What problem does it solve? Estimating biological brain age from neuroimaging features requires leakage-safe cross-validation and predicted-age bias correction, which are easy to get wrong when done manually. This Skill trains predicted-age models on tabular brain features, fits the age-bias corrector on each training fold only, and exports held-out raw age, corrected age, and Brain-PAD values for downstream group analysis. ## Core Features & Use Cases - Leakage-safe cross-validation: Fits the bias corrector on training folds only and applies it to held-out folds, keeping site, cohort, or family groups intact. - Multiple regressors: Supports ridge, OLS, elastic net, SVR, and optional XGBoost estimators reused from the statistical-ml layer. - Complete provenance: Outputs predictions.csv, fold_assignments.csv, metrics.json, per-fold checkpoints, config.json, and run_manifest.json. - Use Case: Given a CSV of subjects with chronological age and features like cortical thickness and hippocampal volume, train a ridge brain-age model with 5-fold cross-validation and obtain held-out Brain-PAD values for comparing accelerated aging across clinical groups. ## Quick Start Train a ridge brain-age model on my brain_features.csv file with 5-fold cross-validation grouped by site and export the held-out Brain-PAD predictions.
