating models quick and easy.

Step 1) create a SAS Model Studio project:

 As you can see I've opted for the famous HMEQ dataset. Mostly everyone have seen that before.

Step 2) Select a target variable. Then add an impute node, some models and enable the model interpretability of your choice:

 Or do like me, and only select your favorite model!

Step 3) Run the pipeline and open the output of a model node:

Note that the imputed variables are used for interpreting the model. But that’s not what we need to convince the business folks to trust our model. Let's fix that!

Step 4) Add an ensemble node to a model and enable the same model interpretability:

The ensemble node can also be used to connect multiple models, and then use i.e. the average of the model output as scores. In this case, we use it only for interpretation purposes. 

Step 5) Run the flow and have a look at the model output:

Voila! The original variables are now used to interpret the model.

more information about Model Studio and the interpretations can be found here: SAS Help Center: Model Interpretability

I wish you a merry Christmas and a Happy New Year!