3.2 Predictive Knowledge
To evaluate the selected models, we adopted 20 runs of the more robust 5-fold cross-validation, in a total of 20×5=100 experiments for each tested configu- ration. The results are summarized in Table 2. The test set errors are shown in terms of the mean and 95% confidence intervals. Three metrics are present: MAD, the classification accuracy for different tolerances (i.e. T = 0.25, 0.5 and 1.0) and Kappa (T = 0.5). The selected models are described in terms of the average number of inputs (I) and hyperparameter value (H or γ). The last row shows the total computational time required in seconds.