Beta power before fixation raised mean ridge prediction error by 5.1 percent in blocked tests of eight SEEG participants from ds004473 v1.0.2

We tested whether beta power before fixation improved response time prediction in eight public SEEG recordings. We used OpenNeuro ds004473, version 1.0.2, with 1,990 retained response trials. The baseline model used task fields, session trends, and prior response time. The added model used all eligible bipolar beta features and a ridge penalty chosen within training data. Five blocked outer folds tested separate time periods with five-trial gaps. Adding beta power increased mean relative squared error in log response time by 5.05 percent. Each participant had equal weight. Six participants had higher error, one had lower error, and one had unchanged error. The exact two-sided participant sign-flip test gave p = 0.078125. This test did not establish a population increase at the 0.05 level. The result applies to this retained sample, feature window, and model. It does not prove absence of a neural relation. The package preserves source arrays, code, trial links, predictions, folds, tuning results, checks, and retrieval instructions.

stereoelectroencephalography · beta power · response time · ridge regression · nested validation · prediction