A performance comparison of auto-encoder and its variants for classification

In this paper, we present auto-encoder (AE), stacked auto-encoder (SAE) and sparse auto-encoder (SPAE) to classify gaits of horse riding for real riding coaching. The parameters of each auto-encoder are adjusted to compare the performance. The data is collected from 16 inertial sensors attached to a motion capture suit to construct a motion database. We build the motion features as the method of gaits classification with the database. The experiment shows that the performance is 95% when applied AE. SPAE was the best in terms of time and AE was the best in performance. We can apply to coaching system by each horse gait for rider under real or horse simulator environments using the SPAE algorithm.