Explainable AI reached about 75% sex-classification accuracy for wild mountain gazelles
WHAT THE EVIDENCE SAYS
Researchers trained an explainable prototype-based image classifier on 1,833 cropped mountain-gazelle detections. Overall sex-classification accuracy and F1 were about 75%, and the visual prototypes highlighted biologically plausible features such as body and leg morphology and male horn structure.
WHY IT MATTERS / OUR READ
The interesting step is not just automation but auditability: wildlife classifiers become more useful when managers can see which visual features drive a result rather than receiving an opaque label. That can help target field validation and diagnose where models fail.
CAVEATS / UNCERTAINTY
About 75% accuracy is not census-grade certainty, and performance was evaluated on cropped detections from one population and camera-trap system. End-to-end performance also depends on animal detection, image quality and population-specific morphology, so the model is not yet a drop-in replacement for expert classification.
SOURCE
Scientific Reports
View original evidence ↗
A Jerusalem camera-trap study used prototype-based deep learning to make sex classification more interpretable, with the model focusing on body, leg, head and horn features.
Read the Wild Lands assessment →