With fixed ridge models, four selected SEEG contacts had lower mean accuracy than all usable contacts on the last matched block in eight epilepsy patients

We tested response-period detection in public SEEG data from eight epilepsy patients. A fixed ridge model used three power bands. Contacts were selected from training trials. The test used each patient’s last matched task block. Four selected contacts gave 62.5% mean balanced accuracy. All 119 to 126 usable contacts gave 68.8%. The paired loss was 6.3 percentage points, with a 95% participant interval of 0.9 to 12.0 points. Random splits gave no clear mean gain over the last-block test. The package preserves 2,059 trial pairs, features, source byte hashes, models, predictions, figures, and reproduction code. This is a test within one session. It does not establish movement-intent decoding or clinical BCI performance.

stereoelectroencephalography · SEEG · brain-computer interface · ridge regression · contact selection · response detection · temporal generalization · balanced accuracy