Selection by weights and score patterns gave mean Brier errors of 0.198 and 0.194 for four SEEG pairs in later trials from eight recordings

We compared three training pair rules in OpenNeuro ds004473 version 1.0.2. Each rule retained four of eight candidate SEEG pairs. Models learned from early trials and tested on later trials in eight recordings from people with epilepsy. Mean Brier error was 0.1983 for the weight rule and 0.1938 for the score-pattern rule. Pattern minus weight was -0.0045, with a 95% paired person interval from -0.0126 to +0.0034. The paired interval includes zero. This small test does not establish a mean error advantage for either rule. We preserved all 70 four-pair set scores, fixed penalty and budget checks, models, predictions, figures, code and source records. One public raw trial was rebuilt for each person. This extends an existing feature sample. It tests offline response and rest windows, not continuous control or stimulation effects.

stereoelectroencephalography · brain-computer interface · feature selection · score patterns · decoder weights · Brier error · logistic regression · probability calibration