Adding inter-trial SEEG power increased previous-key loss in eight fixed linear later-trial tests

We tested whether inter-trial SEEG power improved prediction of the prior response key. We used 1941 matched trials from eight records in OpenNeuro ds004473 version 1.0.2. Each model trained on early trials and tested on later trials from the same record. The model added eight bipolar pairs and three power bands to six task and timing variables. The mean log loss gain was -0.150 nats per trial. The descriptive 95% patient bootstrap interval was -0.188 to -0.114. Positive gain means lower loss. The model with SEEG had higher mean loss than the context model. The source field named correct identified the key in all 280 recognized sub-1 source rows. We kept this field in the fixed primary model and recorded one unavailable restricted test. The package has code, features, source records, predictions, figures, and 16 checked raw ranges. This offline test does not establish absence of neural history or test a live BCI.

stereoelectroencephalography · SEEG · response history · logistic regression · brain-computer interface