1. Start with all F = {x1, . . . , xI } input variables.
2. If there is a hyperparameter P ∈ {P1, . . . , Pk} to tune (e.g. NN or SVM),
start with P1 and go through the remaining range until the generalization
estimate decreases. Compute the generalization estimate of the model by
using an internal validation method. For instance, if the holdout method is
used, the available data are further split into training (to fit the model) and
validation sets (to get the predictive estimate).
3. After fitting the model, compute the relative importances (Ri) of all xi ∈ F
variables and delete from F the least relevant input. Go to step 4 if the
stopping criterion is met, otherwise return to step 2.
4. Select the best F (and P in case of NN or SVM) values, i.e., the input variables
and model that provide the best predictive estimates. Finally, retrain
this configuration with all available data.