Splsda error rate


I have a score(18method) for a sample of 760 people. (row=760, col=18) Samples are divided into yes and no for diabetes (yes = 29).
I did spls-da but the error rate is high and it shows comp=1. I expect score(18method) to be divided into 2 comps. But there is no result. How can I do it?

Thank you.

hi @seung,

First, note that these tuning criteria are based on one-sided t-test of the classification error rate. You can also choose your own parameters if you feel that you would like to add another component when you run your final model (you plot seems to show that ncomp = 2 could also be appropriate).

The PCA seems to show that the groups of samples (that are quite unbalanced) will be challenging to discriminate, I would definitely look at the balanced error rate to guide the parameters choice.