Machine Learning Prediction of Oxalic Acid-Equivalent Residue on Honey Bees (Apis mellifera) After Extended-Release Oxalic Acid–Glycerin Treatments

The western honey bee, Apis mellifera, is an essential pollinator that supports agricultural productivity and ecosystem stability but faces a severe threat from the parasitic mite Varroa destructor, a major contributor to colony losses. Although synthetic acaricides are widely used, increasing mite resistance limits their long-term efficacy. Extended-release oxalic acid (OAE) formulations offer a promising organic alternative; however, further research is needed to characterize oxalic acid (OA)-equivalent residue dynamics under varying environmental and hive conditions. This study investigated how hive relative humidity (RH), the oxalic acid-to-glycerin (OA:GLY) ratio, and exposure duration influence OA-equivalent residue deposition (µg/bee). Residue levels were quantified experimentally under simulated hive conditions and in field colonies. Factorial analysis of variance (ANOVA) revealed that the three-way interaction among RH, formulation ratio, and exposure duration had the largest effect size (partial η2 = 0.3284). A random forest machine learning model was developed to predict residue levels based on these variables and their interactions. Following data cleaning, outlier removal, and controlled noise-based augmentation, the model achieved a 5-fold cross-validation R2 of 0.730 ± 0.058, a validation R2 of 0.758, a mean absolute error (MAE) of 0.6 µg/bee, and a root mean squared error (RMSE) of 1.1 µg/bee. Feature importance analysis identified RH as the most influential predictor (28.65%), followed by RH–formulation interactions (23.28%). These findings suggest that interactions between hive environmental conditions and treatment formulations can influence residue variability.