Figure 8.

Boxplots for the predictive performance of MLPs and CNNs using different hyperparameters (simulated pig dataset). Models were tested on five data partitions of the simulated pig dataset. Statistics represent the external (cross) validations by fitting the same model 30 times. The left three boxes are for MLP models, and the right three boxes are for CNN models. Null box means the model did not converge. Random, random hyperparameters; Perez, hyperparameters recommended by Pérez-Enciso and Zingaretti (2019); Opt, optimized hyperparameters using DE. Abbreviations stand for the same meaning in Figure 9 and Figure 10.