“…This is, in fact, equivalent to treating all training data for different individuals as if they belong to the same subject [28,29,30,59]. In intra-subject approaches, an independent model is built for each subject si [27,31,32,40,41], by considering only training data that belong to that particular subject (scriptTsi). The high accuracy achieved by some subject-independent models (e.g., [28,29,59]) suggests that some relations between features and emotions hold for most individuals.…”