Exploring the Relationship Between Emotion and Bodily Activity With Machine Learning.

Human experiences of bodily activity relate to their emotional state; however, exactly how is yet to be fully explored. We aim to conceptually model this relationship between bodily activity and subjective emotional experience using machine learning models. Bodily activity, including internal and peripheral physiological activity and body movements, has been implicated as an important part of a person's emotional state. However, practical investigation of the relationship between bodily activity and self-reported emotional state relies on suitable choices of emotional-state measures. Furthermore, the moment-by-moment associations of measures of bodily activity with a person's self-reported emotional state are still unclear. This study used recordings of the full-body movements and physiological activity of participants in dyadic interactions discussing various positive or negative topics. For each conversation, the dyads (N = 38) reported how they were feeling on three measures. We used a machine learning model trained on each different measure of bodily activity and self-reported emotion measure to investigate how well each measure of bodily activity predicts each measure of self-reported emotion within the machine learning model. Linear mixed models examining those results showed an interaction between how well bodily activity and emotion measures were predicted. Results also showed that within emotion measures, the intensity of some emotions was easier to predict than others. These results demonstrate that self-reports using summative ratings are easier to predict with machine learning models. In addition, the results replicated the predictive relevance of physiological activity over other measures for someone's emotional state.