Fig. 3.

Learning procedure of the proposed comprehensive ensemble. The individual i-th learning algorithm ${\mathcal{ℒ}}_i$ outputs its prediction probability P_i for the training dataset through 5-fold cross-validation. The n diverse learning algorithms produce n prediction probabilities (P₁,P₂,⋯,P_n). The probabilities are concatenated and then used as input to the second-level learning algorithm $\mathcal{ℒ}$, which makes a final decision $ŷ$. a First-level learning. b Second-level learning