Beta episode occupancy raised mean recall forecast error for 13 SEEG patients under assumed source trial order

We tested whether early beta episode occupancy adds value to later recall-error forecasts in public OpenNeuro ds006136 v1.0.1. All 13 patients, 14 sessions, and 143 released channels were used. Increasing source trial IDs were assumed to give task order. The first 65% of matched trials trained each session model. The later 751 trials were tested. The detector used 12 to 30 Hz activity above a training threshold for at least 150 ms. Occupancy is the fraction of time in these episodes. Thresholds came from training data. Forward signal features ended at 4 s. The control included history, signal availability, inherited power, and power from the same filter and windows. Adding eight occupancy fields raised the mean Brier error across patients from 0.179688 to 0.183474. The mean gain was -0.003786; the 95% patient bootstrap interval was [-0.007071, -0.001141]. This result applies to the fixed detector and regional median design. It does not establish equivalence or live BCI performance. The source was processed and screened offline. The package preserves features, predictions, fitted models, three figures, source hashes, code, and a checked raw-data retrieval and rebuild recipe.

beta episodes · SEEG · recall-error forecasting · Brier error · working memory · signal processing · brain-computer interface