Table 2.

Cross-validated average performance metrics of classic ML models on each full-features dataset.

Feature Extractor	Best model	Accuracy	Precision	Recall	F1-score
MobileNetV3-Small	Linear SVM	0.974	0.985	0.965	0.973
MobileNetV1	Linear SVM	0.970	0.981	0.949	0.961
MobileNetV2	Linear SVM	0.951	0.964	0.924	0.937
MobileNetV3-Large	Linear SVM	0.953	0.969	0.928	0.942

Four models acting as feature extractors were tested together with five different classical machine learning algorithms acting as classifiers. The accuracy, precision, recall, and f1-score for all hybrid models, on the full-features dataset, are reported below.