From Captive to Wild Jaguars: Automated Behavioral Classification and Space Use Monitoring with Machine Learning

Monitoring both captive animals and wild populations is necessary to ensure adequate animal welfare and conservation. This could be achieved via camera traps, yet it could prove time-consuming and labor-intensive if handled manually. In this regard, machine learning (ML) is an active topic of scientific development. Here, ML models were made with LabGym and trained on footage of three captive jaguars to detect individuals and assess active and inactive behavior for both wild and captive jaguars in Bigai, Ecuador, and Randers Regnskov, Tropical Zoo (Randers Regnskov), respectively. The space use of the captive jaguars was also assessed. Footage of wild jaguars was received from camera traps in Bigai, and 123.8 hours of video footage were recorded of the enclosure in Randers Regnskov with three captive jaguars over 6 consecutive days. The ML model for individual recognition analyzed all videos from Randers Regnskov containing jaguars on November 1st, 2025, and another ML model trained to detect wild jaguars analyzed 67 videos from Bigai. The ML model showed clear patrolling behavior for captive jaguars on heatmaps. Captive jaguars exhibited a large amount of inactive behavior. Captive jaguars did not exhibit natural bimodal nocturnal or crepuscular hunter activity patterns. The ML model also showed promise in inferring behaviors of wild jaguars when trained on footage of captive jaguars. The results demonstrate that ML methods can provide a valuable tool in detection, behavior classification, and monitoring of space use.