We show that the experimental design of neuroimaging studies may cause subclass-specific structures in EEG and fMRI data.
We present RSLDA, a novel method that allows to exploit subclass structure in a classification task.
We show that RSLDA outperforms a gold-standard classification method for both, EEG and fMRI data.
RSLDA outputs regularization profiles that allow to interpret the underlying subclass structure in the data.
Regularization parameters are estimated with Multi-Target Shrinkage, enabling a highly efficient method.