Clock time weights increased mean SEEG forecast error at 20–500 ms with matched decay rates in eight sessions

Clock time weights increased mean SEEG waveform forecast error by 0.944% relative to pulse order weights. The patient bootstrap 95% interval was 0.109% to 2.485%. The test used public OpenNeuro ds004696 data from eight patient sessions. It evaluated the final five responses at 86 sites, with 886 bipolar site-readout pairs. Both methods used all earlier retained responses. The clock half-life equaled five times the median initial pulse interval. The order half-life was five pulses. Each patient had a positive mean error increase. An equal mean of all past responses gave lower mean error than either decay rule. Only eight of 955 retained gaps exceeded 30 seconds. The result applies to these sessions and decay rules. A proof shows that normalized exponential weights with a fixed half-life cannot change a forecast during a pause without a new observation. The package includes three figures, compact data, a streaming mean tool with separate age status, source hashes, code, detailed results, and repeat instructions. Two full public source derivatives were rebuilt and matched the inherited products exactly.

SEEG · forecasting · electrical stimulation · exponential weighting · brain-computer interfaces · neuroscience