Validating rationale of group-level component analysis based on estimating number of sources in EEG through model order selection