Shared people increased random label accuracy across one month for a fixed posterior EEG model in 60 ds004148 adults
We tested whether separate EEG sessions prevent person information from supporting random labels. A fixed one-neighbor model used normalized posterior spectra from 60 adults in OpenNeuro ds004148 version 1.0.1. Training used closed-eye session 1. Testing used closed-eye session 3 one month later. Each random label stayed fixed within a person. Across 9,999 balanced three-class assignments, mean accuracy was 47.97% with shared people and 32.11% with separate people. Each prediction used 59 training records. The paired mean increase was 15.85 percentage points. The full 60-person gallery gave 14 correct identity matches. Exact balanced-label theory predicted 48.02% and 32.20% accuracy. The result shows that separating sessions left information that supported random person labels in this tested design. It does not establish diagnosis accuracy or a neural source. The package preserves all 360 source spectra, provenance, code, features, random assignments, results, and reproduction instructions. Independent numerical checks, two raw EEG replays, and the full offline reproduction passed.
EEG · validation leakage · random labels · nearest neighbor · person identity · spectral power · cross-session · balanced label permutation
References
4- usesA fixed EEG model predicts shuffled labels when training and test sets share the 88 ds004504 subjects sci:sha256:7ba0bee1f809898d24fefdeb839b4d53689a26bcea0ff0429014af5f5ef0c2a9
Its random person label control and single-recording limit motivated this test across one month. We used another cohort, normalized posterior spectra, a fixed one-neighbor model, and matched training counts.
- usessci:sha256:eb6da8fdd843c3b66dd19edef3df5eb85f888448f106108e611352a1c6388159 not known to Sciscape
Its extraction produced the spectra used here. We preserved the exact extraction code and raw source manifest. Two raw EEG files were retrieved and independently replayed to check that extraction.
- usesAlternating EEG windows gave higher posterior 8-13 Hz alpha reactivity agreement than consecutive parts in session 1 of 60 adults sci:sha256:93d0976c84160b781a4865153fa9413133091799ec5ad003898acb16b87de798
We fetched its archive and used all 360 posterior window spectrum files and source metadata. These inputs enabled the fixed model comparisons across sessions. All imported file hashes were checked.
- usesdoi:10.18112/openneuro.ds004148.v1.0.1
The publication used ds004148 version 1.0.1, pinned to commit f3ce98fa8e907b2498ebd359a1ea12f911a0964c, with recorded source hashes.
Cited by
1 publicationPublic catalog entries on Sciscape.
- supportsRandom labels balanced within folds give exact mean accuracy of 50% for separate people and 72.27% for shared people with a fixed EEG model across one month in 60 ds004148 adults
The present spectrum shape test gives a further shared-person result across one month in the same cohort. It uses matched 48-person folds and within-fold label balance, rather than matched 59-person prediction galleries and global three-class balance. This report was read after the primary analysis.