Early linear forecasts showed no clear gain over earlier means at low-power gates in nine SEEG rest records at 70 to below 110 Hz

We tested whether the current low-power value improved a forecast of the next 0.5 s SEEG window beyond an earlier mean. The public data were nine rest records and 539 good adjacent contact pairs from OpenNeuro ds003688 version 1.0.7. Each pair used 88 s of earlier data for its mean, gate, and linear fit. The main band was 70 to below 110 Hz. A fixed later test gave normalized mean squared error of 1.033 for the earlier mean and 0.997 for the linear forecast. The paired difference was -0.0365. Its 95% person bootstrap interval was -0.1207 to +0.0127. The percent change was -3.53%, with an interval of -12.00% to +1.21%. Only three of nine people had lower error with the linear forecast. The current-value forecast had normalized error of 2.894 and mean error of 2.94 dB at the same gates. The earlier mean left a mean error of -0.20 dB. These records had no task stimulation. The data support an earlier-mean comparison for selected power changes, but do not establish a linear forecast gain or a stimulation effect. This is a new forecast test in an existing cohort. The package includes a local plan, compact inputs, source identities, fit and forecast arrays, code, figures, and successful checks against all nine complete raw files.

SEEG · intracranial EEG · power forecast · forecast error · low-power gate · human neuroscience · resting state