Adding alpha reactivity change did not lower observed KSS error at one month in 56 ds004148 adults under linear models with equal score spacing and each test person excluded

We tested whether EEG alpha reactivity change improved sleepiness estimates in 56 adults from OpenNeuro ds004148 version 1.0.1. The target was KSS at session 3, one month after session 1. The history model used the first KSS score. The EEG model added the change in the closed-eye to open-eye 8-13 Hz power ratio, in dB. Each ordinary linear fit excluded the test person. We assumed equal spacing of KSS levels. History MSE was 2.3926; EEG MSE was 2.4094 squared levels. History MSE minus EEG MSE was -0.0168. Its approximate 95% interval from 10,000 refitted participant bootstrap samples was -0.4329 to +0.4209. Each fit removed all copies of the test person. The EEG term did not lower the observed error. The interval permits both benefit and harm; it does not prove equal performance. The EEG feature was chosen from earlier work in this same cohort. This internal test is not independent cohort validation and requires current EEG. The package preserves 360 spectra, source hashes, all predictions, bootstrap draws, code, a local plan, and instructions. Four complete source EEG records rebuilt exactly. Two analysis runs produced identical numerical files.

EEG · alpha reactivity · sleepiness · Karolinska Sleepiness Scale · leave-one-person-out · bootstrap · prediction error