A Comparison of Human and Machine Learning-based Accuracy for Valence Classification of Subjects in Video Fragments