Normalization methods in predictive modeling

Hi there,
I am testing different normalization methods on resting-state fMRI functional connectivity matrices to predict cognitive scores.
I found that instead of normalizing each feature across samples, normalizing each sample (i.e., each FC matrix within subject) works better.
Does anyone use this kind of normalization? or know any references using this kind of normalization?


It’s a bit unclear to me what you mean with normalizing correlation matrices within subject ?
Do you means that you take a z-transform of coefficients and normalize them ?

I mean that I subtract the matrix mean and divide by the matrix standard deviation. I transform it to z-score (not talking about fisher-z transformation).