Fig. 4.

Model accuracy and loss. Training and validation accuracy and loss, with moderate data augmentation. a.) The training accuracy ranges from 0.7186 to 0.9695, while the validation accuracy from 0.5208 to as high as 0.9844. b.) The training loss ranges from 0.1029 to 0.7192, while the validation loss from 0.019 to 2.1918. The biggest validation loss, 2.1918, occurs at the epoch 33, while the validation accuracy hits the lowest level, 0.5208. The blue line in figure a and figure b represents early stopping point since after epoch 60, the validation loss is no longer improving. Only the best model is saved during the whole training.