A method for mining combined data from in-hive sensors, weather and apiary inspections to forecast the health status of honey bee colonies

Abstract Bees are the main pollinators of most insect-pollinated wild plant species and are essential for the maintenance of plant ecosystems and food production. However, over the past three decades they have been suffering from numerous health challenges, including changes in habitat, pollutants and toxins, pests and diseases, and competition for resources. An attempt to mitigate this problem is to estimate the health status of colonies and indicate an imminent collapsing state to beekeepers. To estimate the health status of bee colonies, we propose a method for calibrating a classification algorithm based on a supervised machine learning approach. We trained, validated, and tested three well-known and distinguished classification algorithms (k-Nearest Neighbors, Random Forest, and Neural Networks) and used real datasets from 6 apiaries, 27 Western honeybee (Apis mellifera) beehives monitored over three years (2016, 2017 and 2018). We used internal temperature and beehive weight as well as weather data (temperature, dew point, wind direction, wind speed, rainfall, and daylight). We also used 703 in loco apiary inspections made on a weekly basis to add to the data from sensors for the labeling needed in the training phase of supervised learning algorithms. The results suggest that we are proposing a high precision classification model (hit rate over 90%), which can be useful to self-predict healthy, unhealthy, and collapsing bee colony health states.