Print

Print
Hi Isaac,

PCA is done before ICA and not all principal components are retained. Those that explain little of the variance, i.e., those with tiny eigenvalues, are removed. The number represents the amount of variance from the original data with which ICA starts.

I'm not sure there is a specific value that would be ideal but I think most datasets would retain between 50 and 70% with 20 PCs.

All the best,

Anderson


On 4 August 2016 at 00:13, SUBSCRIBE FSL Isaac <[log in to unmask]> wrote:
Dear FSL experts,

My question is regarding the percent of variability retained during PCA by melodic. Could someone please explain to me what this number means/represents and whether there is a certain percentage we should be aiming to retain.

Many thanks,
Isaac