Application of Supervised Machine Learning Algorithms to Morphological Classification of Bufo bufo (Linnaeus, 1758) and B. verrucosissimus (Pallas, 1814) Fr

ABSTRACT

Morphological discrimination between closely related amphibian species is often complicated by overlapping phenotypic variation, geographic structuring, and hybridization. In this study, supervised machine learning (ML) algorithms were applied to evaluate morphological differentiation between B . bufo  and B. verrucosissimus using morphometric data from 285 adult specimens (77 B. verrucosissimus , 208 B. bufo ) measured for 28 morphometric variables. Six supervised ML algorithms (k‐Nearest Neighbors, Artificial Neural Networks, Support Vector Machines, Naive Bayes, Decision Tree, and Random Forest) were trained using repeated cross‐validation on the training dataset and evaluated on an independent test set. Model performance was compared using accuracy, balanced accuracy, sensitivity, specificity, and area under the ROC curve (AUC). Among the evaluated models, Random Forest achieved the highest test‐set accuracy (82.46%), specificity (100%), balanced accuracy (75.0%), and AUC (0.871), whereas Support Vector Machines showed the highest sensitivity for detecting B. verrucosissimus (53.3%) among the better‐performing models. Variable importance analyses consistently identified parotoid gland morphology, particularly left and right parotoid width (LPW and RPW) and parotoid angle (PA), as the most informative characters for species discrimination. Additional informative variables included metatarsal tubercle measurements, interorbital distance, and radioulnar length. The results support previous morphometric studies while indicating that ML approaches can detect subtle multivariate morphological patterns that are difficult to capture with conventional analyses alone. These findings suggest that ML may serve as a useful complementary tool for morphological assessment within integrative taxonomic frameworks, although broader geographic sampling and external validation will be necessary to evaluate the generality of the observed classification performance.