Machine learning-based affect detection within the context of human–horse interaction