Effect of train set size on the ET models’ predictive performance: Coefficient of determination R² and MAPE as a function of the percentage of training instances used to train the models for both the clean and unclean samples. a) and c) correspond to the predictions of fluxes at 24 μm, while b) and d) to those at 8 μm. The symbols and error bars represent the mean and standard deviation of the ten-fold cross-validated metrics obtained for ten independent random samplings.