PLSDA tuning question

1

Hi there,

I am wondering what the “right” thing is to do here. I am running an (s)PLSDA so the first step was to tune for the number of components. 2 components were suggested according to max distance. Then, when I tune to select metabolites, 1 component only was suggested. Should I go with 1 or with 2?

I’ve had a look at this post, and I’m wondering if I should run both models and choose based on the results of perf.assess() on the final model?

Thank you in advance for your time!

Best wishes,
Evelyn

2

Hi @windsnowflake,

Yes I think running both models and then assessing the final performance would be a good way to decide which number of components to use. This is assuming that your overall goal is to create the best performing model possible. If instead your goal is to explore the data and/or select important variables, you could also do this with 2 components even if its not the best performance because this allows you to create plots which can be very informative.

Hope that helps,
Eva