On the Prediction of Varroa Mite Infestations in Honeybee Colonies via Acoustic Monitoring

Honey bees play a vital role in global ecosystems and agriculture through their pollination services. However, the Varroa destructor mite remains one of the most damaging parasites affecting honeybee health, contributing significantly to colony decline. Early detection of Varroa infestation is essential for effective beekeeping management. In this study, we propose a machine learning approach for predicting Varroa mite infestation levels based on acoustic analysis of hive recordings. We introduce a set of nine spectral shape descriptors (SSD) computed from conventional and modulation spectrograms to characterize the acoustic environment within hives. We evaluate the predictive power of these features against traditional cepstral features across multiple classifiers, including Support Vector Machines, Random Forests, and k-Nearest Neighbors. Our results demonstrate that the proposed SSDs significantly improve classification performance relative to conventional features across numerous figures-of-merit, particularly in hive-independent test settings. This work highlights the potential of spectral acoustic monitoring combined with supervised learning as a scalable and non-invasive tool for precision beekeeping and early detection of parasitic threats.