Cross Validated is a question and answer site for people interested in statistics, machine learning, data analysis, data mining, and data visualization. It only takes a minute to sign up.
Sign up to join this community
I created a PLS model with three dependent variables using mdatools. Variable A gets the best results when using two components. However for variables B and C it would be better to use four components.
Does it make sense to use different component counts for different variables or is this not allowed in PLS?
And how is this handled in PLS-DA? If a PLS-DA model has one variable with multiple groups. Is it allowed to use different component counts for different groups?
The first case where there are 3 dependent variables, it is in fact the usual way to have different number of components for each of them. This is also called PLS1. There is, however, another method where you can model multiple dependent variables and is called PLS2. PLS1 usually provides better prediction performance, but for some cases PLS2 can be advantageous i.e. when the dependent variables have collinearity. If unsure, you can test both.
PLS-DA can be defined as PLS regression where each class is treated as a dependent variables.
In single class(group) case, only for PLS-DA, PLS1 is equivalent to PLS2. For more than 2 classes, PLS-DA is basically applying PLS2 regression. For detailed explanation see here.
