Figure A6.

Illustrates a comparison of confusion matrices obtained using cascade generalization, utilizing the probability outputs of the EfficientNet-B1 model. (a) Confusion matrix for classification using only the EfficientNet-B1 model (DL-only, level-0); (b) Confusion matrix for classification utilizing the EfficientNet-B0 probability outputs and hand-crafted features, obtained through an ensemble of conventional classifiers (averaging); (c) Confusion matrix for classification using EfficientNet-B1 (level-0) probability outputs and hand-crafted features for the best conventional classifier, random forest. All confusion matrices are generated after applying a threshold of 0.5 to the model’s probability outputs to obtain the final predictions.