Mean normalized test error increased with a previous trial theta input in 12 SEEG patients at penalty 10

We tested whether theta power from the previous stored trial improves prediction of the next trial's hippocampal fixation power. We used OpenNeuro ds004752 version 1.0.1. Fixed models trained in the first half of numbered sessions and tested in the remaining sessions. The baseline used task history and mean power over five past trials. At penalty 10, the added input increased mean normalized test error from 1.407844 to 1.432007 in 12 patients and 996 test rows. The mean gain was -0.024163. Its 95 percent patient interval was -0.096260 to 0.027458. This test did not establish an improvement. Penalty and reference changes affected the result. The package supplies public features, source pins, code, all fixed settings, fitted models, predictions, and repeat instructions. The tests do not establish a neural mechanism or clinical benefit.

hippocampus · theta power · SEEG · intracranial EEG · sequential prediction · ridge regression · held-out sessions · verbal working memory