Machine learning with neural data
Turn a recording into samples such as patient-level sessions, compute features from training data and evaluate on held-out people.
Loading video…
Turn a recording into samples such as patient-level sessions, compute features from training data and evaluate on held-out people. Signal preprocessing, windowing and artifact rejection must be documented. Adjacent windows from one recording are correlated, so splitting them across train and test leaks identity. Classifier accuracy shows predictive separation under the evaluation design; it does not identify a neurological cause.
Worked example
Compute mean alpha power per person, split by person and compare a simple classifier with the majority-class baseline. Report the confusion matrix and small-sample uncertainty.