Local forecasts reduced normalized joint limit width by 16.7 percent versus a mean at a 90 percent target for simulated 50 ms gaps in 215 SEEG records from seven people

In 215 later SEEG records from seven people with epilepsy, a local forecast gave narrower complete-gap limits than a fit mean. Each simulated gap had 20 native voltage values: two hippocampal channels and ten samples over 50 ms. A ten second fit and a separate nine second calibration interval set the predictor and limits before the test. At a 90 percent target, joint forecast limits covered 89.2 percent of complete gaps. Their normalized full width was 16.7 percent lower than the mean baseline width (descriptive 95 percent person bootstrap interval: 14.6 to 19.3 percent). The mean baseline covered 88.5 percent of gaps. Forecast limits set separately for each value covered 90.0 percent of values but only 44.1 percent of gaps. In 78 of 215 records, joint forecast coverage was below 90 percent. All six fixed calibration checks are reported. A 95 percent target with 18 calibration scores gave unbounded limits. This is an offline test on intact pre-stimulation voltage. It does not establish coverage during real stimulation. The conformal rank method is established and requires exchangeability for its probability claim. The package contains raw selected files, exact provenance, code, predictions, limits, tables, figures, and reproduction instructions.

split conformal prediction · prediction intervals · stereo-EEG · SEEG · brain-computer interface · signal forecasting · coverage · calibration