At the moment I apply 10-fold cross validation to make the selection of criteria more robust. However, the selections is still very dependent on the set of reporting points, for instance, by simply changing the seed in the random selection.
The issue must be related to the fact that in a 10-fold cross validation the variability among the subsets of stations is limited (10%). What if we apply bootstrap aggregation instead?
At the moment I apply 10-fold cross validation to make the selection of criteria more robust. However, the selections is still very dependent on the set of reporting points, for instance, by simply changing the seed in the random selection.
The issue must be related to the fact that in a 10-fold cross validation the variability among the subsets of stations is limited (10%). What if we apply bootstrap aggregation instead?