Assessing animal emotions is critical for welfare, yet reliable, objective indicators of canine fear remain debated. To address this gap, this study used machine learning to classify firework-related fear in dogs, comparing traditional manual behavioral coding with automated facial landmark analysis. Models were trained on videos of dogs experiencing calm baseline conditions versus stressful fireworks. While both approaches successfully detected distress, the model based on manual ethogram coding achieved the highest accuracy. It highlighted backwards-directed ears and increased blinking as the strongest predictors of fear. Notably, the machine learning algorithms recognized blinking as a key behavioral cue that traditional statistical analyses had previously overlooked.
Supervised Neural Style Transfer as an Augmentation Technique for Facial Landmark Detection
A major challenge in training accurate facial landmark detection models
