What if silence could save your harvest? (I am building BeeBuzz, a bee classifier)
I'm building BeeBuzz for ML Zoomcamp 2026 and it's still rough. The idea: every bee species has a characteristic buzz. If a species goes quiet in an area, you could notice weeks before the flowers stay empty — and still have time to move hives and change the plan. Right now it takes short lab buzzes and gives back species and confidence in seconds. I'm working with 200 Mendeley lab recordings in…
I am developing a project called BeeBuzz for the ML Zoomcamp 2026, which is still in its early stages. The concept behind BeeBuzz is that every bee species has a unique buzz, and if a species becomes silent in a particular area, it may be possible to detect the absence weeks before the flowers start to remain empty. This could provide sufficient time to relocate hives and adjust plans accordingly.
Currently, the system analyzes short lab recordings and provides the species identification and confidence level within seconds.
The dataset comprises 200 recordings from the Mendeley lab, involving five different bee species. Out of these, 40 recordings are held back for testing purposes while the system is being refined. On average, the model achieves 80% accuracy on the test set. However, it should be noted that the weakest species, Centris cinza, with an F1 score of 0.57, requires additional data to improve its performance.
Two important caveats to consider are: the confidence scores tend to be overoptimistic, as they represent vote shares rather than calibrated probabilities; and the classifier's accuracy reaches 100% solely due to an artifact related to the recording setup, not because of the inherent biology of the bees. The current dataset consists of lab recordings without the inclusion of field noise. The next step involves field validation to evaluate the system's performance in real-world scenarios.
The project is open-source, and interested individuals can experiment with the loop live by uploading a WAV file, which will display the waveform, frequency, and prediction immediately. The project repository is accessible at https://github.com/marcos-Nieves-24/beebuzz, and a live demo is available at https://beebuzz-csqzda6ff5wtdlwicwuzcj.streamlit.app/. The creator extends gratitude to Alexey Grigorev and the DataTalks.Club team for their support and contribution to the community.
Written by urgent.news from Dev.to's reporting — not their text. Machine-written — may contain errors; check the original before relying on it.