A fixed contact geometry term raised mean SEEG prediction error by 0.21% at 20 to 500 ms in eight tests that each excluded one person

We tested whether four contact positions improve a SEEG response-size forecast beyond distance, current, and contact spacing. We used OpenNeuro ds004696 version 1.0.1. The test included eight people, 85 stimulation sites, and 879 stimulation-to-recording pair instances. The target was the natural log of the post/pre RMS ratio for a five-pulse mean. The post window was 20 to 500 ms. Each outer ridge test excluded one complete person from model fitting and penalty choice. One fixed coordinate term raised mean squared error from 0.62559 to 0.62690, an increase of 0.21%. Error rose for six of eight people. The selected recording set changed at one of 85 sites. Five local calibration pulses gave error 54.6% below the base metadata rule. This comparator used a different information budget. The result applies to the tested feature and model family in this clinical cohort. The package preserves compact data, source hashes, all forecasts, fitted models, fixed checks, figures, and replay code. Fresh source rebuilds cover two people. The other six voltage inputs are inherited.

stereoelectroencephalography · electrical stimulation · contact geometry · ridge regression · response prediction · held-person validation