MVPA Classifier Explorer

k-Fold Cross-Validation

In MVPA, we use cross-validation to estimate how well the classifier generalizes. Each fold holds out a different run's data as the test set, training on the remaining runs.

Below is a simulated 5-fold CV on the current data:

Mean CV accuracy: —

Why cross-validate?

With limited fMRI data (often n=8–12 runs), leave-one-run-out CV is standard. It prevents inflated accuracy from overfitting while maximizing training data. The boundary you see is trained on ALL data — CV accuracy will typically be lower.