Air-based detection of honey bee colony diseases using AI-driven in-hive data pattern analysis: system overview and validation framework

Parasites and pathogens are an increasing threat to bee colonies, yet practical, noninvasive tools for their early detection are lacking. Currently used monitoring methods rely mainly on visual inspections or single-parameter sensing, often identifying disease in the advanced development stage. Here we present and validate an integrated framework for continuous, air‑based monitoring of honey bee colony health that combines in‑hive volatile organic compound (VOC) sensing, microclimatic measurements, and machine-learning-based pattern recognition. Using controlled laboratory cage experiments, we present that Varroa destructor, Nosema ceranae, and chalkbrood can be clearly distinguished based on characteristic signatures in in-hive air composition, accurately classified by the developed learning models. Additionally, we analyze four main parameters, temperature, gas resistance, VOC, and NOx concentration, indicating their relation with a particular disease. The best-performing models achieved F1-scores exceeding 0.90 across all diseases under laboratory conditions. These results demonstrate that VOC-centered, AI-driven analysis can be effectively used for early, noninvasive disease detection in honey bee colonies and provide a reliable foundation for scalable precision apiculture and field-level validation.