Figure 2
Illustration of the machine learning pipeline. Features (e.g., brain volume, genetics) were extracted from the PPMI database. RReliefF based feature ranking and support vector regression with a polynomial kernel were applied to each fold in the tenfold cross-validation. This procedure was repeated for iteratively reduced features sets (lowest ranked features were removed one at a time) until only one feature was left. The feature set with highest R2 across folds was considered the best model.
