An XGBoost-Based Morphometric Classification System for Automatic Subspecies Identification of Apis mellifera

Accurate subspecies assignment of the western honey bee (Apis mellifera) is central to conservation and breeding but remains labor‐intensive with classical morphometrics and costly with molecular assays. We developed an XGBoost-based classifier built on a compact set of routinely measurable characters and evaluated it against standard machine-learning baselines. A curated dataset of labeled workers was measured under harmonized protocols, features were screened by embedded importance, and model performance was assessed with five-fold cross-validation. Using the top 10 characters—forewing width; venation angles A4, N23, D7 and J16; abdominal plate metrics (S6 and T4); distance between wax mirrors on sternite 3; and hind-leg tibia length—the classifier achieved an area under the receiver operating characteristic curve (AUC) of ~0.99 (95% CI ≈ 0.969–0.995), accuracy ≈ 0.988, and F1 ≈ 0.993, outperforming random forest and support vector machine baselines. SHAP analyses indicated that forewing venation and abdominal metrics contributed most to discrimination, and error inspection suggested misclassifications concentrated in morphologically overlapping lineages. The model retained high discrimination when restricted to field-practical measurements, supporting its use as a rapid triage tool alongside genetics. Our results demonstrate that standardized linear morphometrics paired with gradient-boosting yields near-state-of-the-art accuracy for subspecies identification while preserving interpretability and low cost, and the implementation can be ported to other taxa with comparable measurement schemes and representative training data.