Automatic horse gait classification offers insights into training intensity, but directsensor attachment to horses raises concerns about discomfort, behavioral disruption, andentanglement risks. To address this, our study leverages rider-centric accelerometers formovement classification. The position of a sensor, sampling frequency, and window size ofsegmented signal data have a major impact on classification accuracy in activity recognition.Yet, there are no studies that have evaluated the effect of all these factors simultaneouslyusing accelerometer data from four distinct rider locations (the knee, backbone, chest, andarm) across five riders and seven horses performing three gaits. A total of eight modelswere compared, and an LSTM-convolutional network (ConvLSTM2D) achieved the highestaccuracy, with an average accuracy of 89.72% considering four movements (halt, walk,trot, and canter). The model performed best with an interval width of four seconds anda sampling frequency of 25 Hz. Additionally, an F1-score of 86.18% was achieved andvalidated using LOSOCV (Leave One Subject Out Cross-Validation).
Keywords: accelerometer sensor; animal activity recognition; convolutional LSTM network; equines; machine learning.