Traditionally, horse gait classification has depended on sensors attached to the horse itself. Mobile phones offer a more practical alternative, but the effectiveness of gait models based on such sensors has been underexplored. In this study, we applied deep learning to classify horse gaits using data from smartphones carried by riders. Seventeen horses and fourteen riders participated, with data collected simultaneously from the rider’s phone and a four-sensor horse-mounted system. Using this approach to generate labeled data efficiently, we trained a Bi-LSTM model that relied solely on 50 Hz accelerometer and gyroscope signals aligned to the horse’s frame of reference. The model successfully distinguished the five gaits of Icelandic horses with 94.4% accuracy. These results suggest that mobile phones can facilitate large-scale monitoring of horse movement. Future research should examine how factors such as rider style, equipment, and phone placement influence classification accuracy, which will further enhance understanding of equine gait and its applications in equestrian activities.
