OxiGuard AI: A Machine Learning Model for Predicting Oxalic Acid Residue to Optimize Varroa Mite Control in Honeybees

Overview

Honeybees (Apis mellifera) are essential pollinators sustaining agricultural productivity and ecosystem stability but are severely threatened by the parasitic mite Varroa destructor, a primary driver of colony collapse disorder. Varroa infestations cause over $600 million in annual U.S. beekeeping losses and endanger billions of dollars in crop pollination services. Although synthetic acaricides are widely used, increasing mite resistance limits long-term efficacy. Oxalic acid extended-release (OAE) formulations offer a promising organic-compliant alternative; however, OAE remains unregistered in California due to limited residue and environmental safety data under diverse hive conditions. This study investigated how hive relative humidity, oxalic acid:glycerin ratio, and exposure duration influence oxalic acid residue deposition (µg/bee). Residue levels were experimentally quantified across simulated cage and real-world field conditions from San Ramon backyard hives and Grass Valley apiaries. Factorial ANOVA identified the three-way interaction among humidity, formulation ratio, and exposure time as explaining the greatest variance in residue levels (32.85%). A Random Forest machine learning model was developed to predict residue outcomes. After data cleaning, outlier removal, and controlled noise-based augmentation, the model achieved 5-fold cross-validation R² = 0.730 ± 0.058, validation R² = 0.758, MAE = 0.6 µg/bee, and RMSE = 1.1 µg/bee. Fourteen-day OAE field tests in Grass Valley, California, further validated the model, successfully predicting average residue levels with MAE = 0.21 µg/bee. The predictive model, OxiGuard AI, provides a data-driven framework to optimize OAE application and supports safer, scalable mite management and future regulatory evaluation in California.