Half transfer reduced prediction error for shaft omission in paired 60 s 1–80 Hz SEEG excerpts from 35 patients with epilepsy at shrinkage weight 0.1

We tested whether shaft omission effects repeated in two interictal SEEG files. We used paired 60 s excerpts from 35 patients with epilepsy in OpenNeuro ds004100 version 1.1.3. No contact was reused between selected bipolar channels. Partial correlation used shrinkage weight 0.1. The baseline used the mean omission effect for each link in the training file. The tested model added half of the shaft effect beyond that mean. It reduced mean squared prediction error by 31.7 percent. Each patient had equal weight. The 95 percent patient bootstrap interval was 25.0 to 37.6 percent. All 35 patients had lower error. The interval assumes independent, exchangeable patients. Source byte checks, covariance rebuilds, separate matrix checks, and fresh output reproduction passed. The package preserves data, code, pinned retrieval recipes, and all fixed secondary tests. This study reuses one clinical cohort. It tests dependence on available channels. It does not test direct neural connections or clinical outcomes.

stereoelectroencephalography · partial correlation · electrode omission · network analysis · prediction error · epilepsy