Challenges of automated behavior observation in animals
Automated monitoring of movements and behavior of animals can be applied to classify species units of behavior. Monitoring wild species has shown to be problematic, and instead the behavior or a closely related species can be monitored and applied to close relating species.
Accuracies are grouped into either overall or threshold accuracy calculations due to the lack of defined solid standards. When training and testing behavior of 21 dogs and 7 wolves, accuracies ranged from 51 to 60% for classifying a total of 8 behaviors between the a single species, and 41 to 51% between the two species.
It was shown through this study that using data fro dogs to predict wolf behavior is feasible. The model optimized to similar overall and threshold accuracies, and with a common goal of animal behavior recognition, overall accuracy should be the preferred metric.
Read the full article by clicking the link below:
