Added channel features increased delay beta test error by 18.3% per patient on average in retained trials from six ds006136 patients at ridge penalty 0.1

We tested whether separate channel features reduced delay beta power error beyond a mean of other channels. We used retained SEEG trials from six patients in ds006136 version 1.0.1. Each model learned from earlier trials and was tested on later trials. At fixed ridge penalty 0.1, added channel deviations and presence indicators increased mean squared error by 18.3% per patient on average. Two of six patients had less error. The 95% patient bootstrap interval for relative gain was -49.3% to +12.2%. Under a symmetric patient-gain null, the exact sign-flip test gave p=0.34375. Planned checks changed direction with stronger regularization or omission of presence indicators. The small sample does not establish a population error increase. The package preserves exact signal windows, source pins, code, fitted models, all predictions, and reproduction commands. The result applies to retained observed targets and does not validate restoration of artifact-rejected signals.

SEEG · beta power · ridge regression · missing data · channel features · model comparison