PLS-DA

[#1]

Hi

I'm using JMP pro. I have a set of variables that I would like to use to predict 2 groups (i.e. grp A and grp B). Some variables are highly correlated, hence my interest in trying PLS-DA. I can run the model and see some nice plots that make sense e.g. coefficient plots identify variables with high values that match what I would anticipate from domain knowledge.

I would like to get out the predictions from the model, and, if possible, a ROC curve. Is this possible? It doesn't seem to matter how I cross validation the model (e.g. Leave one out, k fold etc), I don't seem to be able to find the group predictions or ROC, but I'd have thought it possible to.... I think I must be missing something obvious but I can't seem to figure it out.

Thanks

[#2 · Accepted solution]

You can get predictions from your model in a couple of ways.

One way is to go to the red triangle popup menu for your model and select Save Columns > Save Prediction Formula. This will add a column to your data table that provides predictions for all observations. Adding a new observation and typing in the values will give you a prediction.

The other way (probably easier) is to go to that same red triangle popup and choose Profiler. That will give you an interactive picture of the model. Dragging any of the red lines for the factors will allow you specify the factors and the predictions will be updated on the graph.

As for ROC, that does not seem to be available in PLS-DA. I guess it is a platform designed for more than a binary response. I'm not sure how to get it unless you wanted to build it yourself.

[#3]

Thats very helpful, thanks.