Parameters influencing queen body mass and their importance as determined by machine learning in honey bees (Apis mellifera carnica)

Abstract

Most parameters describing queen bee quality are reflected in the queen’s body mass, which is in turn considered a robust measure and the best indicator of queen quality. State-of-the-art machine learning was used for the first time to jointly evaluate both biological and rearing parameters influencing queen body mass. Three different models were developed using different combinations of parameters. Regardless of the model composition, we achieved high precision of classification. The parameters “ovary mass” and “breeder” were the most important factors for model predictions. Differences in rearing practices and vegetation were masked by “breeder,” demonstrating the pitfall of this method. Separate analysis confirmed the importance of the time spent in the hive after mating and the phytogeographical region as an indirect indication of food sources. Rearing practices together with phytogeographical information are not enough to explain variation in queen body mass, yet they can contribute to the prediction of queen body mass if “breeder” is excluded from the model.