Validating the Nordic Thingy:53 as an Accessible Tool for Equine Behavioural Analysis using Edge AI

Understanding animal movement and behaviour is integral to ensuring welfare needs are met, a central goal within the scientific community. Among domesticated animals, equids are unique; they serve as both companion and production animals, with estimates showing around 112 million working equids supporting 600 million people worldwide. Furthermore, the horse sport industry is a global multi-billion-dollar sector. However, these roles inevitably expose horses to potential physical and psychological harm. Consequently, being able to monitor behaviour accurately is critical to meeting their welfare needs. Here, we describe the design and evaluation of machine learning models capable of recognizing equine gaits and (semi-)natural behaviours using the Edge AI platform, Edge Impulse. Accelerometer motion data were collected using the Nordic Thingy:53 prototyping platform located on the chest of the horse. Both gaits and behaviours were detected with high mean training and testing accuracies (>95.00%), including idle, walk, trot, canter, fly-twitch, eating hay, head down or up, grazing, and rolling. While transformation types (Fast Fourier Transform (FFT) vs. Wavelet transform) were compared, they were not found to have a drastic effect on model performance. Instead, model accuracy and sensitivity were most impacted by the volume of training data included. This suggests that while transformations may refine model selection, sufficient training data is the single most important factor in defining effective Edge AI models. This study extends the evidence detailing the utility of accelerometers for equine movement and demonstrates that the accessible Nordic Thingy:53 platform can be used to develop highly accurate models for behavioural analysis. This is an important first step towards building effective monitoring methods that can recognise subtle behavioural changes that indicate alterations to emotional or physical states in equines.