Dataset for a Multimodal Machine Learning System for Non-Invasive Detection of Varroa Destructor Infestations in Honey Bee Colonies

This dataset contains processed field deployment data associated with a multimodal machine learning system for non-invasive detection of Varroa destructor infestations in Apis mellifera (honey bee) colonies. The dataset includes three real-world deployments representing distinct colony health conditions: A 30-day continuous deployment on a confirmed healthy colony (0 mites per 100 bees), used to validate baseline system behavior and false positive rates. A 30-day deployment on a mildly infested colony (2–4 mites per 100 bees), capturing early-stage distributional shifts without threshold-triggered alerts. A 1-day deployment on a clearly infested colony (8+ mites per 100 bees), demonstrating threshold exceedance and alert generation under high infestation conditions. Each dataset consists of timestamped fused probability scores generated from a confidence-weighted fusion of multiple sensing modalities, along with corresponding alert flags and hive status labels. The fused score represents the system’s estimate of infestation likelihood at each observation interval. Due to the large size of raw multimodal recordings (~300 GB), this release provides a processed and downsampled dataset suitable for reproducing the analysis and results reported in the associated study. This dataset enables evaluation of system performance across varying infestation levels and provides evidence of real-world deployment behavior under continuous monitoring conditions.