tl;dr - ml model trained to fit steps, achieves good accuracy. we want to track small mammals, which is a bit less invasive (as opposed to tagging) (harder to track as opposed to larger more famous ones) ; tested on two near-identical species of sengi and found to be up to 96% accurate, other methods would be dna analysis, much more expensive and time consuming. Even while collectiing samples, they found one of the species outside of its expected domain (more evidence as to why we want to improve situation). Sengi steps were traced by using charcoal dust and paper. after finding statistically significant variations in them, it was tested on more raw data, and found 94-06% accuracy. Study is still small, limited to 2 species, so for general identification it would need lots of data, but serves as proof of concept. A Non-Invasive Footprint Technique for Accurate Identification of Cryptic Small Mammal Species: A Sengi Case Study, Frontiers in Ecology and Evolution (2026). https://dx.doi.org/10.3389/fevo.2025.1719684
Local good boys and girls: No, ML is bad! Think about all that water! (Chewing on some overprocessed shit, packed in 4 layers of plastic and delivered from the other side of the planet)
ml has existed for like 50 or so years, since basically first computers. if your model does not have billions of parameters (in this case, for 2 species, only 9 such parameters were identified), it uses far less compute.
Nobody cares. ML is bad. Period. [insert important name here] says so on YouTube! Because of water and nature. Poor rabbits! Why do you hate those cute little bunnies so much?!