Identifying foraging spatial cues in beehive sound activity using machine learning methods

Abstract

The beehive sound, a continuous signal produced by bees within the hive, has been found to correlate with different behavioral states of the colony, like being queenless and swarming. We investigated the possibility of identifying foraging spatial cues in this signal. We recorded a colonys sound while foraging from food sources located at three different distances from the hive, one at a time. The recordings were split into frames to obtain six statistics of their Mel Frequency Cepstral Coefficients. Then, we evaluated different autoencoding networks to obtain a latent space that allowed frames from different foraging distances to be easily differentiable. The high accuracy, silhouette score, and F1 score shown in the obtained latent spaces strongly support our approach for identifying foraging spatial cues in beehive sound activity.