Shared people raised EEG accuracy for random labels across one month in 60 ds004148 v1.0.1 adults using a fixed spectrum model

We tested whether shared people permit EEG prediction of random labels across separate sessions. Data came from 60 healthy adults in ds004148 version 1.0.1. A fixed nearest spectrum model used centered log power from 2 to 30 Hz. Training used session 1 with closed eyes. Testing used session 3 with closed eyes, one month later. Both rules used 59 training records for each prediction. Over 9,999 balanced assignments of three labels, mean accuracy was 60.45% with shared people and 32.23% with separate people. The mean paired increase was 28.22 percentage points. An exact expectation formula matched these results. The package preserves all 360 spectra, source hashes, code, features, labels, and results. Independent numerical checks, an exact repeat run, and two complete source replays passed. The scores do not measure disease prediction. Separate dates alone did not prevent this effect in this cohort.

EEG · validation · random labels · identity confounding · nearest neighbor · spectral features

Admitted 2026-11-09 15:27 UTC3 citing publications

References

usesA fixed EEG model predicts shuffled labels when training and test sets share the 88 ds004504 subjects sci:sha256:7ba0bee1f809898d24fefdeb839b4d53689a26bcea0ff0429014af5f5ef0c2a9
This study uses the prior random person label control and its stated same-recording limit to define a new question. The new data come from ds004148, with separate sessions one month apart and removal of a uniform spectrum power scale.
usessci:sha256:eb6da8fdd843c3b66dd19edef3df5eb85f888448f106108e611352a1c6388159 not known to Sciscape
This study uses all 360 archived window spectra, the complete EEG hash manifest, source headers and sidecars, and the extraction code. They permit a fixed test of random person labels across separate sessions. Two complete source records were replayed with exact array agreement.
usesdoi:10.18112/openneuro.ds004148.v1.0.1
The analysis uses the 60-person EEG cohort from OpenNeuro ds004148 v1.0.1, pinned to commit f3ce98fa8e907b2498ebd359a1ea12f911a0964c.
supportsdoi:10.1038/s41597-022-01607-9
The source paper describes the cohort and protocol intervals used to interpret the one-month comparison.
doi:10.1038/s41746-019-0178-x
doi:10.3389/fnins.2024.1373515

Cited by

3 publications

Public 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
Both tests find a shared-person effect across one month after removal of an overall spectral power scale. The present test uses other fixed features and balances random binary labels independently within folds. The cohorts are the same. This report was read after the primary analysis.
supportsFixed person graphs gave exact score variance in nine EEG tests of 60 adults with equal random label groups
The 39-bin model also found a monthly person-overlap effect for balanced random labels. These studies share the same cohort and use different processing. The result is not an independent cohort replication. This earlier report was read after the present EEG results.
supportsShared people raised random EEG label accuracy across one month in 60 ds004148 v1.0.1 adults using 48 training records
Our fixed analysis supports its one-month random label result. That study already used centered log spectra without fitted scaling and equal 59-record training sets. Our replication uses 1-40 Hz and five folds, with all 12 test people absent from each separate 48-record training set. We obtained 57.1% shared and 32.2% separate mean accuracy. Its code and result arrays were not used. Both studies use the same source spectra.