The learning process involves three steps: (a) learning the random forests, (b) constructing the second-level training set, and (c) learning the stacked classifier.