Sudden Queen Loss Event in an Africanized Honeybee Colony

The dataset consists of 23 features extracted from audio recordings of an Africanized honeybee hive in Fortaleza-CE, Brazil. The first feature is the recording date, and the last is the label indicating the queen's presence status. The label can take two values: "QR" for queenright (presence of queen) or "QL" for queenless (absence of queen). The remaining features are directly extracted from the audio signal, divided into three groups: time-domain features (zcr, energy, and energy entropy), spectral features (centroid, spread, entropy, flux, and rolloff), and 13 MFCC coefficients. For further details on the meaning of each feature, please refer to https://doi.org/10.1371/journal.pone.0144610.t002. The data were collected from daily recordings over a 6-day period, with the queen bee removed from the dataset on the last day. Consequently, the QR and QL classes are unbalanced, with QL representing only 1/6 of the data. This situation is common in this type of monitoring, where the hive's functioning is expected to remain within normal well-being parameters most of the time. Naturally, anomalies such as the sudden queen loss are uncommon and therefore represent a smaller portion of the data. The experiment and the data aim to replicate and incorporate these conditions for greater fidelity to the addressed problem. Such issues can be addressed using techniques such as anomaly detection, one-class classification, or incremental learning. Additionally, techniques for handling unbalanced data in classification problems, such as data augmentation and resampling, can be employed. Using OC-SVM, we achieved results with 96% accuracy and 99% precision.