Normalized personal 2–30 Hz EEG spectra reduced after-task squared Hellinger loss in 60 ds005385 adults when mixed equally with peer age-group spectra

We tested a fixed EEG prediction rule in 60 adults from OpenNeuro ds005385 version 1.0.3. We used normalized posterior spectra from 2 through 30 Hz. The prediction mixed the person's before-task spectrum equally with the after-task mean of 11 age-group peers. The target after-task record was excluded from its own calibration. Mean squared Hellinger loss fell from 0.08550 to 0.04188, a 51.0% reduction. The gain was 0.04362; its 95% refit-bootstrap interval was 0.03732 to 0.06206. Loss was lower for 56 of 60 people. Correct-person records also had lower loss than other records from the same age group. All planned checks gave positive gain, including removal of 7–14 Hz. The package preserves all 120 input spectra, source hashes, code, predictions, bootstrap draws, and reproduction instructions. Scalar checks, a raw EEG pair replay, and an identical complete repeat passed. This is an exploratory retrospective test with peer after-task calibration. Bootstrap population coverage is not established. The result does not identify a neural or task cause or establish transfer to a new cohort.

EEG · spectral prediction · Hellinger distance · personal reference · OpenNeuro ds005385